Patterns and Process in Late Roman Republican Coin Hoards, 157-2 BC 9781407301648, 9781407332192

Citation preview

BAR S1733 2007 LOCKYEAR

Patterns and Process in Late Roman Republican Coin Hoards, 157-2 BC

PATTERNS AND PROCESS IN LATE ROMAN REPUBLICAN COIN HOARDS, 157-2 BC

Kris Lockyear

BAR International Series 1733 2007

B A R Gonella 1780 cover.indd 1

19/10/2009 12:48:54

Patterns and Process in Late Roman Republican Coin Hoards, 157-2 BC Kris Lockyear

BAR International Series 1733 2007

Published in 2016 by BAR Publishing, Oxford BAR International Series 1733 Patterns and Process in Late Roman Republican Coin Hoards, 157-2 BC © K Lockyear and the Publisher 2007 The author's moral rights under the 1988 UK Copyright, Designs and Patents Act are hereby expressly asserted. All rights reserved. No part of this work may be copied, reproduced, stored, sold, distributed, scanned, saved in any form of digital format or transmitted in any form digitally, without the written permission of the Publisher.

ISBN 9781407301648 paperback ISBN 9781407332192 e-format DOI https://doi.org/10.30861/9781407301648 A catalogue record for this book is available from the British Library BAR Publishing is the trading name of British Archaeological Reports (Oxford) Ltd. British Archaeological Reports was first incorporated in 1974 to publish the BAR Series, International and British. In 1992 Hadrian Books Ltd became part of the BAR group. This volume was originally published by Archaeopress in conjunction with British Archaeological Reports (Oxford) Ltd / Hadrian Books Ltd, the Series principal publisher, in 2007. This present volume is published by BAR Publishing, 2016.

BAR PUBLISHING BAR titles are available from:

E MAIL P HONE F AX

BAR Publishing 122 Banbury Rd, Oxford, OX2 7BP, UK [email protected] +44 (0)1865 310431 +44 (0)1865 316916 www.barpublishing.com

In loving memory of too many friends and family, especially my Mum and Dad, Hilda and Harry Lockyear.

Contents

List of Figures

v

List of Tables

ix

Preface

xi

1 Statistics, numismatics and archaeology 1.1 Introduction . . . . . . . . . . . . . . . . 1.2 Aims and Methods . . . . . . . . . . . . 1.3 Statistics and coinage . . . . . . . . . . 1.4 The structure of this book . . . . . . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

1 1 2 2 4

2 The data 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Roman Republican Coin Hoards and the chrr database 2.2.1 The database . . . . . . . . . . . . . . . . . . . . 2.2.2 Categories of data . . . . . . . . . . . . . . . . . 2.2.3 Sources of data . . . . . . . . . . . . . . . . . . . 2.2.4 Problems in the computerisation of coin types . . 2.3 Storage and data manipulation strategies . . . . . . . . 2.3.1 The database . . . . . . . . . . . . . . . . . . . . 2.3.2 Accuracy codes . . . . . . . . . . . . . . . . . . . 2.3.3 General Categories . . . . . . . . . . . . . . . . . 2.4 The coverage of the database . . . . . . . . . . . . . . . 2.4.1 Future development . . . . . . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

5 5 5 5 6 7 11 12 13 17 19 19 20

3 Models of coin supply and circulation 3.1 Introduction . . . . . . . . . . . . . . . . . . . 3.2 The life of a coin . . . . . . . . . . . . . . . . 3.3 Supply and distribution within a discrete area 3.4 Types of supply . . . . . . . . . . . . . . . . . 3.5 Inter-regional patterns . . . . . . . . . . . . . 3.6 Other factors . . . . . . . . . . . . . . . . . . 3.7 Hoards and the coinage pool . . . . . . . . . . 3.8 Summary . . . . . . . . . . . . . . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

21 21 21 24 27 27 28 28 28

i

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

Contents

4 The pattern of hoarding 4.1 Introduction . . . . . . . . 4.2 Testing the coverage . . . 4.2.1 Italy . . . . . . . . 4.2.2 Spain and Portugal 4.2.3 Romania . . . . . . 4.2.4 Other regions . . . 4.3 The pattern . . . . . . . . 4.4 Conclusions . . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

29 29 29 31 31 33 34 34 38

5 Comparing hoards — Correspondence Analysis 5.1 Aims and methods . . . . . . . . . . . . . . . . . . . . . 5.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . 5.1.2 Techniques . . . . . . . . . . . . . . . . . . . . . 5.1.3 Software . . . . . . . . . . . . . . . . . . . . . . . 5.2 CA and the analysis of coin hoards and site assemblages 5.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . 5.2.2 An informal description of CA . . . . . . . . . . 5.2.3 The ‘horseshoe curve’ or ‘Guttman effect’ . . . . 5.2.4 Dividing the data — selecting hoards for analysis 5.2.5 CA — a worked example and further problems . 5.2.6 Diagnostic statistics . . . . . . . . . . . . . . . . 5.2.7 Testing for significance . . . . . . . . . . . . . . . 5.2.8 PCA and CA — an empirical comparison . . . . 5.3 A usable methodology . . . . . . . . . . . . . . . . . . . 5.4 The Analyses . . . . . . . . . . . . . . . . . . . . . . . . 5.4.1 Introduction . . . . . . . . . . . . . . . . . . . . . 5.4.2 Hoards closing 147–118 bc . . . . . . . . . . . . . 5.4.3 Hoards closing 118–108 bc . . . . . . . . . . . . . 5.4.4 Hoards closing 105–97 bc . . . . . . . . . . . . . 5.4.5 Re-examining hoards from 118–97 bc by issues . 5.4.6 Hoards closing 92–87 bc . . . . . . . . . . . . . . 5.4.7 Hoards closing 87–81 bc . . . . . . . . . . . . . . 5.4.8 Hoards closing 80–79 bc . . . . . . . . . . . . . . 5.4.9 Hoards closing 78–75 bc . . . . . . . . . . . . . . 5.4.10 Hoards closing in 74 bc . . . . . . . . . . . . . . 5.4.11 Hoards closing 73–69 bc . . . . . . . . . . . . . . 5.4.12 Hoards closing 63–56 bc . . . . . . . . . . . . . . 5.4.13 Hoards closing 56–54 bc . . . . . . . . . . . . . . 5.4.14 Hoards closing 51–47 bc . . . . . . . . . . . . . . 5.4.15 Hoards closing in 46 bc . . . . . . . . . . . . . . 5.4.16 Hoards closing 45–43 bc . . . . . . . . . . . . . . 5.4.17 Hoards closing in 42 bc . . . . . . . . . . . . . . 5.4.18 Hoards closing 41–40 bc . . . . . . . . . . . . . . 5.4.19 Hoards closing 40–36 bc . . . . . . . . . . . . . . 5.4.20 Hoards closing 32–29 bc . . . . . . . . . . . . . . 5.4.21 Hoards closing 29–28 bc . . . . . . . . . . . . . . 5.4.22 Hoards closing 19–15 bc . . . . . . . . . . . . . . 5.4.23 Hoards closing 15–11 bc . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

39 39 39 40 41 42 42 42 43 43 46 57 60 60 64 64 64 65 69 72 77 79 82 89 89 96 102 104 107 112 116 119 123 127 131 134 139 145 149

. . . . . . . .

. . . . . . . .

. . . . . . . .

ii

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

Contents

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

149 154 157 171 174

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

179 179 180 180 182 183 195 197 200 200 202

7 Inter-hoard variability and the speed of coin circulation 7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2 An alternative approach . . . . . . . . . . . . . . . . . . . 7.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3.1 The first analysis — hoards from 87–81 bc . . . . 7.3.2 The second analysis — hoards from 74 bc . . . . . 7.4 Issue size and inter-hoard variability . . . . . . . . . . . . 7.5 Ramifications . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

205 205 207 209 209 214 216 223

8 Modelling coinage production and loss 8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . 8.2 Crawford’s method . . . . . . . . . . . . . . . . . . . 8.3 The late Republican coinage pool . . . . . . . . . . . 8.4 The simulation study . . . . . . . . . . . . . . . . . . 8.4.1 Constructing the theoretical coin populations 8.4.2 Comparing the results . . . . . . . . . . . . . 8.4.3 Testing Crawford’s risc figures (ρ) . . . . . . 8.4.4 Examining the decay rate (δ) . . . . . . . . . 8.4.5 Examining the introduction delay (ι) . . . . . 8.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . 8.6 Postscript . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

225 225 225 228 229 229 232 232 236 239 242 245

5.5

5.4.24 Hoards closing 8–2 bc . . . . . . . . . . . . Summary, conclusions and problems . . . . . . . . 5.5.1 Regional patterns . . . . . . . . . . . . . . . 5.5.2 Numismatic aspects . . . . . . . . . . . . . 5.5.3 Further comments on the CA of hoard data

6 Comparing hoards — Cluster Analysis 6.1 Introduction . . . . . . . . . . . . . . . . . . . 6.2 Dmax-based cluster analysis . . . . . . . . . . 6.2.1 Informal description . . . . . . . . . . 6.2.2 Technical issues . . . . . . . . . . . . . 6.3 The analysis . . . . . . . . . . . . . . . . . . . 6.4 Discussion of results . . . . . . . . . . . . . . 6.5 Principal Coordinates Analysis . . . . . . . . 6.6 Conclusions and discussion . . . . . . . . . . . 6.6.1 Archaeological and numismatic results 6.6.2 The statistical method . . . . . . . . .

9 Summary

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

249

A The Hoards 252 A.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252 A.2 The data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252

iii

Contents

B Concordances B.1 Introduction . . . . . . . . . . . . B.2 Database codes with Appendix A B.3 RRCH with Appendix A . . . . . B.4 RRCH with hoard identifiers . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

285 285 285 287 288

Bibliography

291

Index

317

iv

List of Figures

3.1 3.2 3.3 3.4

Model of coin circulation I . . . . . . . . . . . Model of coin circulation II . . . . . . . . . . . Idealised coin distribution between two mints . Down the line movement model . . . . . . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

22 24 26 26

4.1 4.2 4.3 4.4 4.5 4.6

Frequency of hoards in Italy during the later Republic . Denarius hoards per annum, entire database . . . . . . Denarius hoards per annum from Italy . . . . . . . . . Denarius hoards per annum from the Iberian peninsula Denarius hoards per annum from Romania . . . . . . . Denarius hoards per annum from Spain and Portugal .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

30 35 35 36 36 37

5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 5.10 5.11 5.12 5.13 5.14 5.15 5.16 5.17 5.18 5.19 5.20 5.21 5.22 5.23 5.24 5.25 5.26

Horseshoe curves . . . . . . . . . . . . . . . . . . . . . . . Sample score map from CA of 297 hoards . . . . . . . . . . Variable score map from CA of 297 hoards . . . . . . . . . First CA of test data, variable map . . . . . . . . . . . . . First CA of test data, object map . . . . . . . . . . . . . . Second CA of test data, variable map . . . . . . . . . . . . Second CA of test data, object map . . . . . . . . . . . . . Variable score map from third CA of the test data set . . . Sample map from third CA of the test data set . . . . . . Cumulative percentage curves for 13 hoards closing 46 bc . CA of 12 hoards closing in 46 bc, variable map . . . . . . CA of 12 hoards closing in 46 bc, sample map . . . . . . . Procrustes Analysis comparing CAs of the test data set . . Object loading map from PCA of the test data set . . . . . PCA of test data set, object loading map . . . . . . . . . . PCA of test data set, variable map . . . . . . . . . . . . . Cumulative percentage graph of hoards closing 147–118 bc CA of hoards closing 147–118 bc, variable map . . . . . . CA of hoards closing in 147–118 bc, sample map . . . . . Cumulative percentage graph of hoards closing 118–108 bc CA of hoards closing in 118–108 bc, variable map . . . . . CA of hoards closing in 118–108 bc, sample map . . . . . Cumulative percentage graphs of hoards closing 105–97 bc CA of hoards closing in 105–97 bc, variable map . . . . . . CA of hoards closing in 105–97 bc, sample map . . . . . . Map from CA of hoards closing 118–97 bc . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

44 45 46 49 49 50 50 51 52 53 54 55 56 61 63 63 66 67 68 70 71 71 74 75 75 76

v

. . . .

. . . .

. . . .

. . . .

List of Figures

5.27 5.28 5.29 5.30 5.31 5.32 5.33 5.34 5.35 5.36 5.37 5.38 5.39 5.40 5.41 5.42 5.43 5.44 5.45 5.46 5.47 5.48 5.49 5.50 5.51 5.52 5.53 5.54 5.55 5.56 5.57 5.58 5.59 5.60 5.61 5.62 5.63 5.64 5.65 5.66 5.67 5.68 5.69 5.70 5.71 5.72 5.73 5.74 5.75

Map from CA of issues from hoards closing 118–97 bc . . Percentage of RRC 286 . . . . . . . . . . . . . . . . . . . Cumulative percentage graph of hoards closing 92–87 bc CA of hoards closing in 92–87 bc, variable map . . . . . CA of hoards closing in 92–87 bc, sample map . . . . . . Cumulative percentage graph of hoards closing 87–81 bc CA of hoards closing in 87–81 bc, variable map . . . . . CA of hoards closing in 87–81 bc, sample map . . . . . . Variable map from CA of hoards closing 87–81 bc . . . . Sample map from CA of hoards closing 87–81 bc . . . . Map of Italian hoards closing 87–81 bc . . . . . . . . . . Cumulative percentage graph of hoards closing 80–79 bc CA of hoards closing 80–79 bc, variable map . . . . . . . CA of hoards closing 80–79 bc, sample map . . . . . . . Procrustes Analysis of hoards closing 80–79 bc . . . . . . Map of Italian hoards closing 80–79 bc . . . . . . . . . . Cumulative percentage graph of hoards closing 78–75 bc CA of hoards closing 78–75 bc, variable map . . . . . . . CA of hoards closing in 78–75 bc, sample map . . . . . . Cumulative percentage graphs of hoards closing in 74 bc CA of hoards closing in 74 bc, variable map . . . . . . . CA of hoards closing in 74 bc, sample map . . . . . . . . CA of hoards closing in 74 bc, variable map . . . . . . . CA of hoards closing in 74 bc, sample map . . . . . . . . Cumulative percentage graph of hoards closing 73–69 bc CA of hoards closing 73–69 bc, variable map . . . . . . . CA of hoards closing 73–69 bc, sample map . . . . . . . Cumulative percentage graph of hoards closing 63–56 bc CA of hoards closing 63–56 bc, variable map . . . . . . . CA of hoards closing 63–56 bc, sample map . . . . . . . Cumulative percentage graph of hoards closing 56–54 bc CA of hoards closing 56–54 bc, variable map . . . . . . . CA of hoards closing 56–54 bc, sample map . . . . . . . Cumulative percentage graphs of hoards closing 51–47 bc Variable map from CA of hoards closing 51–47 bc . . . . Object map from CA of hoards closing 51–47 bc . . . . . Procrustes Analysis of hoards closing 51–47 bc . . . . . . Cumulative percentage graphs of hoards closing in 46 bc CA of hoards closing in 46 bc, variable map . . . . . . . CA of hoards closing in 46 bc, sample map . . . . . . . . Cumulative percentage graph of hoards closing 45–43 bc CA of hoards closing in 45–43 bc, variable map . . . . . CA of hoards closing in 45–43 bc, sample map map . . . Cumulative percentage graph of hoards closing in 42 bc . CA of hoards closing in 42 bc, variable map . . . . . . . CA of hoards closing in 42 bc, variable map . . . . . . . CA of hoards closing in 42 bc, sample map . . . . . . . . Cumulative percentage graph of hoards closing 41–40 bc CA of hoards closing 41–40 bc, variable map . . . . . . . vi

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

77 78 80 81 81 83 85 85 86 87 88 90 91 92 93 94 95 95 96 98 99 100 101 101 103 104 105 106 108 108 109 110 111 113 114 115 117 118 119 120 121 122 123 124 125 126 127 128 129

List of Figures

5.76 5.77 5.78 5.79 5.80 5.81 5.82 5.83 5.84 5.85 5.86 5.87 5.88 5.89 5.90 5.91 5.92 5.93 5.94 5.95 5.96 5.97 5.98 5.99 5.100

CA of hoards closing 41–40 bc, sample map . . . . . . . Cumulative percentage graph of hoards closing 40–36 bc Variable map from CA of hoards closing 40–36 bc . . . . Object map from CA of hoards closing 40–36 bc . . . . . Percentage of RRC 443 . . . . . . . . . . . . . . . . . . . Percentage of RRC 468 . . . . . . . . . . . . . . . . . . . Percentage of RRC 494 . . . . . . . . . . . . . . . . . . . Cumulative percentage graph of hoards closing in 32 bc . CA of hoards closing 32–29 bc, variable map . . . . . . . CA of hoards closing 32–29 bc, sample map . . . . . . . Cumulative percentage graphs of hoards closing 29–28 bc CA of hoards closing 29–28 bc, variable map . . . . . . . CA of hoards closing 29–28 bc, sample map . . . . . . . Procrustes Analysis of CAs of hoards closing 29–28 bc . Cumulative percentage graphs of hoards closing 19–15 bc CA of hoards closing 19–15 bc, variable map . . . . . . . CA of hoards closing 19–15 bc, sample map . . . . . . . Cumulative percentage graphs of hoards closing 11–11 bc CA of hoards closing in 15–11 bc, variable map . . . . . CA of hoards closing in 15–11 bc, sample map . . . . . . Cumulative percentage graphs of hoards closing 8–2 bc . CA of hoards closing 8–2 bc, variable map . . . . . . . . CA of hoards closing 8–2 bc, sample map . . . . . . . . . CA of hoards closing 8–2 bc, variable map . . . . . . . . CA of hoards closing 8–2 bc, sample map . . . . . . . . .

6.1 6.2 6.3 6.4 6.5 6.6 6.7

Graph showing calculation of Dmax . . Example dendrogram . . . . . . . . . . Dendrogram from cluster analysis using Boxplots of group members . . . . . . . Cluster analysis groups a, n and o . . . PCO analysis plot — 1st and 2nd axes PCO analysis plot — 2nd and 3rd axes

7.1 7.2 7.3 7.4 7.5 7.6 7.7 7.8 7.9 7.10 7.11 7.12

Weights of coins of L Piso Frugi compared . . . . . . . . Proposed zones within coin hoards . . . . . . . . . . . . . . Line graph of scores from CA of 12 hoards from 87–81 bc . Partial CAs of hoards from 87–81 bc (I) . . . . . . . . . . Partial CAs of hoards from 87–81 bc (II) . . . . . . . . . . Partial CAs of hoards from 87–81 bc (III) . . . . . . . . . Partial CAs of hoards from 87–81 bc (IV) . . . . . . . . . Partial CAs of hoards from 87–81 bc (V) . . . . . . . . . . Partial CAs of hoards from 87–81 bc (IV) . . . . . . . . . Line graph of scores from CA of hoards from 74 bc . . . . Comparison of the size of issues from 90 and 74 bc . . . . Coin distributions over time . . . . . . . . . . . . . . . . .

8.1 8.2 8.3

A battleship curve—see text for explanation. . . . . . . . . . . . 229 Simulated population for 72 bc . . . . . . . . . . . . . . . . . . 235 Simulated population for 50 bc . . . . . . . . . . . . . . . . . . 235 vii

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

130 132 133 134 135 135 136 137 138 138 141 142 143 144 146 147 148 150 151 152 153 155 156 157 158

. . . . . . . . . . . . . . . . . . . . . . . . . . . . the average link method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

181 181 184 195 197 198 199

. . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

207 208 210 211 211 212 212 213 213 215 217 219

List of Figures

8.4 8.5 8.6 8.7 8.8 8.9 8.10

Simulated population for 100 bc . . . . . . . . . . . Varying the decay rate . . . . . . . . . . . . . . . . Simulated populations for 82 bc, archaic hoards . . Simulated populations for 82 bc, ‘middling hoards’ Simulated populations for 82 bc, modern hoards . . Simulated populations for 87 bc, modern hoards . . Total numbers of denarii in circulation . . . . . . .

viii

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

237 237 240 240 241 242 244

List of Tables

2.1

Accuracy codes in the database . . . . . . . . . . . . . . . . . .

19

4.1 4.2 4.3

Number of hoards by region in the chrr database . . . . . . . . Number of denarii in hoards by region . . . . . . . . . . . . . . Total numbers of denarius hoards by region . . . . . . . . . . .

32 33 33

5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 5.10 5.11 5.12 5.13 5.14 5.15 5.16 5.17 5.18 5.19 5.20 5.21 5.22 5.23 5.24 5.25 5.26 5.27 5.28 5.29 5.30

Hoards in the test data set . . . . . . . . . . . . . . Eigenvalues etc. from CA and PCA of the test data Diagnostic statistics for ‘samples’ . . . . . . . . . . Diagnostic statistics for variables . . . . . . . . . . Eigenvalues from CA analyses . . . . . . . . . . . . Hoards closing 147–118 bc . . . . . . . . . . . . . . Hoards closing 118–108 bc . . . . . . . . . . . . . . Hoards closing 105–97 bc . . . . . . . . . . . . . . . Hoards closing 92–87 bc . . . . . . . . . . . . . . . Hoards closing 87–81 bc . . . . . . . . . . . . . . . Hoards closing 80–79 bc . . . . . . . . . . . . . . . Hoards closing 78–75 bc . . . . . . . . . . . . . . . Hoards closing in 74 bc . . . . . . . . . . . . . . . . Hoards closing 73–69 bc . . . . . . . . . . . . . . . Hoards closing 63–56 bc . . . . . . . . . . . . . . . Hoards closing 56–54 bc . . . . . . . . . . . . . . . Hoards closing 51–47 bc . . . . . . . . . . . . . . . Hoards closing in 46 bc . . . . . . . . . . . . . . . . Hoards closing 45–43 bc . . . . . . . . . . . . . . . Hoards closing in 42 bc . . . . . . . . . . . . . . . . Hoards closing 41–40 bc . . . . . . . . . . . . . . . Hoards closing 40–36 bc . . . . . . . . . . . . . . . Hoards closing in 32 bc . . . . . . . . . . . . . . . . Imperial ‘issues’ . . . . . . . . . . . . . . . . . . . . Hoards closing 29–28 bc . . . . . . . . . . . . . . . Hoards closing 19–15 bc . . . . . . . . . . . . . . . Hoards closing 15–11 bc . . . . . . . . . . . . . . . Hoards closing 8–2 bc . . . . . . . . . . . . . . . . . Hoards with Roman Republican & Iberian denarii . Ten hypothetical coin hoards . . . . . . . . . . . . .

6.1 6.2

Summary of cluster analysis results . . . . . . . . . . . . . . . . 185 Detailed cluster analysis results . . . . . . . . . . . . . . . . . . 187 ix

. . set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

47 48 58 59 65 66 70 73 80 83 90 93 97 103 107 109 112 117 120 124 128 132 137 140 140 146 150 153 163 177

List of Tables

6.3 6.4

Cluster analysis date ranges . . . . . . . . . . . . . . . . . . . . 194 Cluster analysis ‘supergroups’ . . . . . . . . . . . . . . . . . . . 196

7.1 7.2 7.3 7.4 7.5

Hoards closing 87–81 bc . . . . . . . . . . . . . . Eigenvalues etc. from partial CAs . . . . . . . . . Hoards closing in 74 bc . . . . . . . . . . . . . . . Coins of 90 bc as a percentage of the ‘good total’ Coins of 74 bc as a percentage of the ‘good total’

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

209 210 215 218 220

8.1 8.2 8.3 8.4 8.5 8.6 8.7

Example of the results of the simulation program Hoards used in simulation study . . . . . . . . . . Dsum comparisons, 72 bc . . . . . . . . . . . . . Dsum comparisons, 51 bc . . . . . . . . . . . . . Dsum comparisons, 100 bc . . . . . . . . . . . . . Dsum comparisons, 82 bc . . . . . . . . . . . . . Total numbers of denarii in circulation . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

230 233 238 238 238 238 244

x

Preface Over the last eighteen years friends and colleagues have often asked, why Roman coins? Why use statistics? Why the Roman Republic? My interest in Roman coins stems from the teaching of John Casey at Durham, particularly his coinage course which I took in 1986–7. During 1988–9 I undertook the M.Sc. course in archaeological computing at the University of Southampton. It became quickly obvious that the most efficient method for handling large quantities of coinage data was the construction of a relational database. I also took courses in statistics taught by Stephen Shennan, during which I was taken with the beauty of multivariate statistics, and their applicability to coinage studies. For my M.Sc. dissertation Richard Reece suggested that I examine the Crawford–Buttrey debate, a topic which had the advantage that a body of good quality data was available in the form of Table L in Roman Republican Coinage. At the end of this dissertation, it was clear that further work would be profitable and I went on the expand the database and the analyses which formed the core of my PhD thesis submitted in 1996. Unfortunately, work on the Celtic Inscribed Stones Project prevented me from immediately recasting the work into two proposed books, of which this is the first. Lecturing, and running the Noviodunum Archaeological Project (Lockyear et al. 2007) delayed the volumes even further. This book derives from the central part of that thesis, although it has been almost entirely rewritten and most of the analyses have been rerun using a larger set of data. The work on the late Iron Age in Romania has yet to appear although I have summarised some of it in print (Lockyear 2004). Funding for my thesis was provided by the SERC to whom my thanks are due. This book was completed during the Autumn term 2006 during a sabbatical kindly granted to me by the Institute of Archaeology, UCL. In conducting the research for my thesis and thus this book, I have incurred many debts of gratitude and I would like to take this opportunity to acknowledge the help I have received. I would like to thank my supervisors and now colleagues, Clive Orton and Stephen Shennan for all their help and encouragement. I would also like to thank: Michael Crawford for allowing me to use his archives and data, now stored in the British Museum; the then staff of the Dept. of Coins and Medals, primarily Andrew Burnett and Roger Bland, for allowing me such easy access both to those archives and to the library of the Department; Richard Reece, who kindly provided the original idea for my M.Sc. dissertation and spent much time discussing my ideas and work, and was my examiner; Matthew Ponting for many useful and interesting discussions; Terrence Volk and Ted Buttrey, both from Cambridge, for providing off-prints and discussing my work; Mike Baxter, Christian Beardah, Richard Wright and Morven Leese for discussing and helping with the statistical aspects of my work; Nick xi

List of Tables

Ryan for his help with the database and for so promptly sending me copies of his article and book, and Sebastian Rahtz for acting as my unpaid, unofficial but willing LATEX guru; Dale Serjeantson for proof-reading almost the entire original thesis. I also wish to thank Tim Sly, Sally, Jonathan and Lindle Markwell for all the support they gave me when I wrote the original thesis for which I am very grateful. Many people have helped me cope with the large numbers of languages involved in this work, including Federica Massagrande (Italian, Spanish), Rosemary Burton (Spanish), Dale Serjeantson (French), Genevieve Stone (Russian) and Doina Whitaker (Romanian). My work in Romania could not have taken place, and certainly would not have been so pleasant, without the kind help and co-operation of a great many people to whom my thanks are willingly given, foremost of whom are the late Gh. Poenaru Bordea and Adrian Popescu formerly of the Institutul de Arheologie “Vasile Pârvan” in Bucureşti, and Virgil Mihăilescu-Bîrliba of the Institutul de Arheologie, Iaşi. I would particularly like to thank Adrian’s and Virgil’s families for their repeated hospitality over the years. I would also like to thank: Constanţa Ştirbu (formerly of the Muzeul de Istorie al România), the late Radu Ocheşeanu (formerly of the Severeanu Museum, Bucharest), Catalin and Geta Borţun (Muzeului Judeţean Teleorman, Alexandria), Ecaterina Ţînţăreanu (Muzeului Judeţean Teleorman, Alexandria), Aniko Mitruly, Lili Ardea and Doina Comşa, (Mediaş Museum), Ioan Caproşu (Universitatea Al. I. Cuza, Iaşi), Dan Gh. and Silvia Teodor, Victor Spinei, Costel Chiriac, Ion Ioniţa, and all the researchers and staff at the Institutul de Arheologie, Iaşi, and Nicu Bolohan and Marius Alexianu and their families (Piatra Neamţ). This book was largely written during my sabbatical in 2006 which I spent in South Carolina with Ellen Shlasko. It was then typeset by myself in LATEX during the autumn of 2007 when I had a broken leg, and spent some of the time recuperating in South Carolina. Thank you Ellen, for everything.

A note on place names and locations I am afraid I have been inconsistent with place names in this volume, which is largely a reflection of the inconsistencies in my sources. On the whole, I have stuck with the names of hoards as given by Michael Crawford (1969c), hence Rome and Padova rather than Roma and Padova or Rome and Padua. In some cases the name given by Crawford has changed substantially due to changes in borders, thus Nagykágya (RRCH 411) is now in Romania and is called Cadea. In those cases I have placed the new name in parentheses after the old name in the catalogue and the tables. After the fall of communism the Romanian Academy decided to change how Romanian words were spelt, particularly the letter î was changed to â. I have generally, however, stuck to the older spellings which are more familiar to non-Romanian speakers. Hoards with a very nonspecific location are usually shown in inverted commas (e.g., ‘Bahrfeldt’). The countries from which hoards derive are also a little inconsistent. I have split Italy in several areas, particularly listing Sicily separate from mainland Italy so as to provide a quick method of assessing regional variation. Hoards from the former Yugoslavia, however, have been grouped under that heading rather than split into the current states. Kris Lockyear, Institute of Archaeology, UCL xii

1

Statistics, numismatics and archaeology

1.1

Introduction

This book uses multivariate statistics to investigate late Roman Republican coin hoards from an archaeological perspective. This approach has many strengths. Trying to identify patterns in 100,000 coins is certainly possible without using multivariate statistics, but it would be very hard work and some of the more subtle patterns are likely to be missed. Numismatic evidence has much to offer archaeology in terms of dating evidence, economic developments, the use of coins, attitudes to value and so forth and archaeology helps provide numismatists with a context for the coins recovered. The weakness of the approach is that few archaeologists, and fewer numismatists understand statistics very well (Thomas 1978). Statisticians do not always appreciate the problems with the data or the aims of the analysis, and archaeologists and numismatists often have problems communicating with each other (Casey 1986). Similarly, the advantages of a properly constructed relational database system often add to the confusion (Lockyear 2007b). Developments in archaeological theory have had a negative impact on the use of statistics in archaeology (Baxter 2003, pp. 3–8) and Renfrew may have been overly optimistic when he opined “The days of the innumerate are numbered” (quoted by Shennan 1997, p. 1). Even the doyen of numerical techniques in the study of Roman coinage has felt more recently the need to write a book with “no numbers” (Reece 2002, esp. p. 9–10). My intention in this volume is not only to explore the complex patterns within Roman Republican coin hoards but also to demonstrate by example the advantages and beauty of multivariate statistics. This volume is not intended as a statistical textbook for which there are many examples. For an archaeological audience that by Shennan (1997) is excellent. An approachable textbook for multivariate statistics is that by Bartholomew et al. (2002). Baxter (1994) provides an in-depth discussion of some of the methods and he has also provided a broad overview of statistics in archaeology (Baxter 2003). However, I have tried to write this book in such as way as to allow those without specialist knowledge to understand the analyses. I have tried to clearly demarcate the sections which are more technical from more informal descriptions of the techniques. Some material is also included which is of more relevance to debates within archaeological statistics, and those too are clearly indicated. The reader is urged, therefore, to skip the parts not relevant to their level of expertise.

1

1. Statistics, numismatics and archaeology

1.2

Aims and Methods

Just as there are schools of archaeological thought there are schools of statistical thought (Baxter 2003, p. 3). It is not uncommon for statistics to be viewed as a way of testing a quite specific hypothesis which is accepted or rejected on the basis of the results. An alternative approach is to view statistics as a method for exploring data. Tukey (1977) is a seminal volume that defined this approach and describes a variety of simple tools for doing so. With the development of computers, the application of more complex multivariate tools has grown, but the aim of ‘exploring’ the data is similar. Rather than having a specific testable hypothesis such as ‘hoards from Spain have more coins of type A than hoards in Italy’, the question tackled in this book is: ‘what patterning is there in the hoard data?’ The methods chosen, mainly Correspondence Analysis and Cluster Analysis, were selected as those most likely to answer that general question. What those patterns mean take us from the realm of statistics into the realm of numismatic and archaeological interpretation. Archaeologically and historically, the principal aim is to examine the reasons for the differences between hoards such as the pattern of supply of coinage, or differences in the use of coinage. Historical events obviously impact on coin hoards. For example, very large quantities of coinage were struck during the Civil Wars at the end of the Republic and this can be seen in the hoard evidence. There is also an increase in hoarding, or at least hoard non-recovery, as shown by Crawford (1969a). At a more detailed level, distribution patterns between issues struck by the different protagonists persist in the hoard evidence, not just at the time but some years later. We can see, for example, that the legionary denarii of Mark Antony are prolific in the east when struck, then become more common in Italy as they circulated, and finally arrive in Spain. They are relatively under represented in hoards from Romania at the time which begs the question ‘why?’ We must also, however, be aware that some of the patterns we can see in the evidence may be the result of more prosaic factors, such as simple random variation in samples.

1.3

Statistics and coinage

Like mathematics, numismatics can be divided into two broad overlapping subject areas which could be defined as ‘pure’ and ‘applied.’ ‘Pure’ numismatics concerns itself with the date of the coin issues, who struck them, what the images on the coins mean, what weight they were intended to be struck at, the composition of the metal they are made of, and so on. ‘Applied’ numismatics is the use of coin data for more historical or archaeological purposes and often concerns the analysis of finds from archaeological excavations (Lockyear 2007b). Coin hoards are a key piece of evidence for both strands of numismatics. For example, the presence or absence of specific coin types from hoards is the foundation of most dating schemes. Similarly, the quantity of hoards over time or the spread of them geographically has been used to examine periods of economic or political unrest (Crawford 1969a; Kent 1988). Within ‘pure’ numismatics three topics have received detailed statistical treatment (Metcalf 1981): the intended weight of the coins when stuck, the analysis of their metallurgical composition and the estimation of the size of coinage 2

1.3. Statistics and coinage

issues, usually from an analysis of coin dies. Of these three subject areas the last has received most attention. Esty (1986) surveyed the various methods proposed up to the mid eighties, and the various volumes of the Survey of Numismatic Research contain updates (Esty 1997, 2003). This topic has become an obsession amongst numismatists even though the method proposed by Esty (1984) some time ago seems perfectly acceptable. The estimation of the size of the issues of the Roman Republic by Crawford (1974) has received most attention and Chapter 8 examines this in detail. The work of ‘applied’ numismatists receives less attention in the numismatic literature. Although the analysis of site finds has a long history (e.g., St. O’Neil 1935) the foundations of the modern subject were laid in the late 1960s and early 1970s (Casey & Reece 1974). For many, producing a summary of the coin finds via a histogram or barchart and comparing them to others is sufficient, e.g., the Lydney report (Casey & Hoffman 1999). Richard Reece, however, has developed a series of different techniques for comparing sites which culminates in his method using cumulative percentage curves (Reece 1995; see Lockyear 1996b, pp. 61–72 for an overview). In this method the coin lists are converted to per mills (i.e., per 1000 or %) and then cumulative curves calculated. An average profile is then treated in the same way and subtracted from each site or hoard, and the curves plotted on a graph. For sites, this is likely to be the average coin loss for the region; for hoards it might be the average contents of hoards closing in a specific period. This average is actually rather unimportant, its function is to make groupings in the assemblage clearer. This technique has much to commend it as it is simple and easy to understand although there are some issues in the interpretation of the diagrams (Lockyear 2007b) and it can be quite labour intensive (cf. chapter 6). The technique has gained acceptance by other scholars (e.g., Esmonde Cleary 2001). Multivariate methods have been used much less frequently. Ryan (1982, 1988) undertook Principal Components Analysis of fourth century coin finds from Britannia. Duncan-Jones (1989) performed Factor Analysis on nine coin hoards but the analysis is reported so poorly it is impossible to assess the results. Reece (1995, pp. 181–2) noted that these more formal statistical methods, as applied by students undertaking the numerical methods course at the Institute of Archaeology, London, had been unsatisfactory in the analysis of his 140 sites data (Reece 1991) and were always overly influenced by sample size. I demonstrated, however, that in the hands of a more experienced analyst these techniques — in this case Cluster Analysis and Correspondence Analysis — could be very powerful indeed (Lockyear 2000b). The problem was not variable sample size, although it was an issue, but the extremely unusual profile of coin loss from the site at Fishbourne, England. An interesting sequence of papers are those by Creighton (1994), van Arsdell (1996) and Orton (1997). Van Arsdell (1996) attempted to counter the suggestions of Creighton (1994) by using significance testing. Unfortunately, the methodology he used was poor prompting Orton’s (1997) paper Testing significance or testing credulity? In this paper Orton showed how the patterning in late Iron Age hoards could be analysed using Correspondence Analysis. Unfortunately, it was unclear that the first two papers may have used the same hoards, but had used different typological and dating schemes (Creighton, pers. comm.)! As Creighton did not present his raw data, Orton had used the figures given 3

1. Statistics, numismatics and archaeology

by van Arsdell. It would be interesting to reanalyse the data using Creighton’s numbers and compare the two. Beyond applied numismatics, multivariate analysis has seen application in a wide variety of archaeological fields. Baxter (1994) gives an extensive bibliography. For example, Peacock (1989) applied Cluster Analysis to the mills of Pompeii, and Correspondence Analysis has been used to examine glass assemblages (Cool & Baxter 1999), small finds assemblages (Cool & Baxter 1995, 2002), animal bones (Orton 1996) and pottery assemblages (Pitts & Perring 2006) as well as my re-examination of Reece’s coin data (Lockyear 2000b). If these techniques are so powerful, why have coinage studies not employed them more extensively? There are various possibilities. Firstly, there is the perception that the techniques are ‘difficult’ and ‘complicated’ (Baxter 2003, pp. 16–18) This is only partly true. If we insist on understanding the detailed mathematics behind the methods then this is possibly true but if we can treat the technique as a ‘black box’ then we can use the techniques effectively. This does not mean we can throw any old data at the methods and hope they work. A basic understanding of what they do is needed, what the method’s assumptions are, and how to interpret the output is obviously important. Learning these methods requires some work, but any archaeological expertise is gained as the result of work and practice. Criticising CA on the basis that it ‘can position hoards in a two-dimensional plane according to unspecified similarities’ (Esty 1997, p. 819) reveals a lack of understanding of the method. (The similarities are based on an analysis of χ2 distances in a table of data; Greenacre 1993.) All research, archaeological, numismatic or otherwise requires the scholar to make ‘numerous decisions about what to include and what to omit’ (Esty 2003, p. 924) and is not a cogent argument for avoiding multivariate statistics. Curiously, later in the same review, Esty (2003, pp. 925–6) states that the paper by Orton (1996) described above ‘discuss[es] the proper methods for comparing hoards to see if they are drawn from the same population.’ The proper method was, as we have seen, Correspondence Analysis. Secondly, and perhaps more importantly, numismatists are generally more interested in other problems, such as die estimation, and thus these techniques do not answer the questions they wish to ask. This cannot be used as a criticism of the method, but simply a reflection of academic diversity.

1.4

The structure of this book

The first three chapters provide the background to the book examining firstly the data used, then models of coinage supply and the distribution of hoards geographically and temporally. The next two chapters provide a detailed analysis of up to 300 coin hoards using CA and Cluster Analysis. A detailed introduction to the two methods and their application to coin data is provided, as well as a summary of the conclusions from each. The next two chapters provide two case studies which build on these analyses, one examining the speed of coin circulation and the second re-examining Crawford’s estimates of the size of Republican issues. A summary of the major conclusions is then given, followed by a abbreviated catalogue of the hoards used and concordances to the standard catalogue of these hoards (Crawford 1969c).

4

2

The data

2.1

Introduction

This chapter outlines the nature, sources and problems with the data used in this book. More often than not, the difficulties in gathering, organising, correcting and manipulating data in a study such as this are not discussed in any detail. As a result, future scholars often have problems reconciling different publications, or recreating the analyses. In numismatics this often shows itself by the same hoard having different totals in different publications, or a different closing date, or even a different name. Thus, if we compare the data published by Crawford (1974, Table L) with Backendorf (1998) we find sometimes significant differences. This is because Table L does not publish coins earlier than 157 bc or after the end of the Republic, whereas Backendorf lists the entire hoard. Some of the hoards in Table L have, however, been examined by Crawford and where “inspection has led me to correct the published record, the correct figures have simply been used for the Table without explicit note” (Crawford 1974, p. 643, n. 3). My listings may vary once more because I had access to Crawford’s personal records housed in the British Museum, which were not consulted by Backendorf. I have, therefore, decided to take the unusual step of discussing the sources of data, how the database was created and so forth. This could be seen, in a way, as a part of what archaeologists would term formation processes. In this way future scholars reading this book or using the database will have a head-start in understanding the issues. Thus, this chapter is essential for understanding: • the project archive; • the project database, which will be made available to interested parties; • the reasons for the manner in which the analyses were performed and their possible shortcomings. Outlined are the strategies adopted for the collection, input and manipulation of the data, the sources of data, and the structure of the Coin Hoards of the Roman Republic (chrr) database.

2.2 2.2.1

Roman Republican Coin Hoards and the chrr database The database

Before briefly outlining the history of the database we need to be sure of our terminology. Contrary to popular usage, a database is an orderly collection of 5

2. The data

data which need not be stored on a computer. The software used to manipulate a computerised database is a DBMS or Database Management System. There are a number of models for organising data in such a way that it can be efficiently input, stored and manipulated on a computer, the most common of which is the relational model (Date 2000) and software that uses this system is called a RDMS (Relational Database Management System). The chrr database started life as part of my MSc dissertation in 1989 (Lockyear 1989, appendix C) and at that time consisted solely of the hoard data published in Roman Republican Coinage (henceforth RRC; Crawford 1974, Table L) along with relevant information on the different types of coin contained within the main catalogue. This database was implemented using the Ingres RDMS running on a network of Sun workstations. During my PhD the database was greatly expanded using using the data discussed below. The database structure was altered and implemented using the MS-DOS database package dBase III+. The data on coin types was imported from the Ingres database, although problems in reconciling Table L with other sources of data resulted in this data not being immediately included. This database is described in full in my thesis (Lockyear 1996b, Chapter 5). Subsequent to completing the thesis the database was not updated substantially until 2006, although it was used as an example database for my students undertaking the database course at the Institute of Archaeology, UCL. In using the database as a teaching tool, it became clear that some of the ‘table’ and ‘column’ names were confusing, and so these were changed for clarity. In 2006 the database was imported into Microsoft Access. I decided to update the data using three volumes which had been published since the completion of the thesis (Backendorf 1998; Moisil & Depeyrot 2003; Paunov & Prokopov 2002). None of these volumes were unproblematic and will be discussed further below.

2.2.2

Categories of data

The main category of data is hoards of the Roman Republic and the early Principate. The working definition used is similar to that used by Michael H. Crawford (Roman Republican Coin Hoards, henceforth RRCH, Crawford 1969c). A hoard is defined as any two or more coins deliberately deposited together. Many hoards of this period contain coins from several different issuing authorities. Therefore, a hoard is included if it contains at least one Roman coin. Also following Crawford, hoards are included down to the issues of C. L. Caesares (c. 2 bc to ad 4; Sutherland 1984, henceforth RIC 1(2) , pp. 55–56, nos. 205–212). The earliest hoards are those containing denarii (i.e., post 211 bc). Pre-denarius hoards are not considered. In the main data tables of the chrr database, issues of the Republic and the early Principate are recorded to as exact a catalogue reference as possible. Non-Roman and poorly identified coins are recorded as general categories, e.g., Iberian denarii, miscellaneous Republican asses. For the rest of this work I shall refer to the coinage under investigation as Republican despite the fact it also contains some of the early Imperial material. Throughout this work the ‘primary key’ for a hoard, a three letter code, will be given in small capitals whenever a hoard is mentioned. This is in order that there can be no doubt as to which hoard is being referred to as the full hoard name is not necessarily unique. For example, there are several hoards

6

2.2. Roman Republican Coin Hoards and the chrr database

from Padova and two from Carbonara. A concordance between the code and the data listed in Appendix A is given in Appendix B. Although it was originally hoped to collect other data about the hoards such as how they were found, whether they were in a vessel or not or accompanied by other items, this was very difficult to do in practice and although the project archive contains copies of many of the relevant articles and notes, the data has not been systematically extracted or computerised. The second major category of data is information about the individual coin types. The cointype table (see below) is directly derived from Roman Republican Coinage (RRC) and Roman Imperial Coinage, vol. 1, second edition (RIC 1(2) ). The sorts of data included in these catalogues include the type, legends, place of minting, denomination, dates and moneyers. At present the only data used are the dates, the typological scheme and the denominations.

2.2.3

Sources of data

The sources of hoard data for this project are varied. The majority of information is from Michael Crawford’s records now housed in the British Museum. In the discussion below regarding problems with the data, these should not be seen as criticisms. This data set was collected for Crawford’s own purposes. These comments should be seen solely as background information to the chrr database. The sources of hoard data fall into several categories: 1. unpublished hoards (a) with detailed list (usually compiled by Crawford) (b) with no detailed information 2. published hoards (a) articles about individual hoards, or a small group of hoards (b) corpora of hoards but with only summary information (e.g., RRCH) i. by period ii. by region iii. by type or issue of coin (c) corpora of hoards but with detailed information e.g., that by Chiţescu (1981) or Chaves Tristán (1996). These can be divided as in 2b. (d) other works which contain details of hoards, e.g., RRC and Coinage and Money under the Roman Republic (Crawford 1985). 3. derived data — these are lists derived from reports. Crawford has for many hoards produced a detailed list of coins with Sydenham (1952) reference numbers. The British Museum archive It is necessary here to explain the archive in the British Museum, as this information is vital for understanding the archive for this project. The main task was to cross reference the data stored in four locations. These are RRCH, a filing

7

2. The data

cabinet with hoard records collected by Crawford and now housed in the Department of Coins and Medals, Crawford’s personal card index which accompanies the filing cabinet, and the various publications in the Department’s library. RRCH is a published list of hoards with references, closing dates, locations etc. of 549 coin hoards and 18 other coin finds. This work forms the basic list of hoards, and RRCH numbers are given as standard reference numbers in a number of other works. For many of the hoards in RRCH the detailed coin list or publication is stored in Crawford’s filing cabinet. These data take several forms: 1. offprints or photocopies of the original articles. The lists of coins are either given with references to one of the many catalogues, or simply with a description of each coin. 2. index cards with a detailed list of coins catalogued using Sydenham (1952). These can either be original lists from the coins themselves or derived from publications. These are generally pre-RRCH/RRC. 3. handwritten lists. Some of these are as 2 but many simply record as little information as is necessary for the identification of the coin type. Only rarely are references ambiguous. In many cases it is possible to decide between two possible references on the basis of the position of the reference in the list which is usually in, or nearly in, the same order as Sydenham. Therefore the list usually looks like this: 2 NATTA 1 NAT 1 SAVF

etc. These lists take some considerable time to deal with as the coins have to be looked up in RRC, all possible ambiguities sorted out, and then input to a separate datafile and ‘uploaded’ (see page 12). 4. typed lists. These are usually hoards which have been prepared for publication. For example, most of the hoards that appeared in RRC, Table L, are typed up. 5. letters, lists or computer listings that have been sent to Crawford from other numismatists. Not all the coin hoards in RRCH are contained in this filing cabinet and not all the hoards in the filing cabinet are in RRCH. Those hoards which are not in RRCH but are in the filing cabinet also have a card in Crawford’s card index. This index has a card per hoard for finds which have come to his attention since the publication of RRCH. These cards may simply have a reference, or may have a detailed list of coins if this is small enough to fit onto one card. It follows, therefore, that not all the cards in this index have entries in the filing cabinet. For those hoards which are either in RRCH or the card index or in another source but not in the filing cabinet the original report had to be located. For many of these they were obtained from either the excellent library of the Royal 8

2.2. Roman Republican Coin Hoards and the chrr database

Numismatic Society or the library of the Department of Coins and Medals. There were some, however, not available in either which were obtained from other sources. Other sources of hoard information As well as this information there are a number of other corpora. These include published volumes by Săşianu (1980), Chiţescu (1981), Blázquez (1987–1988), Chaves Tristán (1996), Backendorf (1998), Paunov & Prokopov (2002) and Moisil & Depeyrot (2003). Sometimes these volumes are cross referenced with RRCH, or are in Crawford’s records as described above, but there are often difficulties. For example, in the volume by Chiţescu (1981) references are sometimes given to RRCH. There are many more hoards from Romania listed by Chiţescu than are in RRCH but there are two hoards in RRCH which do not seem to be in her corpus. Romanian hoards frequently have two names especially when they come from Transylvania which was for a long time part of the Austro-Hungarian Empire. For example, RRCH 411 is recorded as a hoard from Nagykágya (nag), Hungary, and contained 169 denarii, 22 barbarous imitations of denarii and was found with silver ornaments. Chiţescu (1981) records the hoard under the name of Cadea, the Romanian name for the town and lists 171 denarii of which there are records for 131! Similar difficulties occur when a single place, such as Padova, has multiple hoards and confusions exist as to which hoard is which. Occasionally, therefore, it seemed as though new hoards not in Crawford’s records had been published, but in reality I already had the information under a different name. Careful cross-checking was necessary. As well as cross-referencing the hoards by name there was also the task of cross-referencing them by the various corpus numbers, e.g., RRCH, IGCH (Thompson et al. 1973), Săşianu (1980), Chiţescu (1981), IRRCHBulg (Paunov & Prokopov 2002) and Moisil & Depeyrot (2003). I have provided a concordance between RRCH and my database codes and the list of hoards (see Appendix B). Once we have some information about a hoard, there are various problems which have to be taken into account. One major area of concern is that very few of the hoards have been recovered in controlled conditions. For example, the Montiano hoard (RRCH 266, mnt) was found during agricultural work; the hoard from Barranco de Romero (bdr) was found during building works. The Cosa hoard (cos) is an exception as it was found during an excavation. Many hoards have no details at all about their origins. Crawford records if the hoard is out of a larger one, or is x coins out of a known number. For example Oleggio (RRCH 241, ole) has 228 denarii out of 527 ‘but including all issues represented in the total’. If the coins for which we have a record are a random selection from the hoard then there are no problems. However, if the list represents a carefully selected collection of coins from the hoard then this could easily create problems during the analyses. In many ways there is less distortion of the composition of the hoard if the hoard was found by workmen when compared to selected sorting by a numismatist. When hoards are seen to be ‘odd’ in an analysis, the first question must be ‘what is the data quality like?’ Another severe problem is locating the hoard’s find spot. The locations of hoards in Spain and Romania is not a major problem as Blázquez (1987–1988) and many Romanian scholars publish distribution maps (e.g., Glodariu 1976;

9

2. The data

Moisil & Depeyrot 2003). Backendorf (1998) has published maps of the hoards from Italy excluding Sicily. Many of the hoard names are of very small villages and this can be a problem — often the original publication has to be consulted. For example, of the 24 hoards in RRC (Lockyear 1989, Fig. 2.1) locations of eleven were found in atlases, one in Coinage and Money (Crawford 1985), six in a world Gazetteer, four by going back to the original references in the British Library, one from Mackenzie (1986) and one in an Italian directory of postal codes. For the present work the maps are presented only when they add to the interpretation of an analysis (e.g., section 5.4.7). Probably the most difficult and error-prone exercise of all is when there are several different sources of information for the contents of a hoard which do not agree with each other, or have somehow to be combined. The hoards published in RRC (Table L) are a good example. Table L does not have a complete listing of coins: only those after RRC 157 bc and before the Principate are included. For most of the hoards in Table L the data are incomplete and the information already in the RRC database (Lockyear 1989) had to be correlated with the records in the British Museum or publications. In most cases this was merely tedious, but in some it was difficult or impossible. For example, the Alvignano hoard (RRCH 417, alv) is recorded as having 2,317 denarii, 1 victoriatus and 3 quinarii in RRCH, as well as four extraneous denarii. However, the original RRC database (Lockyear 1989, Appendix C), and Table L from which this was derived, lists 2,334 denarii. The typescript in the British Museum has the total 2,321 (all the coins) written at the bottom, which agrees with RRCH, but when the list is added up it comes to 2,338 (all coins). Backendorf (1998) uses the data from RRC and publishes the total as 2,334. As this hoard is otherwise unpublished, either the total is wrong and was carried over to RRCH or the coin list contains a mistake which was carried over to RRC; it is impossible to tell. The updating of the database in 2006 was intended to be a ‘quick and dirty’ grab of extra data from three new corpora but it was more difficult than expected. The data given by Backendorf (1998) proved most useful. Cross-referencing the database and that volume was difficult as he often uses different names and no concordance to RRCH was provided. In some cases the only data available to Backendorf were the incomplete listings given in RRCH or RRC, whereas fuller listings are included in the chrr database, which was derived from from Crawford’s records. The biggest problem were hoards listed by Backendorf also included in RRC, Table L as they can be contradictory. In a few cases I have used Backendorf’s listings and these are noted in Appendix A. In the process of computerisation some small errors were noted in Backendorf’s listings and these have been corrected, with a note made in the database’s log books. The reasons for differences between Backendorf’s listings and others is usually explained in his extensive footnotes. The corpus by Paunov & Prokopov (2002) for the Bulgarian material is a welcome addition. Unfortunately, inconsistencies in the data presented in Tables 1–15, and between those tables and the main catalogue, resulted in no new additions to the database. It is hoped these problems can be resolved and new data will be added when available. The final corpus, that by Moisil & Depeyrot (2003) can be seen as an update to that by Chiţescu (1981). Unfortunately, many errors in the latter work have been carried over to the new volume, and many more new errors were introduced. 10

2.2. Roman Republican Coin Hoards and the chrr database

The database’s coverage of Romania was very good in any case, and the majority of the hoards could be cross-referenced to the new volume. A few new hoards, where the problems were obvious typographic errors, were input, but most of the new hoards will have to be input from the original publications.

2.2.4

Problems in the computerisation of coin types

To catalogue the coins found in a hoard scholars have used one of many different catalogues of coins of the Republic. These are of two sorts: catalogues which describe the different types struck by the Republic, and catalogues of specific collections of coins. The former are generally of more use unless the collection is very large and comprehensive. Many hoard reports use one (and sometimes more) of the former type of catalogue. Occasionally the latter type is used, usually Grueber’s Coins of the Roman Republic in the British Museum (Grueber 1910). The data for coin types used in the chrr database are derived directly from RRC and RIC 1(2) . The information included consists of the catalogue number, the date range and the denomination of the coin. Alternative dating schemes etc. are stored in separate data tables and can be linked to the main tables via the primary key, a unique number given to every coin type. One difficulty in computerising the hoard data is the fact the coins can be listed by the various catalogues mentioned above, or may only be listed by a description, and these have to be converted to RRC references. Those listed by descriptions only can vary in quality. For example, the hoard from Valdesalor (vld, Callejo Serrano 1965), has very full-looking descriptions, but when it comes to cataloguing them according to RRC it is very difficult to assign some of the coins to a precise reference. This is not helped by the fact that they are not in any order, so that identical coin types are spread over the report, and the quality of the photographs is so poor they are of little use. It is possible to describe Republican coins with very few details (as Crawford does on his handwritten lists) and get to an exact reference, but this can be difficult to decipher at first when one is less than totally familiar with the material. Even when a published list uses RRC references, they are not always complete. For example, the references used by Crawford in RRC Table L do not relate exactly to the main catalogue: RRC 197/1 in Table L is an incomplete reference as there are in fact types 197/1a and 1b. When entering the data these discrepancies have to be picked-up and corrected and this was achieved originally by using dBase program, and latterly via SQL queries. Cross referencing RRC with the other catalogues of coin types required some care. Sometimes coin types in RRC simply do not exist in other catalogues. If the report states that there is a variation on an older catalogue reference it may be possible to give that coin an RRC reference. Some coin types are subdivided into a large number of separate types in older catalogues but are grouped under a smaller number of types in RRC. For example the denarii of L. Piso L.F. L.N. Frvgi only have one reference in RRC (340/1) but can be between 650 and 671d in Sydenham (1952), and Calpurnia 6–11 in Babelon (1885, 1886). Although there is a minor loss of information converting the reference from these older catalogues to RRC, this is not particularly a problem. The reverse situation, when issues in an older catalogue are divided into several RRC issues, is more of a problem. Usually, these are minor variants of issues which are given an

11

2. The data

alphabetic sub-division of a reference, e.g., RRC 197/1a and 197/1b In this case the coin is given the first possible RRC reference and this fact is noted via the concept of an ‘accuracy code’ (see page 17). These codes are a method whereby the reliability and/or accuracy, and/or status of a coin can be encoded in the database. Concordances exist between the major catalogues, such as Sydenham (1952) and RRC, but for some of the older catalogues no concordance yet exists. In these cases the coins in the hoard lists were identified manually, rather than attempt to construct a concordance for the whole of that catalogue. These references were then added to the coinrefs table — see below, page 16. Some coin types are difficult to distinguish on the basis on either descriptions or early catalogue references. This is especially true of types from the earlier periods. The various anonymous dioscuri denarii are generally impossible to separate out unless the hoard had been examined by Crawford, or by another reliable numismatist, since the publication of RRC. Crawford has only identified these denarii to exact types for the earlier hoards when he was studying their chronology. It should be noted that as well as the those coin issues where all the types are anonymous, there are a number of anonymous denarii ‘in the style of. . . ’ For example, RRC 110/1a is an anonymous dioscuri denarius with a wreath; 110/1b is ‘similar, but no wreath.’ Even with anonymous denarii with symbols there can be some confusion. For example, there are two denarius issues with a crescent (RRC 57/2 and 137/1). Some reports list symbols not listed in RRC and these are sometimes difficult to assign to one of Crawford’s symbols, e.g., denarius ‘with flower.’ Wherever there is a major ambiguity the coin type has been recorded as being the first possible type with an accuracy code of 2 (this is equivalent to ‘as. . . ’, Reece 1975). As a result most of the anonymous dioscuri denarii are recorded as being ‘as RRC 44/5’, the earliest denarius issue which is dated to 211 bc. This leads to either a) an unrealistically high peak of coins in 211 bc when the hoard is graphed (if all coins are used) or b) hoards which appear to have far too few of the earliest coin types when only coins that can be identified precisely, or almost precisely (accuracy codes 1 or 5, see below), are plotted. This means that any analysis performed on these earlier coins and hoards would have to use quite general information, or be very careful as to which hoards are used. As well as these anonymous issues there are some later issues that are frequently lumped together by older reports, but have been separated out by Crawford. For example the various CN. DO, CN. DOM, CN. DOMI issues (RRC 147, 261 and 285) are often conflated in the earlier reports. Issues of Caesar, Mark Antony etc. are often also difficult to assign to correct references. As can be seen the use of secondary data is fraught with problems. However, if one wishes to take a broad overview of any field in archaeology then their use cannot be avoided It is necessary, therefore, to be constantly aware of the limitations and problems with the data available.

2.3

Storage and data manipulation strategies

At the time of the original project, the only well designed relational database of coin assemblages known to the author was that by Ryan (1988). The structure of the chrr database is an adaptation of that database. The chrr database 12

2.3. Storage and data manipulation strategies

discussed below has a highly flexible structure which has allowed the easy manipulation and extraction of data. Although constructing the database has taken some time, the extraction of data for analysis was made easy by the design and vindicates the effort taken. The database used by Backendorf was also originally implemented in dBase. In this case all the data were in one huge table. This led to a huge replication of data (‘data redundancy’ in database jargon) and difficulties in manipulating it. For example, calculating the percentage of a type from a hoard was done manually (Backendorf, pers. comm.). By separating the coin types and coin references into separate tables, it was possible to check the accuracy of data input to the chrr database before uploading the data to the main tables. In this way, many minor errors in Backendorf’s (1998) listings were identified. Volk’s database of Republican coin hoards is extremely similar in many respects to the chrr database described here (Volk 1994–5). Volk’s database was also originally implemented using dBase III+, but was imported to Paradox for Windows in c. 1995. Despite the obvious advantages of relational database structures, few numismatists seem prepared to expend effort in the correct structuring of their data (e.g., Guest & Wells 2007, data available from http://ads.ahds.ac.uk/ catalogue/search/fr.cfm?rcn=IARCW07-1; cf. Lockyear 2007b). This is shortsighted and the difficulty many then experience in using the data effectively, compared to the relative ease that data can be extracted from a database such as that described below, amply illustrates the value of careful database design.

2.3.1

The database

Where possible, each hoard was input directly into the database. This was achieved in three stages: 1. for each hoard, data were input to a separate small data file using whatever catalogue references (if any) were given 2. these data files were then ‘uploaded’ to the main database, either by a specially written dBase program or, after the database was imported to Access, by importing the Excel spreadsheet. 3. Various checks were then undertaken to identify any errors. This three-stage strategy was adopted for a number of reasons: • it enabled the hoards to be input with whatever catalogue had been originally used and then converted and uploaded later; • by having small data files this did not slow down the small 8088 single disc portable computer used for the majority of the data capture; • by having a conversion/uploading process the data could be checked for publication and input errors; • data in difficult formats could be input leaving the conversion process for later thus minimising the amount of time required in the British Museum.

13

2. The data

Even using this system there were many hoards where the data were not amenable to immediate input to a data file, usually due to a lack of any reference numbers. These were photocopied and the list constructed from RRC by hand. The original DBMS used was dBase III+. Although dBase III+ is not a relational database management system (RDMS), the structure of the chrr database largely conforms to the ‘normal forms’ required of such a database (Date 2000; Rolland 1998). There were a few fields included which should not be part of a true relational database structure but were created for a variety of ad hoc purposes. These are not included in the available version of the database. A suite of database programs were written to facilitate the use of the data. Subsequent to the transfer of the database to Access these have largely been replaced by SQL queries, or by manipulating subsets of the data in the spreadsheet package Excel or the statistical analysis system R. The chrr database consists of three main tables, and a number of subsidiary ones. The one serious break with the relational database model was the use of a free-text memo field in the findspot table for the recording of various miscellaneous facts. The main tables are: cointype this table contains data concerning the coin types under consideration. At present the basis for this table is RRC and RIC 1(2) . Additional coin types include general categories. Fields include: ctype

numeric primary key. This is an internal number and would not generally be used directly.

issue

Crawford issue number. For example, P. Crepusi is issue 361, whereas individual coin types have a full reference of 361/1a etc. This enable the extraction of hoards by the overall issue.

date_from the earliest date for that coin type. date_to

the latest date for that coin type. Frequently this field and the previous one are identical. Dates are derived from RRC and RIC 1(2) . Alternative dates, such as those suggested by Mattingly (2004) can be stored in separate tables for comparative analyses of dating schemes.

denom

denomination, i.e., denarius, victoriatus. Four letter codes used which link to the denom table.

by

the issuing authority e.g., Roman Republican, Roman Imperial, Greek etc. Helpful with general categories.

This table consists of 2962 tuples of data (as of December 2006), the majority from RRC and RIC 1(2) . coins

this contains the detailed lists of hoards. Fields include: sitecode a three character alphanumeric code for each hoard. Links to many tables especially findspot. ctype

numeric code for the coin type. Links to the cointype table described above.

14

2.3. Storage and data manipulation strategies

total

total number of coins for that coin type, hoard, and accuracy code.

accuracy a single digit numeric code as discussed in detail below. The primary key for this table is a combination of sitecode, ctype and accuracy. As of December 2006 this table contained 29,966 tuples of data. findspot this table contains general details about the hoard and its find spot etc. Some of the fields should not be part of the database in a strict interpretation of the relational database model. Most of these irregular fields either contain data derived from the other tables in the database and are a legacy from the dBase implementation. Most have been deleted from the version of the database available for distribution, and are excluded from the description below. The fields are: name

the name of the hoard.

country

some countries have been sub-divided e.g., Italy and Sicily. Countries formed after 1989 have been left as pre-1989, e.g., Former USSR, Yugoslavia, Czechoslovakia.

page

page in the project archive logbooks. Every hoard in the database has an entry in the log books recording any relevant information especially as regards the status of the hoard in the database and any problems that might exist with the hoard.

weights

logical field — are coin weights given in the publication? As table coins has one tuple per coin type, not per coin, it is not possible to store weight information there. For part of the original project, publications with weights were needed and this field enabled them to be found quickly, although as this information was only collected relatively late in the day, a ‘no’ means ‘unknown’.

extract

logical field — used to enable easy selection when the hoards required do not fit any easy extraction criteria.

findcode the primary key for the hoard. This is a three character alphanumeric code and links to the findcode field of the coins table. source

this gives the source of the coin list input to the database (therefore not necessarily the publication). For example mhchw is one of Crawford’s handwritten lists without reference numbers.

notes

originally a dBase memo field containing odd notes usually for parts of the database yet to be implemented. This field is not part of the relational database design but is a development aid.

As of December 2006 the table contains 643 tuples of data. 15

2. The data

These three tables contain the main data and can logically stand on their own. However, there are a number of other tables, the most important of which is: coinrefs this table is a concordance between the internal reference numbers used in the coins and cointype tables and the various catalogues of Republican coins. Coin catalogues currently included are RRC, Sydenham (1952), Babelon (1885), Babelon (1886), Grueber (1910), and the Augustan issues from Mattingly & Sydenham (1923) and RIC 1(2) (Sutherland 1984). Selected parts of Riccio (1843), Cohen (1857) and Fabretti (1876) have been input as necessary. Fields include: ctype

the unique number in the cointype table used in the hoards table.

cat

the catalogue, e.g., syd represents Sydenham (1952).

name

this is the name of the moneyer, or Emperor, or moneyer’s family. For most catalogues this is not necessary and the field is set to n/a. It is necessary, for example, for Babelon (moneyers family) and RIC 1(2) (Emperor).

denom

with some catalogues, notably RIC 1, first edition (Mattingly & Sydenham 1923), the same reference number has been given to several denominations and thus the necessity for this field. Where the field is unnecessary it is set to n/a.

ref

the reference number (e.g., 340/1).

accuracy see below. This table is essential for the uploading and conversion process as the temporary data files use the original catalogue references. The table, as of December 2006, contains 13,231 tuples of data. This table can be used as a concordance between RRC and the other catalogues but cannot necessarily be used as a direct concordance between two other catalogues without careful consideration of the accuracy codes as discussed in detail below. Other tables include: kris_cat this is a table of my general coin categories such as ‘misc. unidentified Republican asses’. For details of this data see below. Fields include: ctype

links to cointype table

descrip

description of the general category

ref

unique number

accuracy accuracy code. This is always 8 — see below. As of December 2006 there are 49 tuples in this table.

16

2.3. Storage and data manipulation strategies

denom

the meaning of the denomination codes used in the cointype table. Fields include: denom

the four character code

name

the denomination

accuracy_cd the meaning of the accuracy codes outlined below corpora relates site code to various corpora, e.g., RRCH, Chiţescu (1981). hrdrefs links site codes to BibTEX data files for generation of bibliographic references. Many of the original programs used temporary datafiles to speed up data manipulation or for import into other computer packages but these have now been rendered obsolete. Construction cross-tabulations (also known as contingency tables or pivot tables) is extremely easy in Access. Recently, it has been found useful to copy the results of queries to Excel files as an intermediate step when, for example, importing data to R or canoco or creating the tables for this book.

2.3.2

Accuracy codes

The major problem of accurately identifying coins either from the lists or from the coins themselves as discussed above was tackled by the use of an accuracy code.1 This idea was originally developed by Ryan (1988) and modified for this project. It is a variant of the recording terminology used by Reece (1975) which has become standard for the publication of coin lists from archaeological excavations. Each entry in the coins table has an accuracy code attached to it giving the status of the coin or coins. The four main codes are: 1 exact reference; i.e., the coin is 361/1a; 2 as reference; i.e., the nearest reference obtainable is 361/1a but it need not necessarily be that coin; 3 copy of 361/1a; i.e., a copy of exactly that coin; 4 copy as 361/1a; i.e., the nearest genuine coin that the copy may have used as a design. There are, however, a large number of references for coins of this period which cannot be converted to exactly one RRC reference, but can be assigned to one issue. For example, if the reference given in the hoard report is to Babelon, Valeria 18, this can be either be RRC 474/2a or 474/2b. The difference between these coins of L. Valerivs Aciscvlvs is minor: they both have an owl with a Corinthian helmet, a shield and spear(s) as the reverse type: 474/2a has one spear, 474/2b has two. They are minted in the same place (Rome) and at the same date (45 bc).2 It would therefore be a great waste of information to record 1

These were originally called ‘query codes’ but this created a confusion between the codes and the concept of a database query, so the name was changed to better reflect the function of these codes. 2 It should be noted that Babelon divides the issue into two (Valeria 18 and 19). Valeria 19 is the equivalent of RRC 474/2c. However, in Crawford’s concordance he only records these two types of Babelon as being 474/2.

17

2. The data

such a coin as code 2, but inaccurate to use a code 1. Therefore I have introduced a further code: 5 almost exact reference; i.e., the coin is of the same moneyer, place of issue and date. Only in very rare cases is this not the same issue. In the above example a coin with an original reference of Valeria 18 would have an internal reference 1913, accuracy code 5 (i.e., almost exactly as 474/2a). It therefore follows that all references with a code 2 must have enough doubt around them for them not to be assigned to this code. This can occur when a) the coin is badly worn or damaged or b) the description given is not detailed enough to assign it to an issue. The latter is common with the early issues as discussed above. For example an anonymous dioscuri denarius with cornucopiae (either 58/2, 207 bc or 157/1, 179–170 bc) would be given a code 2 when the only information provided is the description. This coding took place at two different stages: during data input to the small temporary files and during the uploading process. In the first stage case the code is given as a result of information in the hoard list. The second stage case took place during the original uploading process in dBase where the program used updated the code if necessary. The program could only decrease the data quality (i.e., a coin coded 1 could be changed to a 5, but a coin coded 2 could not be). This was achieved via the accuracy column in the coinrefs table. Therefore, if the hoard contains a coin with a reference ‘Babelon, Valeria 18’, it was entered to its temporary table with an accuracy code of 1. This was appended to the main tables during a run of the uploading program with the correct code of 5 (ctype 1913). This is achieved by every entry in the coinrefs table having an accuracy code attached to it indicating how accurate the cross reference with RRC was. It therefore follows that all RRC references in the coinrefs table have an accuracy code of 1. More recent additions to the database in 2006 were all from publications using RRC as the primary catalogue which made the task somewhat easier and was achieved using an SQL append query. Data for the coinrefs table was scanned into text files from the concordances in RRC. This had to be carefully edited as the variable base lines used in the publication caused some problems. The scanned files were input to a dBase file and then the final table constructed using a program. In many cases the RRC number in the concordance was incomplete or did not exist. These entries had to be checked by hand and the appropriate action taken. Where entries did not exist they were obvious misprints or errors. Where the reference was incomplete this was due to it having one of several possible minor differences (e.g., 474/1a or 1b), and these were assigned an accuracy code of 5. In some cases (e.g., Babelon, Valeria 19) it is possible to obtain a more accurate reference by referring back to the original catalogue.3 Some other codes are also used. Coins with an accuracy code of 6 are those which are considered by Romanian scholars to be copies, e.g., the coins from Poroschia (prs Chiţescu 1968b, 1980). This enables data sets including or excluding these coins to be easily extracted for comparative statistical analyses. Other codes include: 3

Referring back to the original catalogue in all cases would be extremely time consuming and given the slight increase in accuracy in was decided that it was not economical to do so unless the error was noticed during some other task (such as writing this chapter).

18

2.4. The coverage of the database

code

meaning

1 2 3 4 5 6 7 8 9

exactly identified coin inexactly identified coin copy of a specific coin copy of a general type of coin of which the reference is an example Almost exactly identified coin, e.g., either RRC 408/1a or 408/1b Coin in a Romanian hoard which is suspected to be a copy considered extraneous, usually by Crawford a general coin type, e.g., miscellaneous Iberian denarius total in hoard unknown, i.e., only presence/absence of type

Table 2.1: Meaning of the various accuracy codes used in the chrr database.

7 considered by Crawford to be extraneous; 8 general reference; see table kris_cat; 9 entry in the total column of the coins table only indicates presence/absence. This is usually set to 1. A summary of these codes is provided in Table 2.1.

2.3.3

General Categories

A method by which general entries such as ‘otherwise unidentified Republican denarii’ or ‘denarii of Juba I’ was developed. These general categories are stored as references to a pseudo-catalogue ‘KL’ which has its data stored in table kris_cat. This table is of a form compatible with the coinrefs table and has been appended to it. However, for ease of development it is also stored as a separate file. Being a member of a general category such as this is somewhat different from being ‘as’ or ‘almost. . . ’ (codes 2 or 5, see above), and thus all genuine coins recorded as having a KL reference also have an accuracy code of 8. Copies of general categories are recorded with an accuracy code of 4.

2.4

The coverage of the database

No attempt was made during the process of uploading hoards to ensure that the coverage of the database was representative either by country or by period. Hoards were input when sufficient information regarding their contents were available with priority given to hoards with more than 30 denarii. There are also serious problems of coverage by country due to regional traditions in publication. For example, Romania stands out as having an enormous number of hoards compared to other areas. However, the relative lack of Bulgarian hoards is due to a lack of publication, especially in the detail required for uploading to the coins table. This problem is exacerbated by the author having worked extensively in Romania. In Italy, the tradition was to not to publish hoards of this period in detail although this is partly offset by the large number of lists prepared by Michael Crawford and more recent series of publications. Chapter 4 discusses this matter in detail. The detailed analyses do not, however, require that the coverage across regions be representative in terms of the number of hoards provided the 19

2. The data

quantity of hoards at each period was large enough that the composition of the hoards could be seen to be representative.

2.4.1

Future development

As yet, many pieces of subsidiary data are not input to the database. For example, was the hoard found in a pot or with other finds? What year was it found and how? Much of this is contained in the project log-books but needs computerisation. A database such as this can never be ‘complete.’ New hoards, new publications and old records continually come to light. This database will be periodically updated. Much information has yet to be converted into the detailed format of the coins table, although more than enough data has been input for the purposes of the present project. Appendix A contains the list of hoards currently held on the chrr database, relevant information and their status. I would be very grateful to be notified of any errors in the database, and of any new data that could be included.

20

3

Models of coin supply and circulation

3.1

Introduction

Before starting to analyse the hoard data we need to have some models as to how the supply and circulation of coinage might be reflected by coin hoards. It was this problem that was examined by computer simulation (Lockyear 1989, 1991). Here I wish to outline some models of coinage supply and circulation, and to predict how they might be reflected. For the moment, we can assume that hoards are a random selection of coins from the local coinage pool (Thordeman 1948). Variations within the hoards across time and space should therefore represent variations within the global pool.

3.2

The life of a coin

The life cycle of a coin can be represented as shown in Fig. 3.1. The figure shows the various stages of production, supply and use and was used as the basis for the computer simulation. A similar model was presented by Haselgrove (1987). The first parameter that affects the contents of hoards is the numbers of each type of coin struck. This will be determined by a variety of factors, foremost of which are the availability of bullion and the political desire, or need, to mint coin. The relative sizes of issues in the coinage pool at a certain date can be seen by examining hoards of that date. This does not give the relative sizes of the issues as struck. Each year a proportion of the coinage pool was lost (see below). The absolute numbers of coins struck is difficult to calculate. It has been proposed, although not universally accepted, that the global coinage pool grew through time (Hopkins 1980; cf. Buttrey 1993; Lockyear 1999). These factors combine to make comparisons of the absolute sizes of coin issues difficult. One method which has been used to estimate the absolute size of coin issues is to estimate the number of dies used to strike an issue and to multiply that by a constant. A variety of formulæ and methods have been used (Lockyear 1996b, pp. 93–5). This procedure has been strongly criticised (see Chapter 8; Buttrey & Buttrey 1997; Buttrey 1993, 1994). Once a coin issue had been struck, it is possible that it was not all released by the state at once. However, it seems probable that coins were struck to meet state expenditure demands, and that rarely was there much delay in its release. The global coinage pool is defined as all the coins in the area under study at one time. Within this there will be local variations — that is, a series of local

21

3. Models of coin supply and circulation BULLION

NUMBER OF DIES

NUMBER OF COINS MINTED PER DIE NEW COIN

DELAY IN RELEASE OF COIN FROM MINT

TOTAL COINAGE POOL

DISTANCE FROM THE DISTRIBUTION POINT(S)

REMINTED?

SPEED OF CIRCULATION

LOCAL COINAGE POOL

CASUAL LOSSES

COINS WITHDRAWN? COLLECTION FACTORS

TYPE ONE HOARD

TYPE TWO HOARD

RECOVERY

NON-RECOVERY

LATER (ARCHAEOLOGICAL?) RECOVERY

SITE FINDS

HOARDS AVAILABLE FOR STUDY

KL 1990

Figure 3.1: Model of coin circulation (from Lockyear 1991, Fig. 28.9).

coinage pools. Some of the possible factors which will create or destroy local variation are: 1. Distance from distribution points. If the local pool is a long way from the initial distribution points it will take longer for it to receive the newest coins than areas near to those points. 2. Speed of circulation. The faster coinage circulates the sooner its distribution will be even. 3. Time. The longer the period since issue, the more likely the distribution of that issue will be even. These three factors were combined in the simulation model as the introduction delay (Lockyear 1991). 22

3.2. The life of a coin

Having entered circulation, coins usually fall slowly out of circulation. This can happen as the result of: • accidental losses (i.e., dropped coins); • accidental non-recovery of hoards; • deliberate disposal, e.g., burials or ritual hoards; • melting down of coins for bullion; • export of coins to areas outside of the core area of coin use, e.g., across the Rhine frontier into Germany. Sometimes coin issues fall out of circulation quickly. This can either be via the recall of coinage by the state, or by deliberate disposal due to demonetization of an otherwise worthless, debased, coinage. The rate of loss is known as the decay rate (Lockyear 1991), τ2 (Goulpeau 1981), the sink rate (Volk 1987), the wastage rate (Creighton 1992a; Duncan-Jones 1999) or the attrition rate (Buttrey 1993). The model also provides two extreme theoretical mechanisms for the collection of hoards. The Type One (or emergency hoard) is when the hoard is collected in a relatively short period of time. An example would be a day’s takings from a market stall. The Type Two (or savings hoard) is when the hoard is formed over a longer period of time, e.g., coins being saved for a dowry. computer simulation Computer simulation was used to analyse how these factors would affect hoard structure. It was found that, at some periods, variation in any of these factors — the introduction delay, the decay rate, or the collection method — could account for the observed variation in hoards (Lockyear 1989, 1991). The reason for this being applicable to some periods only became clear during subsequent work and will be discussed in detail in Chapter 7 (see also Lockyear 1993). Haselgrove’s model, which is very similar to this, was criticised by Creighton (1992a, section 2.12) as being ethnocentric. He claims that hoards are seen as an appendage to circulating money, whereas he would see them as ‘dynamic stores of wealth.’ He states that most coin would have spent the majority of its life in a hoard of some form. Creighton’s figure is reproduced here (Fig. 3.2). The hexagon represents a variety of exchanges with coin being kept in small quantities, such as in a purse, to large amounts, such as in an armarium or strongbox. Whilst accepting his criticism that hoards are given a too peripheral rôle I do not accept that this is ethnocentric. If we are to define hoards as widely as Creighton does, then there is little difference between his model for the Roman period and an equivalent model for today. The main difference is that the location of larger hoards has changed to shop safes and bank strong rooms. If we use a definition of money derived from neoclassical economics, we can add chequing accounts and computer memory to his list of static locations of money. His model (Fig. 3.2) is self explanatory. Note that the armarium, or store of wealth, need not be that of an individual. It could be an army pay chest, a tax collector’s revenues, or the state’s financial reserves. Important parts of the system are not represented well in either model. An important missing element from both models, especially when comparing different geographical regions, is that of supply. What mechanisms were used for 23

3. Models of coin supply and circulation

extraction

conversion metal

other metal objects

withdrawal

minting/issue

metal ores

circulation pool

somewhere else

in a pocket

on a counter

circulation pool

Armarium or store of wealth

temporary concealment in the hand

in a purse money looses function

loss

deliberate discard (votive offerings)

accidental discard (loss)

unrecovered hoards

abandoned hoards

Figure 3.2: Model of coin circulation (after Creighton 1992a, Fig. 21.02).

releasing the coin from the mint into circulation? How would coinage move from one area to another? Can we suggest how the different possible mechanisms would affect hoard structure? In the rest of this chapter I will propose, and discuss, different possibilities.

3.3

Supply and distribution within a discrete area

Firstly, we can suggest a model for a discrete area. This could be a small region, or even a town. We can suggest that new coin entering the pool would be issued at a point or points within that area. As an example, let us take a fort near a town. The soldiers’ pay is likely to contain a proportion of new coin. On 24

3.3. Supply and distribution within a discrete area

pay day, the distribution of that coin is limited to soldiers’ purses and private stores, including the fort’s central strong room. The coins have a highly uneven distribution. Over time, the soldier spends his money and the coins enter the ‘hoards’ in shops, bars and brothels. These coins are then in turn passed onto others. After a while, the distribution of the new coins in that town is reasonably even. Towns with no troops, or other reasons for official payments, will only receive these coins as a result of trade and other contacts. We can summarise this as follows: Stage 1: coins being struck; distribution limited to mint. Stage 2: coins used by state for payment; distribution limited to payees. Stage 3: payees use coins; distribution irregular within area where payees live and/or travel to. Stage 4a: coins have been used for a while; distribution within initial area now even. Stage 4b: coins used for a long period; distribution over large areas now even. Stage 5: coins no longer used and are destroyed/recycled, lost or thrown away. Stage 6: some coins recovered and the distribution now limited to loss sites, museums, collectors and archaeological units. The distribution of coins will initially be centred around the points of supply. If the points of supply are widely dispersed it may be difficult to see trends in the hoard data. If there are few points of supply, or those points of supply are limited to a small area, it should be possible see a trend in the distribution at first. A good example is coin hoards between ad 197–238 in Britain. Hoards near military centres such as Segontium or Hadrian’s Wall have a large proportion of the newest coins; hoards in the south have a low proportion (Creighton 1992a, Fig. 25.19). The edges of the coin distribution present some problems. Some ‘edges’ are definite borders beyond which coins are simply not present, or are used in a way that the number and type of exchanges, if there any, are quite different from the model proposed above. A good example would be the distribution of coinage beyond Hadrian’s Wall, or Roman coins in India. Supply to such areas is likely to be highly erratic. Another possibility is where a number of centres are producing coins as part of the same coinage system. Such multiple mint systems occur especially under the Empire in the fourth century. This can lead to the situation where coins of a mint become less common, in comparison to coins of a second mint, as one moves from the region around the first mint to the region around the second mint (see Fig. 3.3). There are cases where a centralised mint produces issues of coins which are exported to specific regions only. Walker (1988, 290–5) identified several issues of bronze coinage which appear to have been struck specifically for Britain, e.g., the Britannia asses struck by Hadrian in c. ad 119. As for the multiple mint system mentioned above, we would expect the distribution of these coins to ‘bleed’ into the surrounding regions over time. 25

abundance

3. Models of coin supply and circulation

A

distance

B

Figure 3.3: Idealised coin distribution between two mints (A and B) which are part of the same monetary system.

N

Figure 3.4: Down the line movement model. At each point coins can move in any direction including back towards the point of origin. This figure contains ten random ‘walks.’ Each starts from the same origin. The length of each step is constant, the direction random.

26

3.4. Types of supply

3.4

Types of supply

How does coin minted, for example in Rome, get to Spain or Romania? We can suggest three simplified possibilities. Down the line movement. This is the sort of movement described above. As coins are exchanged from their point of issue, a decreasing number will move further and further away from that point. The number decreases because some coins at each stage will move back towards the point of issue. Eventually, a physical, political or social border will be encountered restricting further movement. The random walk simulations which have been used to look at the creation of distribution patterns (Elliott et al. 1978) can provide a schematic model to demonstrate this (Fig. 3.4). Public supply. Some areas will receive coinage from the state as payments or subsidies. A common payment would be to troops stationed in an area. This coinage will then be released into the local coinage pool. Some of it might return to the area of production via taxation. The coinage used in payments would presumably consist of a mixture of older coins (from taxation), and the latest coins from the mint. If a hoard was recovered from this payment we could predict that it would have a higher than average quantity of new coin. Private supply. Coin may be moved large distances by private persons. These movements may be for a variety of reasons which include: • trade • loans • other types of exchange between private persons • with the owner; e.g., emigration, or the return of a soldier to his home. Unless the supplier has recently been paid by the state, we can suggest that a hoard from this supply would reflect the coinage pool in the area from which it was withdrawn.

3.5

Inter-regional patterns

If the contacts between the producer region and the target area are regular and large scale, we would expect, all other things being equal, for the composition of their respective coinage pools to be similar. The exception would be in the distribution of the latest coins where some regional dissimilarities would be seen. In the right circumstances, we would be able to see which areas were receiving official supplies of coin. What these ‘right’ circumstances are will be discussed in section 7.5, page 223. If contacts were irregular, or new, we could predict the following: 1. If the contact was in some way official, the coinage received would have a higher proportion of new coin than the coinage pool ‘at home.’ 27

3. Models of coin supply and circulation

2. If the contact was private, e.g., private trader, the coinage received would reflect the pool ‘at home.’ 3. If supply fluctuated, the pool would not reflect the pool at home as it would not receive coins in the same proportions as produced. These possibilities are a simplified set. In reality there will be mixture of the various factors and our best hope is that there is a dominant factor which will show through the likely ‘messy’ pattern.

3.6

Other factors

A number of other important factors also have to be considered. 1. Do the coins perform the same function from area to area? 2. Do they circulate at the same speed from area to area? If coins circulated more slowly in one region than another, the time taken for the distribution of a coin type to even out will vary. 3. Does the target area have coinage of its own? Are the systems compatible? Does it matter? In general, I have here stuck to talking about coins rather than money. In general, coins are are form of money, but money can take a wide variety of forms from coins to cowrie shells, from cattle to antique plates. I do not wish to discuss here approaches to the study of past and present economies, definitions of money and so on beyond to state that I am of the opinion that neoclassical economics is not universally applicable and we should be wary of making ethnocentric assumptions as to the functions of ancient coins. See Lockyear (1996b, chapter 2) for a more detailed discussion.

3.7

Hoards and the coinage pool

At the start of this chapter we made the assumption that hoards are a random selection of coins from the local coinage pool. This assumption is generally false. For example, hoards rarely mix denominations and/or metals. They are thus utterly unrepresentative of the ratio of bronze coins to silver, or silver coins to gold. In periods where there is rapid debasement, even coins of the same denomination but differing fineness may not be accurately represented. Within periods of stability, however, it appears that coin selection within a single denomination is random which allows us to examine the distribution of denarii during the later Republic with some confidence.

3.8

Summary

The models outlined above are a gross simplification of the real situation. However, using these models as a starting point we can start to analyse the available data, and then to attempt to interpret it. The analyses themselves might lead to further refinement, alteration or rejection of the these models, which in turn could lead to further analyses and interpretations. 28

4

The pattern of hoarding

4.1

Introduction

It is well known that hoarding, or perhaps it would be more accurate to say the incidence of the non-recovery of hoards, is not evenly spread over time or geographically. Speculation is common about why a specific hoard was buried where it was, and why it failed to be recovered. For example, the Brescello hoard (bre) was found in 1714 and is thought to have originally consisted of 80,000 gold aurei. The late Republican aureus weighed about 8.03g (Crawford 1974, p. 593) which means the hoard consisted of 642.4kg of gold, or just under two-thirds of a UK Imperial ton. At current prices (as of 30/11/2006) the gold bullion alone is worth £6.8 million. How did this huge quantity of gold come to be buried and lost? Was it someone’s entire personal fortune? Or perhaps a daring robbery? We will never know. If we can rarely know why an individual hoard came to be buried, we can, however, look at patterns in hoard deposition over time, and/or geographically and try to interpret the patterns we observe. In a classic paper, Crawford (1969a) demonstrated that the time distribution of hoards recovered in Italy during the Roman Republic was not even. Peaks in the incidence of hoards correlated well with historically attested episodes; e.g., the Social War (Fig. 4.1). A number of scholars have then tried to use the geographical distribution of hoards to plot the routes of invaders or armies. Kent (1988) neatly demonstrated, by the use of English Civil War hoards, that such a simplistic interpretation of the patterns found is unlikely to be true in most cases. Unfortunately, as discussed above, the chrr database may not be ideally suited to the examination of these distributions (section 2.4, page 19). The first task is, therefore, to test for possible problems in the database coverage.

4.2

Testing the coverage

The database contained, as of November 2006, some information about 643 hoards, although only 452 of those had detailed listings uploaded into the database. In the construction of the database hoards were uploaded when sufficient good quality data had been obtained. Effort was directed towards inputting as many hoards as possible within the time available, with an emphasis on Romanian material. In the following chapters I have restricted the analyses to hoards with more than 30 denarii and this reduced the number of available hoards to 294. This is still a reasonable sample size, although the uneven distribution of hoards over time results in some of the data sets being smaller than I would

29

4. The pattern of hoarding

218 to 216 215 to 211 210 to 206 205 to 201 200 to 196 195 to 191 190 to 186 185 to 181 180 to 176 175 to 171 170 to 166 165 to 161 160 to 156 155 to 151 150 to 146 145 to 141 140 to 136 135 to 131 130 to 126 125 to 121 120 to 116 115 to 111 110 to 106 105 to 101 100 to 96 95 to 91 90 to 86 85 to 81 80 to 76 75 to 71 70 to 66 65 to 61 60 to 56 55 to 51 50 to 46 45 to 41 40 to 36 35 to 31 30 to 26 25 to 21 20 to 16 15 to 11 10 to 6 5 to 3

0

5

10

15

20

25

30

Number of hoards Figure 4.1: Frequency of hoards in Italy during the later Republic (after Crawford 1969a).

30

4.2. Testing the coverage

like. The total number of denarii with an accuracy code of 1, 5, or 6, i.e., well identified, probably genuine coins, was 91,965. This seems a huge number, but only represents a tiny percentage of the total number of coins struck. This is not a problem as sampling theory shows that the absolute size of a sample is important, not the size of the sample relative to the parent population. Tables 4.1–4.2 present the total number of hoards by period and region contained in the chrr database as of the end of November 2006. The periods in the tables are those used in RRCH (Crawford 1969c). From Table 4.1 we can see that Italy, Romania and Spain/Portugal have the most hoards. This is a reflection of the real situation, although Romania is probably overrepresented and Spain/Portugal under-represented. Other regions are definitely under-represented, such as Bulgaria. Germany probably should have many more late Republican/early Imperial hoards than are included in the database. Within the three main countries, we can test the time distribution of hoards in the database by comparing them to published catalogues.

4.2.1

Italy

For Italy, including Sicily, Sardinia, Corsica and Elba, a reasonably representative catalogue is RRCH. For this area we can be confident that RRCH presents a true picture due to Crawford’s long standing personal involvement in collecting data from there. Table 4.3 shows the number of hoards with at least one denarius in RRCH and the chrr database. Comparing these two distributions using the Kolmogorov Smirnov two sample test (Shennan 1997, pp. 57–61) we can see that there is no statistically significant difference between the two at the 0.05 level.1 Looking in detail at the figures we can see that the distribution across periods is slightly more even in the database than in RRCH. This is due to an attempt being made to provide enough hoards for detailed analysis across all periods. The first and last periods (208–151 bc and 26–2 bc) are slightly under-represented. This is due to hoards with early Republican or Imperial issues being more difficult to input and upload to the database than those with only later Republican issues. The complexities of early catalogues of the Augustan coinage result in early hoard reports requiring much more manual intervention during computerisation.

4.2.2

Spain and Portugal

For Spain and Portugal we can compare the evidence with the catalogue by Blázquez (1987–1988) which expands the data given in RRCH. The data are presented in Table 4.3. Comparing the two distributions as above shows no statistically significant difference.2 Looking in detail, however, shows that the period 124–91 bc is relatively over-represented in the database, and 91–78 bc under-represented.3 1

The null hypothesis (H0 ) is that there is no difference. Dmaxobs = 0.044; Dmax0.05 = 0.144. We therefore accept the null hypothesis. 2 The null hypothesis (H0 ) is that there is no difference. Dmaxobs = 0.081; Dmax0.05 = 0.226. We therefore accept the null hypothesis. 3 The work by Chaves Tristán (1996) was not available to me at the time of this study but it is unlikely it would show any differences.

31

32 7

21

1

2

64

2

3 24

1 2

30†

1

4 1

13

1

124–92

3 1

3

150–125

57

102

2

3 11

1 2 1

1

4 42

5 1 3 1 28†

1

1

78–50

1 18

1

30†

1

1

91–79

56

1

3

2 20

1

1 3 1 2 1 19

1

1

49–45

100

2 3

3 2 1

5 29

36 1

6 3 3

1 1 3 1

44–27

44

3

1 1 3

1 17

1

7

3 3

3

1

26–2

452

2 17 48‡ 1 2 2 13

16 128

2 1

1 1 17 8 9 2 166 1

2 1 8 1

3

uploaded

191

1 8 19 1 1 3 21 1 2 2 10

1 3

6 20 5 7 1 54

1 1

17

2 3 1

not uploaded

643

5 3 3 1 25 1 1 1 1 7 37 13 16 3 220 1 1 5 1 1 24 147 1 3 20 69 2 4 4 23

total

Table 4.1: Number of hoards by region in the chrr database as of November 2007. Hoard closing date determined by the latest Roman coin in the hoard. Date ranges those used in RRCH with slight modification. † Includes one hoard with only presence/absence data. ‡ Includes one undatable hoard. Totals include ten hoards with no denarii (3 hoards of victoriati , 4 hoards of bronzes and 3 hoards of aurei ).

Total

uncertain Albania Austria Britain Bulgaria Corfu Corsica Crete Elba Former USSR France Germany Greece Hungary Italy Jersey Morocco Netherlands Is. Pantelleria Poland Portugal Romania San Marino Sardinia Sicily Spain Switzerland Tunisia Turkey Former Yugoslavia

208–151

4. The pattern of hoarding

4.2. Testing the coverage

208–151 uncertain Austria Britain Bulgaria Corfu Elba Former USSR France Germany Greece Hungary Italy Jersey Netherlands Is. Pantelleria Portugal Romania Sardinia Sicily Spain Switzerland Tunisia Turkey Fmr Yugoslavia Total

150–125

124–92

91–79

78–50

49–45

215

44–27

26–2

total

16

426

449 2633

80 1485

3839 3276

252 2506

40 5 166 1395 1119 24

168 2209

509

197 219 61

497

191

4

207

109

132 111

41

657 74 3 940 28 43 1 2969 151 992 52 60466 13 63 40 5902 8952 1413 2195 5543 61 115 132 1160

5342

9110

14631

22655

7663

27128

5392

91965

22 35

52 3 168 28

459

278

43

42 28

4595

4689

11621

198 12 157 1 16543

1 386 13 51 51 4091

1258 84 742

1127 42

16945 13

1954

2 1448 2

1 15

133 2

61 81 1389 18 311 74

115

44

Table 4.2: Number of denarii in hoards by region. Only denarii with a accuracy code of 1, 5 or 6 have been included. Region Italy etc. (RRCH) Italy etc. (database) Iberia (Blázquez) Iberia (database) Romania (database) Romania (M&D)

208–151

150–125

124–92

91–79

78–50

49–45

44–27

26–2

total

7 3 1 0 1 0

13 16 0 1 0 0

28 33 34 26 1 0

36 30 9 2 18 26

31 31 16 15 42 51

14 18 12 5 20 22

34 37 10 7 29 47

17 9 10 4 17 28

180 177 92 60 128 174

Table 4.3: Total numbers of denarius hoards by region. Comparison of catalogues (Blázquez 1987–1988; Crawford 1969c; Moisil & Depeyrot 2003) with the chrr database.

4.2.3

Romania

For Romania there are a number of catalogues we could use. RRCH underrepresents the Romanian material (Poenaru Bordea 1971) and Chiţescu (1981) is difficult to use for this purpose as the coin identifications are according to Sydenham (1952), and the hoards are listed alphabetically. The most recent catalogue by Moisil & Depeyrot (2003) is the easiest to use for our current purposes, although the large numbers of errors make it difficult to use in detailed analysis. Comparing the two distributions once more again shows no statistically significant difference4 although looking in detail shows that the last two periods (44–2 bc) are a little under represented. The two very early hoards (Bugiuleşti, bug and Olteni, olt) only consist of four denarii, two each. Both were categorised 4 The null hypothesis (H0 ) is that there is no difference. Dmaxobs = 0.072; Dmax0.05 = 0.158. We therefore accept the null hypothesis.

33

4. The pattern of hoarding

as individual finds by Chiţescu (1981) but as hoards by Crawford (1969c). As the database was partly created to examine the Romanian evidence, it is not surprising that it is very well represented in the database.

4.2.4

Other regions

No attempt was made to ensure detailed coverage of areas other than those discussed above. No other regions, however, have such large numbers of hoards with the possible exception of France at a late date. Some of the differences between countries are due to publication traditions. Bulgaria is severely under-represented in the literature (Crawford 1977a; Poenaru Bordea 1971) and the database. Until the publication of IRRCHBulg (Paunov & Prokopov 2002) there was no comprehensive listing of hoards from Bulgaria, although Crawford (1985, 328–9) provided a short listing. Unfortunately, even with these additional sources, it was impossible to improve the coverage very much with only eight hoards uploaded in detail to the database compared to the 132 listed by Paunov & Prokopov (2002). In contrast, Romania has a long tradition of detailed hoard reports and corpora of finds. Many French hoards are known from very old publications which are now being listed in the series Trésors Monetaire. Germany has few hoards of an early date, although later hoards are often published in detail. Although British hoards are usually well published, none contain only Republican denarii. The one hoard listed (Weston, wes, RRCH 476) contains three denarii hoarded with British coins and could easily be of a later date. To summarise, the chrr database does generally reflect the distribution of hoards over time although there is a tendency for periods with a low rate of hoarding to be slightly over-represented, and for the Augustan period to be under-represented.

4.3

The pattern

Figures 4.2–4.5 present histograms of the number of hoards per year using data from the chrr database. Comparison of the four histograms shows substantial differences. The absolute height of each bar is less important than the relative changes between bars. The Italian pattern shows major peaks in periods of civil or military unrest: the Social War (91–88 bc), Spartacus’ revolt (73–71 bc) and the Civil Wars of the 40s bc. This is the pattern as shown by Crawford (1969a) and discussed above. A visual comparison between the Italian graph presented here and with Figure 4.1 shows a substantial difference in the earliest period. This is because Crawford’s graph includes non-denarius hoards of which a large number are early hoards of victoriati . Why the lower fineness victoriatus should have been hoarded preferentially to the new denarius during the Second Punic War is unclear. Although the lack of hoards in the last period (26–2 bc) is slightly exaggerated, it is real. Guest (1994) shows that Italy all but ceases to hoard silver coins after the end of the first century bc. This is in contrast to peripheral areas of the Empire, for example Roman Britain (Robertson 2000). The Iberian peninsula material shows comparatively more hoards in the period 124–92 bc and comparatively few hoards from 91–79 bc compared to Italy. 34

7 6 5 4 0

1

2

3

hoards per year

8

9

10

11

4.3. The pattern

208

150

124

91

78

49 44

26

2

26

2

year BC

2 0

1

hoards per year

3

4

Figure 4.2: Denarius hoards per annum in the chrr database.

208

150

124

91

78

49 44

year BC

Figure 4.3: Denarius hoards per annum in the chrr database from Italy, including Sicily, Sardinia, Corsica, San Marino and Elba.

35

0.5 0

hoards per year

1

1.5

4. The pattern of hoarding

208

150

124

91

78

49 44

26

2

year BC

2 1 0

hoards per year

3

4

Figure 4.4: Denarius hoards per annum in the chrr database from Spain and Portugal.

208

150

124

91

78

49 44

26

year BC

Figure 4.5: Denarius hoards per annum in the chrr database from Romania.

36

2

0

1

hoards per year

2

3

4.3. The pattern

208

150

124

91

78

49 44

26

2

year BC

Figure 4.6: Denarius hoards per annum in Spain and Portugal from Blázquez (1987– 1988).

The peak for Italy in the period 91–70 bc has been explained by Crawford by the Social War and Spartacus’ revolt (Crawford 1969a). Two contradictory points could have relevance. Firstly, as both these events took place in Italy, it would be unsurprising that the peak is not large in Spain and Portugal. Conversely, these years, especially the 80s bc, also produced very large quantities of coin and one would have expected a moderate number of smaller hoards to have been dated to that period erroneously. If we compare Fig. 4.4 with Fig. 4.6 we can see that the database has fewer hoards of 91–78 bc than we would expect. The Iberian peninsula also seems to have a slightly less marked peak in the period 49–45 bc. The Civil Wars, which are cited as the cause of the peak in 49–45 bc for Italy, were also partly fought in Spain — the Battle of Munda took place in 45 bc (Scullard 1982, p. 142) — and Republican denarii were minted there in 46–45 bc (RRC 468–471, 477–479). Fig. 4.4 appears to be a little misleading here — as noted above the number of hoards from 78–45 bc in the chrr database is a little high. Plotting the numbers of hoards from Blázquez (1987–1988) produces Fig. 4.6, which has a much higher peak for 49–45 bc. For Italy, the numbers of hoards per year for 49–45 bc is just under 3.5 times that of the previous period, for Romania about 2.75 times, and for Iberia, using Blázquez’s figures, over 4 times. The detailed pattern is even more irregular than shown in Figs. 4.4 and 4.6. All Iberian hoards with 30+ denarii in the period 124–92 bc are dated to 115– 100 bc. In the period 78–50 bc a detailed breakdown of hoards of 30+ denarius shows that 9 of 10 are in the period 78–71 bc, compared to 21 of 29 Italian hoards. These differences between Italy and the Iberian peninsula cause difficulties in the detailed analyses as we are not always able to compare enough hoards of the same date from different regions — see sections 5.4.3–5.4.4 especially.

37

4. The pattern of hoarding

The Romanian graph shows comparatively large number of hoards in the period 91–79 bc. This is misleading. The 20 hoards dating to pre-78 bc contain only 169 ‘good’ denarii (i.e., well identified denarii with a accuracy code of 1, 5 or 6). Of these, 41 denarii come from the Bobaia hoard (bob) which closes in 79 bc (Chirilă & Iaroslavschi 1987–1988). It is extremely likely that these very small hoards were actually concealed later, probably mainly after 80 bc. In the period 78–50 bc substantial variation is masked by the large time period. Twenty-one of 29 Italian hoards over 30 denarii occur between 78–69 bc, compared to only 11 of 24 Romanian hoards. Conversely, eight of 29 Italian hoards occur between 63–51 bc, compared to 13 of 24 Romanian hoards. This concentration of hoards from 63–51 seems the more extraordinary when one considers the small size of the issues of this period. The dating of Romanian hoards is problematic due to the extensive copying of coins (Lockyear 2004, pp. 65–6). Coins of the 40s bc are common, however, in later hoards and this suggests that the earlier hoards were not concealed very much later than their closing date. Romania also has a large rise in hoarding for the period 49–45 bc. This is a real peak, with 20 hoards containing 1,485 good denarii. Although there were fewer hoards buried in Italy in this period, the size of the hoards is larger. Generally, the size of Italian hoards is larger than that of Romanian hoards. The average Italian hoard is 364 well identified denarii compared to only 70 in Romania.

4.4

Conclusions

At a general level, the incidence of hoards in the chrr database appears to reflect real patterns in the data although the large time periods used in these comparisons do mask some variation in the hoarding pattern. The Iberian pattern seems slightly different from the Italian despite both areas being under Roman rule at the period considered. The lack of a peak in the period 91–79 bc may be explicable in terms of political events. The relatively large numbers of hoards from 124–92 bc has been interpreted by Crawford as evidence of Italian and Roman settlers moving into Spain (Crawford 1985, p. 97–102) but this interpretation has not met with universal acceptance (Simon Keay, pers. comm.). The supply of coinage to Iberia seems to have been rather erratic. The Romanian pattern is more similar to the Italian, once the supply of coinage to Romania starts fully. This patterning however, masks important differences in the structure of Romanian and Italian coin hoards which is examined in the next chapter. Also, the problem of copies means that the dating of the hoards is not secure and the possibility exists that some may have been concealed much later.

38

5

Comparing hoards — Correspondence Analysis

5.1

Aims and methods

5.1.1

Introduction

When deciding on which techniques to use in any statistical analysis a number of considerations must be to the fore. Firstly, what is it we are trying to achieve — what is the question? Secondly, do the data meet the requirements of the statistical method to be employed? Lastly, is the technique likely to give optimal results? In the case of the data set at hand we have a very general question. Is there spatial and/or temporal patterning in the structure1 of coin hoards? If so, what is it and can it be characterised? Can we identify aspects of that patterning which are the results of less interesting factors, such as post-recovery biases, or the effects of sampling error, so that we can identify residual variation which could be more interesting? The neglect of the principles of formation processes in the interpretation of coin hoards, as in other classes of archaeological material, has led to serious errors in interpretation (Lockyear 1991, 1993). Of the many aspects of the data it would be possible to examine, it was decided to focus on one, variation in quantity of denarii in hoards. The original analyses (Lockyear 1996b, chapter 8) grouped the denarii by the dates given in RRC.2 For example, the hoard from Cosa (cos) has 6 denarii of 92 bc, 56 of 91 bc, 204 of 90 bc, etc. This is what Creighton calls the ‘age profile’ of a hoard (Creighton 1992a, p. 79). A criticism of those analyses, however, is that the dates Crawford assigns to his issues are not universally acknowledged, and are often far more accurate than the dating evidence warrants. The Mesagne hoard, in particular, necessitated a new dating scheme (Hersh & Walker 1984). It would be just as possible to look for variation in the distribution of particular coin issues but this has many problems such as the sparseness of the resultant tables and the large number of variables. As a result, for the original set of analyses I decided to stick to using years of issue (Lockyear 1996b, chapter 8). However, the structure of the chrr database allowed for data sets to be easily 1

The term ‘structure’ of a coin hoard is used to mean the pattern of certain classes of coins in a hoard. The classes could be exact coin types, dates of coins or mints from which the coins originated. My use of the term derives from the statistical literature where it is often stated that we are looking for latent structure in a data set (e.g., Wright 1989). 2 Where RRC gives a range of dates the start date of the range was used, field date_from in the chrr database.

39

5. Comparing hoards — Correspondence Analysis

created grouped by years or by issues and so it was decided for this study to analyse the data twice, by years and by issues.

5.1.2

Techniques

Simple picture summaries of the data are needed. Histograms are a powerful visual tool for examining data but are limited when several objects (hoards) need to be compared Lockyear (cf. 1989, figs. A2–A25). Reece (1974) used scattergrams for plotting this type of data and this method was followed by Lockyear (1989, figs. A26–A30). Ryan (1988) used cumulative percentage curves. This method, used here, is especially suitable for the representation of coin hoards, allowing the comparison of a number of hoards identified by colour or line style. More recently, Reece (1995) has developed a method in which some form of average is subtracted from the cumulative percentage curve for each site/hoard so that the graph represents deviation from that average. This type of graph requires some care in interpretation, however (Lockyear 2007b). The comparison of hoards using significance tests has been discussed and found to be less useful than might be initially thought (Lockyear 1989). The three main objections are: 1. Given our models of coin circulation, do we ever expect two hoards to be drawn from the same population? Although they may be drawn from the same global coinage pool are the local pools going to be the same? If we do not expect the local pools to be identical, significance tests where the null hypothesis (H0 ) is one of no difference, are not an appropriate method. 2. Given the wide range of sizes of coin hoards, it is difficult to see possible patterns above the effect of sample size (Shennan 1997, pp. 113–5). 3. Given the number of hoards, we need to assess the problem of multiplicity (Mosteller & Tukey 1977, p. 28f.). This becomes a problem of multiple comparisons (O’Neill & Wetherill 1971). In the following significance tests will only be used when appropriate and to answer a specific question. It would be possible to use contingency coefficients (e.g., Creighton 1992a,b) but this method is fraught with problems (Lockyear 1996b, section 3.12.5). They are used in a limited fashion on page 165. It was clear that some form of exploratory multivariate technique was required. Correspondence Analysis (CA) is a technique specifically for the analysis of contingency tables and is thus highly suited to the analysis of this type of data (Baxter 1994, 2003; Bølviken et al. 1982; Greenacre 1984, 1993; Wright 1989). It has been widely used in a wide variety of contexts (see various papers in Greenacre & Blasius 1994). I have used it with some success both in the analysis of hoards and in the analysis of site finds (Lockyear 2000b). Its advantages are that it provides information about the relationships between objects, and between variables, and when used with care between objects and variables. The results can be easily displayed as a scattergram for both objects and variables. This technique is not without its problems. Firstly, it is very sensitive to ‘odd’ samples and variables and these can dominate the results. Secondly, the technique is very good at seriating objects and variables — that is, putting them 40

5.1. Aims and methods

in some form of order according to their distribution along a gradient (Madsen 1989). This is often time, but can be space or social status. With coin hoards of relatively well dated material, this is not an advantage but a disadvantage. Lockyear (1989, section 2.4) found that it was difficult to find variation in the data set which was not connected to time. A worked example is provided below. Other multivariate techniques which have been applied to coin hoards include Principal Components Analysis (Creighton 1992a; Ryan 1988) and Factor Analysis (Duncan-Jones 1989). However, most of these analyses were performed before CA became popular in the late 1980s (Baxter 1994, pp. 133–139).

5.1.3

Software

The majority of the statistical analyses and graphs for this book were produced using the statistical programming system R which is freely available over the internet http://www.r-project.org/. CA is available in a number of R packages. I used the FactoMineR package. As discussed below, CA produces a number of ‘diagnostic statistics’ which help in the interpretation of the analyses but I found that none of the R packages produced these statistics in the form discussed by Greenacre (1993) (see section 5.2.6). I therefore wrote an R macro to recast the diagnostics from FactoMineR’s CA routine into Greenacre’s form.3 The scattergrams from CA are more properly ‘maps’ which means that 1 unit of x must be the same size as 1 unit of y, like a geographical map. Many pieces of software which produce scattergrams, like Microsoft Excel, cannot be forced to make x = y and thus specialist software must be used. R makes very good maps in most circumstances but I found the large numbers of variables in the analyses reported here were difficult to plot well. After some trial with a variety of packages I resorted to returning to the specialist CA program canoco (ter Braak 1987, 1990) and its companion plotting program canodraw (Smilauer 1990). In order to get the exact plots I wanted I stuck to using the MS-DOS version of canoplot and produced PostScript files. These were then edited manually and converted to PDF files using Ghostscript/Ghostview. The PDF files were trimmed using Heiko Oberdiek’s pdfcrop.pl perl script for inclusion in this document. Canoco does not produce the standard diagnostic statistics but does have a wide variety of options. One option, to make variables or samples passive, performs the CA without including those items and then plots their positions on the resulting maps. This can be useful when one wants to see how those items relate to the broader data set without them influencing the results. The CAs discussed below, whether run in R or canoco were ‘ordinary’ CA. The options for automatically down-weighting rare variables was not used, and if particular points were omitted from the analysis they are noted in the text. The plots were produced with symmetric scaling unless noted. 3

Available from the author.

41

5. Comparing hoards — Correspondence Analysis

5.2 5.2.1

Correspondence Analysis and the analysis of coin hoards and site assemblages Introduction

This section will examine CA with a view to developing a methodology for the analysis of coin hoards by their age profile. After a review of some aspects of CA, an example data set will be analysed and discussed in detail as an example of how the method works. Section 5.2.8 then goes on to compare the technique to Principal Components Analysis.

5.2.2

An informal description of CA

Imagine we have collected data about a group of people, say height and weight.4 We can obviously examine the variation on a scattergram without problem. If we had collected data for a third variable, say waist measurement, we have two choices. We can either plot three scattergrams for each pair of variables, or we can create a three dimensional scattergram, either for real with thin wire and balls or points suspended in a glass cube, or by using perspective to plot the three dimensional figure on a flat, two-dimensional piece of paper. If we had a three-dimensional model, we could rotate it until we found a good line of sight, one which enables us to see the patterning most clearly. Some computer packages allow one to rotate 3D diagrams in exactly this way. What happens, however, if we measure four variables?, or five?, or 100? Although it is impossible to visualise 100 dimensional space, we can manipulate it mathematically. What methods such as CA do, in effect, is to rotate this multidimensional scatter of points to find the best ‘view’, and then to project this view onto a two dimensional, flat scattergram. Complex data sets may require several such diagrams to encompass all the important variation. How do we interpret scattergrams produced by methods like CA, however? These scattergrams are more accurately called maps because they should be scaled so that 1x = 1y. When we look at a map resulting from a CA, two points plotted close together are, in some way or other, similar, and vice versa, just as they would be in a normal 2 or 3D scattergram. It is the ‘some way or other’ which needs sorting out in CA. To enable us to do this, CA allows us to plot two maps, one of which is a map of ‘samples’, in our case hoards, the other is the map of ‘species’ or ‘variables’, in our case coin issues or issues grouped by date.5 If two issues are plotted close together on the map, then they are likely to vary together. To go back to our people example, if two of the variables were inside leg measurement and total height, we would expect tall people to have large inside leg measurements, and short people small inside leg measurements. On the map of variables, these two would be plotted close together as they vary together. By looking at the distribution of the variables on each axis, we can come up with an interpretation for each axis, large versus small, early versus late, tool assemblages versus ornament assemblages etc. Having interpreted the 4

For the purposes of this explanation, I have chosen an example which does not use count data, simply because it is one which many find easy to visualise. 5 There is the option to produce ‘asymmetric’ or ‘symmetric’ maps in CA. Although their are technical problems with the former, the practical advantages of the latter are such that the majority of CAs are presented as symmetric maps.

42

5.2. CA and the analysis of coin hoards and site assemblages

variable map, we can look at the object map and see, for example, which samples are large or small, which are early or late, tools or ornaments.6 By looking at the two maps together we hope to be able to see patterning within our variables and samples. A common, and oft striven for, result is for the variables and samples to form a curve (or horseshoe) on the map where the sequence around the curve represents time with the earliest objects and assemblages at one end, and the oldest at the other. How can we tell how good a map is at illustrating the variation in our data? For each new axis a percentage figure can be calculated which can be read as “this axis can show (or ‘explain’) x % of the variation in data”. The higher the percentage for the whole map, the better the display. Unfortunately, it is difficult to use these figures to compare how good two or more CAs are because it depends in part on the number of samples (hoards) and variables (years, coin issues etc.) there are in the analysis.

5.2.3

The ‘horseshoe curve’ or ‘Guttman effect’

The results of CA and other techniques with a similar aim are often presented as a scattergram or map. The distribution of the points on these maps often form a curve similar to an arch or horseshoe. As this pattern will be encountered and discussed below, a brief outline of one reason why this occurs is offered here. In Figure 5.1 the upper graph shows the relative frequency of three objects: A, B and C. The x axis represents a dimension such as time, (i.e., the five samples were taken at different times), space (different places) or social status. Other ‘gradients’ are possible. In terms of hoards the five samples could be hoards with different closing dates or hoards from different areas. Social status as a gradient can be found, for example, in cemetery assemblages or between settlements. The composition of the five samples is given. If we were to perform CA on such a data set a graph similar to the bottom part of Figure 5.1 would result. The horizontal axis could be interpreted as representing the the distribution of A and C, whereas the vertical axis represents with, or without B. The order of the sample points is shown by the dotted horseshoe shaped line. This order, or seriation, is sometimes the aim of CA in archaeology. However, in ecology it is less desirable as the gradient is known — often a sample transect — and techniques to remove the curve have been developed, see page 177. If we were to examine a large set of hoards from the database with a variety of closing dates we would certainly encounter a time gradient which would result in a horseshoe curve. Figures 5.2–5.3 are maps derived from CA of 297 hoards. They all have 30+ well identified denarii, i.e., with an accuracy code of 1, 5 or 6. As can be seen a curve, similar to what was expected, has resulted and this is generally the result of a time gradient.

5.2.4

Dividing the data — selecting hoards for analysis

The next important question to answer is what to analyse. Some 458 hoards were available for analysis at the time of writing. They consisted of 109,445 6 It is common to plot both variable and samples on the same map but I prefer to keep them separate, partly because crowded maps become difficult to interpret, but also for the technical reason that the distance between a variable point and an object point on symmetric maps is undefined, and sometimes leads to erroneous interpretations.

43

frequency

5. Comparing hoards — Correspondence Analysis

A

B

C

time, space, social status?

1 All objects type A

2 Objects of types A and B

3 Mainly objects type B

4 Objects of types B and C

5 All objects type C

with B 3

2

A

C 4

1

5 NO B

Figure 5.1: Diagram illustrating the usual cause of horseshoe curves in CA and PCA.

coins. These hoards are of varying size and data quality, as is the information about their contents. The first choice is to set a minimum size of hoard for analysis. Hoards as small as five coins have been used in the past, but they are too small for both numismatic and statistical reasons. In a hoard of five coins there is a high probability that the closing date of the hoard is a poor reflection of the date of collection of the hoard. The statistical problem which arises is that each coin in a small hoard represents a significant proportion of the hoard, and these small hoards will often, therefore, dominate any statistical analysis. The limit used here was 30 coins. This was chosen not on any theoretical basis, but on experience, and trial and error — hoards of this size did not usually dominate the analyses. This left 300 hoards. In this chapter only well identified denarii, i.e., those with accuracy codes 1, 5 or 6, were analysed and the total included in each analysis is given as the ‘good total’ (see section 2.3.2 and Table 2.1). It would be possible to analyse all the hoards simultaneously. Figures 5.2–5.3 are maps from CA of 297 hoards.7 All variables (coins grouped by ‘date_from’, 7

When this analysis was performed 300 hoards were available but in the first run three

44

+4.0

5.2. CA and the analysis of coin hoards and site assemblages

pra kl1

pis

isa

+2.5

-1.5

-2.5

Figure 5.2: Sample score map from CA of 297 coin hoards. Points represent hoards. First (horizontal) and second axes of inertia.

referred to below simply as ‘years’) contribute to the analysis. The first axis has an eigenvalue of 0.468 and ‘accounts for’ 19.6% of the variance in the data; the second axis has an eigenvalue of 0.292, 12.2% of the variance, giving a total of 31.8%. Given the very large size of the data set this is acceptable. The two maps show a time gradient from the top-left of the curve of points, round to the top-right. The three hoards which stand out on the extreme end of the curve to the right, Vico Pisano, Pravoslav and Köln (I) (pis, pra & kl1) close in 8 or 2 bc, some of the latest hoards in the data set. It is immediately apparent that there are difficulties in interpreting the results. It is a reasonable seriation of the hoards, but this is a confirmation of what we already know. Some oddities can be seen, e.g., the hoard from Işalniţa (isa) stands out in the middle hoards with exceptional numbers of legionary denarii of Mark Antony dominated the results. The analysis was therefore re-run excluding those hoards.

45

+4.0

5. Comparing hoards — Correspondence Analysis

17

2

12 8 11 16 20 15 18 13 19

25 32 33 21

27

28

29 36 31 40 41

142

207 155 157 179 208 199 150 169 194 206 189 141 211 139 135 131 209 128 127

69 61

123 122

58

73

125

67

124

76

119 120 72 79 118 116 103 80 81 114 115 100 117 92 85 83 109 113 108 101 89 88 96 97 90 91

+2.5

68 71 70

-1.5

-2.5

44 46 43 49 47 45 48 39 51 60 56 54 66 59 57 52 53 65

Figure 5.3: Variable score map from CA of 297 coin hoards. Points represent years of coin issue (date_from). First (horizontal) and second axes of inertia.

of the map but it is impossible to work out why. The time gradient, which we know exists, is dominant, but it is clear that other processes are also at work. The coinage of 90s–80s bc forms a large proportion of the coinage pool. This results in the relatively large dispersion of points in the bottom right quadrant. Mixing hoards of different periods makes interpretation difficult in this sort of analysis (contra Creighton 1992a, p. 33). Several factors, most notably time and space, are mixed together and if we want to examine more subtle patterns in the data we have to subdivide the data into smaller groups, either into time periods and/or by region.

5.2.5

CA — a worked example and further problems

The purpose of the following set of analyses is to work through one example in quite some detail to illustrate the advantages and problems inherent in under46

5.2. CA and the analysis of coin hoards and site assemblages

code

name

cr1 cro dra el2 erd fdc gul jae loc mor p06 p07 sen spn sur ti2 vas vll

Carbonara Crotone Dračevica El Centenillo Érd Fuente de Cantos Gulgancy Jaén Locusteni Morrovalle Padova Padova Sendinho da Senhora Spoiano Surbo Tîrnava Văşad Villette

rrch 362 383 379 385 373 — 377 386 367 380 364 391 388 — 381 — — 393

country

closing date

Italy Italy Fmr Yugoslavia Spain Hungary Spain Bulgaria Spain Romania Italy Italy Italy Portugal Italy Italy Romania Romania France

48 46 46 46 46 46 46 46 48 46 48 45 46 46 46 46 46 45

‘good total’ 383 86 109 57 51 387 459 65 88 125 54 655 76 264 138 148 53 340

Table 5.1: Hoards in test data set used in CA and PCA discussed in sections 5.2.5 and 5.2.8.

taking CA on coin hoard data. The results of the CAs presented in section 5.4 are discussed in as concise a manner as possible. A test data set of eighteen coin hoards closing in 49–45 bc, one of the periods used by Crawford in RRCH (Crawford 1969c), was analysed using CA. Details of these hoards are contained in Table 5.1.8 The eigenvalues for the following three analyses are shown in Table 5.2. In all the following analyses ordinary CA was performed with symmetric scaling of maps. Years (variables) 211–158 were made passive (see page 41) due to their rarity in this data set. In this, and the following analyses, the term ‘rare’ should be taken to mean rare or uncommon in the current data set. Likewise, the term ‘abundant’ means common in the current data set. In the case of the latter term it should not be confused with a statistical measure of abundance. Analysis one — all the data The eighteen hoards contained 3,538 denarii. Figures 5.4–5.5 present the variable and sample score maps from this analysis. There are three points to note. Firstly, the hoard from Érd (erd) appears to be very different from the other 17 hoards in this analysis standing clear at the top of the second axis in Figure 5.5. Years 157, 76 and 74 are in a similar position in the variable map (Fig. 5.4) suggesting that these years are particularly associated with Érd. Secondly, there appears to be a classic horseshoe curve centered around the origin of the maps. This is partly a result of the varying closing dates of the hoards. Two hoards closing in 45 bc appear on the right of the plot, and the three hoards closing in 48 bc on the left (marked by a triangle and square respectively). The variable map shows 8

This data set, originally analysed for a public lecture, is only part of the data now available for the period 49–45 bc. It is used an an example as it nicely demonstrates the problems of the time gradient and of ‘odd’ hoards.

47

5. Comparing hoards — Correspondence Analysis

Axis

CA Analysis one CA Analysis two CA Analysis three CA Analysis four PC Analysis one PC Analysis two

1

2

3

4

Total

0.569 0.322 0.323 0.379 0.242 0.284

0.224 0.261 0.137 0.237 0.174 0.151

0.184 0.134 0.091 0.175 0.121 0.117

0.069 0.088 0.087 0.156 0.096 0.086

1.482 1.177 0.935 1.656 — —

Axis

CA Analysis one CA Analysis two CA Analysis three CA Analysis four PC Analysis one PC Analysis two

1

2

3

4

38.4 27.4 34.6 22.9 24.2 28.4

53.5 49.5 49.3 37.2 41.5 43.5

65.9 60.9 59.0 47.8 53.7 55.2

70.5 68.4 68.3 57.2 63.3 63.8

Table 5.2: Eigenvalues (top) and cumulative variance explained (bottom) from the analyses of the data set listed in Table 5.1, see sections 5.2.5 and 5.2.8.

a gradient around the curve from right to left — the order is not perfect but still marked. However, it appears that a second gradient is also present. Hoards Gulgancy (gul) and Văşad (vas) at the left extreme of the curve close in 46 bc along with the majority of the other hoards including Sendinho da Senhora (sen) near the right hand end of the curve. This raises the question: how much of the pattern is the result of the known time gradient and how much the result of another, unknown gradient? Analysis two — hoards closing in 46 bc In order to overcome the shortcomings of the first analysis, the second analysis contained only the thirteen hoards which closed in 46 bc, some 2,018 denarii. Figures 5.6–5.6 present the maps from this analysis. Again, the exceptional hoard from Érd dominates the plot standing clear on the second axis (Fig. 5.7). The curve is no longer clear and the majority of the hoards cluster along the first axis. Some patterning is evident — the four hoards from Italy cluster tightly as a group and all the Romanian and Bulgarian hoards are to the left of the map. It is, however, difficult to interpret these maps beyond the dominance of one hoard in the data set. A common problem in CA is the large influence exerted by odd or unique items. Érd is a good example of this. Looking at the data we can see that Érd has 26 coins with a date of 76 or 74 bc out of a total of 51 (50.9%) whereas other hoards, such as Sendinho da Senhora with 1 out of 76 (1.3%) or Spoiano (spn) with 3 out of 264 (1.1%), have much lower quantities. A common procedure when faced with this problem is to omit the sample or variable causing the problem — in this case Érd — and re-run the analysis. This is not ‘fudging’ as long as the

48

+4.0

+4.0

5.2. CA and the analysis of coin hoards and site assemblages

erd

closing 48 BC closing 46 BC closing 45 BC

74 76

157

79 81 83 89

p06 54

67 57 110

71

100 109 119

fdc

55

ti2

56 49

80 106 131 151

-1.5

65

45

+2.0

-1.5

cr1 cro spn sur mor

el2 dra

sen vll

+2.0

loc

48

jae

46

p07

vas

126

84 92 120 117 136 145

gul

141 148 206 189

-3.0

-3.0

194

Figure 5.4: Variable map from the first CA of the test date set. Points are years of issue, first and second axes of inertia.

Figure 5.5: Object map from the first CA of the test date set. Points are hoards, first and second axes of inertia.

exception is noted, described, and if possible, explained and is sometimes called ‘peeling’ (Cool & Baxter 1999). Indeed, with some data sets the analysis may proceed by successive ‘peels’ being used to slowly carve it up into meaningful groups. Analysis three — omitting Érd In this analysis, the remaining 12 hoards, containing 1967 denarii, were analysed. Figure 5.8 again shows a curve, but this time it is not a result of variation in the closing date of the hoards. However, there is still a time trend from the top right quadrant to top left with the majority of the years from 92–57 in the bottom

49

149

148

96 103 78 101 113 117 86 155 116 109 153 85 142 71 136 114 89 92 120 77 104 84 146 111 64 125 112127 106 80 209 143 211

124

102

79 81 82 70 67 12958 60 57 54 59 56 47 55 87 152 48 208 61 139

83

157

74

76

49

46

+1.5

Figure 5.6: Variable map from the second CA of the test data set. Points are years of issue.

-2.5

141 169 134

128 100 121

132

+4.5 -1.0

50 gul

vas

jae

ti2

sur cro spn mor

fdc

dra

el2

erd

sen

+1.5

Figure 5.7: Object map from the second CA of the test data set. Points are hoards.

-2.5

+4.5 -1.0

Italy Spain Portugal Romania Bulgaria Hungary former Yugoslavia

5. Comparing hoards — Correspondence Analysis

+2.0

5.2. CA and the analysis of coin hoards and site assemblages

46

135

142 141 179 134

147

49

169

47

151

148 149

144

127

154

136

115

125 126

132

114

128

48

123

121

131

75

116 120

96

150 105

117

-2.5

81

122 84

145

140

101

110

209

103 92

80

130

100

76 211

78 153

109

118

86

88

89

59 60

77 83

113

+2.0

61

63

87

90

102 71

62

82 79

111 85

55

56 54

108

129 68

58 66

139 67

57 64

70

-2.0

157

Figure 5.8: Variable score map from CA of hoards closing in 46 bc omitting Érd, see Table 5.1. Data points are years of issue. First (horizontal) and second axes of inertia.

right quadrant. Only three years in this range have a positive value on the second axis (84, 81 & 75). There are some outliers and oddities. The last active variable, 157, occurs at the bottom of the 2nd axis, 135 seems to float in the middle of the curve. Both these are quite rare years with only seven coins between them. We can interpret the first axis as representing the relative numbers of new coins compared to old coins. The second axis can be interpreted as representing the relative numbers of coins of 92–57 compared to coins of other years. Comparing this map to Figure 5.9. it can be seen that the hoards lie in a similar curve. To show how the sample and variable points relate we can examine a few cases. Taking the three hoards at the extremes of the curve (sen, fdc & gul) we can identify the three years plotted nearest to each on a joint plot and

51

+2.0

5. Comparing hoards — Correspondence Analysis

Italy Spain Portugal Romania Bulgaria former Yugoslavia

sen el2

dra

jae gul

vas sur

-2.5

+2.0 spn mor cro ti2

-2.0

fdc

Figure 5.9: Sample map from CA of hoards closing in 46 bc omitting Érd, see Table 5.1. Data points are hoards. First (horizontal) and second axes of inertia.

then calculate the percentage of the hoard totals belonging to those years.9 This gives: Years 146, 137 & 128 89, 86 & 77 49, 47 & 46

gul

fdc

sen

6.5% 1.9% 0.4%

1.0% 8.0% 2.5%

0.0% 1.3% 43.4%

The gradients along the horseshoe curve can be clearly seen. We can conclude that there is a trend from hoards with relatively less new coin but more old, to 9

Theoretically, this process is incorrect as the plots have been produced using symmetric scaling, and thus the distances between a sample point and a variable point on a joint plot are not defined. The correct procedure would be to select variable (years) at the same extremes of the axes as the sample (hoard) points. The rarity of the extreme variable, however, does not lead to such a nice illustration of pattern. The incorrect interpretation of these inter-point distances on symmetric maps can lead to erroneous conclusions. I would like thank Mike Baxter for pointing out this important point.

52

5.2. CA and the analysis of coin hoards and site assemblages 100 90 80

cumulative percentage

70 60 erd 50 gul

vas

jae

dra el2

ti2

sen

40 30 20 10 0 157

147

137

127

117

107

97

87

77

67

57

47

year BC

Figure 5.10: Cumulative percentage curves for 13 hoards closing 46 bc. Solid lines: hoards from Italy; dashed lines: hoards from the Iberian peninsula and the former Yugoslavia; dotted lines: hoards from Hungary, Romania and Bulgaria.

hoards with relatively more middle period coinage when compared to the oldest and newest coins, to hoards with mainly new coin. We can examine the find spots of hoards represented in Figure 5.9 by using different symbols. The four Italian hoards are closely clustered on the first axis although separated out on the second axis. I showed previously that the huge quantities of coinage minted during the Social War (91–89 bc) and shortly after results in a large proportion of the variation in hoard structure for these years being a result of random variation rather than archaeological processes (Lockyear 1993, also see Chapter 7). To some extent the second axis reflects this. The Iberian hoards occur along the length of the curve showing a much greater variation than the Italian hoards. The Romanian and Bulgarian hoards have little of the most recent coin although Tîrnava (ti2) seems not too dissimilar from Fuente de Cantos (fdc) and the Italian hoards. We can compare the results of this analysis with a cumulative percentage curve graph (Fig. 5.10). As can be seen, the Italian hoards form a compact group in the middle of the plot as would be expected from the CA. The Iberian hoards appear both sides of the Italian pattern with the Yugoslavian hoard, Dračevica (dra), also falling below the Italian line. The Romanian and Bulgarian hoards show clearly above the Italian profile with Gulgancy (gul) looking very unlike other hoards. Tîrnava (ti2) has large quantities of coins from the 90s–80s. The remaining Romanian hoard, Văşad (vas) is above Tîrnava having a larger proportion of earlier coins. We can conclude that the gradient revealed by the first two principal axes, and the cumulative percentage curves, is one in the age structure of hoards. This age structure is not a simple reflection of the find spots of hoards, although some spatial factors clearly are present. Italian hoards are all similar, Iberian hoards are more variable and occur above and below the Italian pattern, Romanian 53

Rome Spain Africa moving with Pompey moving with Caesar eastern mint uncertain before issue 425

+4.0

5. Comparing hoards — Correspondence Analysis

469

459

461 468

443

453 427

464 434

431 426

458

-2.5

465

+3.0

442 430 449 448 432

425 433 428

445

-1.5

429

450 463 455, 467 444 452

454

440

Figure 5.11: Variable map from CA of 12 hoards closing in 46 bc. First and second axes of inertia, data points are RRC issue numbers.

and Bulgarian hoards tend to have more old coin, and coin from the 90–80s bc. At a more detailed level, there seems to be no relation between the detailed geographical position of, for example, the Italian hoards and their dispersion along the second axis. The Iberian peninsula hoards are likewise unrelated to their detailed geographical position. El Centenillo and Jaén (el2 & jae) are the closest geographically but not on the CA maps. Following Creighton (1992a) I shall term hoards with large quantities of older coin, relative to other hoards of the same date, ‘archaic’ in structure, those with large quantities of new coin, ‘modern.’ The first two principal axes discussed above represented 49.3% of the variance in the data. We should not, however, ignore the the third principal axis which represents another 10%. I shall not go into this axis here but in some of the analyses discussed below the third axis of inertia has pulled out some more interesting aspects of the data. Analysis four — using issues as variables As noted above, the use of Crawford’s dates is controversial not just in terms of whether the date of any one issue is correct, but also which issues were struck 54

+4.0

5.2. CA and the analysis of coin hoards and site assemblages

Italy Spain Portugal Romania Bulgaria former Yugoslavia

el2

sen

jae

gul vas

-2.5

+3.0

fdc cro sur spn

ti2

dra

-1.5

mor

Figure 5.12: Sample map from CA of 12 hoards closing in 46 bc. First and second axes of inertia, data points are hoards.

in the same year. Therefore, even if we try and argue that the ‘years’ are just convenient groupings, many will disagree. The main problem with analysing the hoards by issues — by which I mean individual RRC numbers, not their subdivisions10 — is that the numbers of variables and the sparseness of the resultant table. In the above analyses there were 1,922 denarii spread over 97 active variables. Analysing the same data set by issues there are 212 variables. This increases the chances the analysis may be distorted by rare variables or odd samples, such as the effect of Érd above. On the other hand, removing the possible errors in the grouping of issues may either reveal other interesting patterns in the data, or cast into doubt patterns revealed by the analysis by years. To test this, the twelve hoards used in the preceding analysis were re-analysed using issues as variables. Figures 5.11–5.12 are the maps from this analysis representing 37.2% of the variance in the data. As can be seen the basic pattern of sites has been maintained (cf. Fig. 5.9). giving us confidence in the results of 10

I know of only a single later issue, RRC 393/1a and b where it has been argued that these do not go together. Earlier issues are more complicated but do not concern us here

55

1000

1500

5. Comparing hoards — Correspondence Analysis

●

500

●

●

●

0

● ●

●

●

−500

● ● ●

−1000

●

−500

0

500

1000

Figure 5.13: Procrustes Analysis comparing CA of the test data set by date v. issues.

the analysis. A formal method of comparison will be used below. The principal difference is that the Italian hoards are now even more clustered, and Dračevica (dra) is plotted close to them rather than half-way between the Italian hoards and the Iberian peninsula hoards. The variable map, now representing the individual issues, is extremely interesting. I have used symbols to represent where the issues from 56–46 bc were struck according to Crawford (1974). RRC 463–469 were all struck in or from 46 bc and had been grouped into a single variable in the previous analyses but RRC 463–7 were struck in Rome and RRC 468–9 were struck in Spain. In this analysis the correlation between the Spanish issues and the Iberian hoards is clear and checking the data we find that there is only one out of 35 coins of these two Spanish issues in the four Italian hoards. Of the three African issues, RRC 459 and 461 are very rare with only three examples all of which in two of the Iberian peninsula hoards as is clearly indicated in the two maps. RRC 458, however, is more common with 14 examples but only three of these occur in Spanish hoard contrasting with 11 in Italian hoards. I will discuss this pattern further below (see section 5.4.15, page 116). Issue 443 which was struck by the mint ‘moving with Caesar’ is also a relatively more common coin with 23 examples in the data set but of these only 6 occur in the four Italian hoards compared to 13 in the El Centenillo and Sendinho da Senhora hoards (el2 & sen). We have, therefore, teased out yet one more cause of variation in the data, i.e., patterns of minting due to the events in the Civil Wars. In this case the comparison between the two sample maps, Figures 5.9 and 5.12 is not too difficult to do visually. This may not always be the case as it is possible to reverse the direction of the axes without affecting the interpretation of the results, or even to rotate them. There exists, however, a neat way of

56

5.2. CA and the analysis of coin hoards and site assemblages

comparing two maps called a Procrustes Analysis (Oksanen 2007). In this two distributions are compared and a best fit found between them that minimises the change in the relative positions of the individual points. The results are then plotted as a map with arrows showing the movement of individual points. The shorter the arrows, the better the fit. Figure 5.13 show the Procrustes Analysis between the CA of the 12 hoards using years v. issues.11

5.2.6

Diagnostic statistics

So far we have concentrated on the intuitive analysis and interpretation of the maps created by CA. CA, however, also produces detailed numeric output which we can call ‘diagnostic statistics’ as they can help us understand the maps in more detail, and help us avoid erroneous interpretations. Formal discussion This section is intended for readers who have been comfortable with the description of CA so far, but are new to the technique. Readers who wish to avoid the technical aspects of CA should skip to the next section, those familiar with the method, should skip to section 5.2.7. Tables 5.3–5.4 present the numeric output from the third analysis of the test data set. As can be seen, when we are looking at the first two axes we get, for every sample (hoard) and variable (year of issue), nine statistics. The abbreviations used in these tables are those used by Greenacre (1993, see chapter 11 especially) although the description of their meaning closely follows that by Underhill & Peisach (1985). The easiest column to understand is mas for mass which is simply size of the sample or variable expressed as a per mill (i.e., per 1000) of the total. Thus the hoard from Fuente de Cantos contained 387 well identified denarii, which forms 201% (i.e., 20.1%) of the total 1,922 denarii dating to 157–46 bc (cf. Table 5.3). For individual years the mass is often very low due to the large numbers of variables in each analysis. The columns k=1 and k=2 are the scores which are plotted in on the maps, usually multiplied by 1000 to keep them as whole numbers.12 For the remainder of the statistics we have to come to grips with the concept of the inertia of a table, and to explain this we need to go back to the concept of the χ2 statistic. The χ2 “statistic measures the discrepancy between the observed frequencies in a contingency table and the expected frequencies calculated under a hypothesis of homogeneity of the row profiles (or of the column profiles)” (Greenacre 1993, p. 31). In this case we are not usually interested in using this statistic to perform a significance test, but as a measure of variation within our table. To do this we can divide χ2 by the total number of items in the table to give the total inertia of that table. In the case of the twelve hoards in Analysis 3 χ2 = 1797.497 and so the total inertia (I) is given by: I = 1797.497/1922 = 0.935 (Table 5.2). Of that total inertia, 34.6% is ‘accounted for’ by the first axis and 14.7% by the second. This means that the maps of the first two axes can represent 11

The analyses were conducted using the vegan package within R. Canoco appears to scale CA slightly differently to other packages and as a result the k= values differ but the maps are identical and so the difference is unimportant. 12

57

5. Comparing hoards — Correspondence Analysis

hoard

qlt

mas

inr

k=1

cor

ctr

k=2

cor

ctr

cro dra el2 fdc gul jae mor sen spn sur ti2 vas

190 518 551 569 966 51 115 808 277 113 255 113

45 57 30 201 222 34 63 40 135 71 76 27

52 53 71 85 225 66 71 111 68 59 56 83

340 586 790 109 −926 −120 292 1139 361 271 −20 −546

107 393 278 30 904 8 82 493 275 95 1 101

16 60 57 7 587 2 17 159 54 16 0 24

−299 331 782 −462 243 279 −185 911 −31 118 −419 182

83 126 273 539 62 43 33 315 2 18 255 11

29 45 132 313 95 19 16 239 1 7 98 6

Table 5.3: Diagnostic statistics for ‘samples’ (hoards) from the third CA of the test data set.

just under half the variance in the data, which for a 12 by 97 row table is pretty good going. The next stage is to examine how much individual samples and variables contribute to the overall variation and the individual axes. The columns labelled inr for inertia represent the proportion of the total inertia, again expressed as a per mill, due to that sample or variable. Thus Table 5.3 shows that 85% of the inertia is due to Fuente de Cantos and Table 5.4 shows 97% is due to coins from 46 bc. The columns labelled qlt for quality shows how well the sample or variable is represented on the two dimensional map. Table 5.3 shows that the map of the first two axes represents Gulgancy extremely well with 966% of its inertia represented, whereas Jaén is very poorly represented with only 51% of its inertia accounted for. In Table 5.4 shows that 933% of the inertia for 49 bc is accounted for by the first two axes, but only 15% of the inertia for 155 bc is accounted for. Thus, when we look at Figure 5.9 we can be happy that Gulgancy and Sendinho da Senhora (gul & sen) are well represented, but Jaén (jae) is very poorly represented. The columns labelled cor give the figures for the relative contributions which “indicate the proportion [here given as a per mill] of the point’s contribution to the inertia that is accounted for by each axis” (Underhill & Peisach 1985, p.53). Thus, in Table 5.3 we can see that only 30% of the inertia due to Fuente de Cantos is explained by the first axis, but 539% is explained by the second axis. This contrasts with Gulgancy where 904% of the inertia due to that hoard is explained by the first axis and only 62% is explained by the second. Note that the quality is the sum of the two relative contributions. Lastly, the columns labelled ctr give the absolute contributions and “indicate the proportion of the inertia explained by the axis which is due to each sample and [variable]” (Underhill & Peisach 1985, p. 53). Thus 587% of the variation accounted for by the first axis is in turn accounted for by Guljancy but only 7% by Fuente de Cantos. The absolute contributions will sum to 1000%. From these statistics we can identify which variables and objects are well represented on the CA maps, and contribute significantly to the results of the analysis. 58

5.2. CA and the analysis of coin hoards and site assemblages

year of issue

qlt

mas

inr

k=1

cor

ctr

k=2

cor

ctr

157 155 154 153 152 151 150 149 148 147 146 145 144 .. .

208 15 243 241 96 126 267 899 766 756 479 482 19 .. .

1 2 3 3 2 2 4 6 10 7 6 3 1 .. .

6 12 12 3 5 9 3 13 23 10 12 4 13 .. .

395 −327 −853 −479 541 −541 −471 −1232 −1203 −756 −890 −836 −242 .. .

27 15 205 213 93 74 259 805 673 419 443 481 5 .. .

1 1 7 2 1 2 2 29 44 12 15 6 0 .. .

−1026 −2 370 −173 −89 453 81 421 447 678 254 44 404 .. .

181 0 39 28 3 52 8 94 93 337 36 1 14 .. .

8 0 3 1 0 3 0 8 14 23 3 0 1 .. .

60 59 58 57 56 55 54 49 48 47 46

133 133 471 198 408 442 572 933 359 739 889

2 2 12 2 14 15 8 33 21 13 41

5 5 13 6 13 17 7 45 31 16 97

394 394 621 445 594 687 659 887 629 695 1012

54 54 382 77 408 442 572 615 287 414 466

1 1 15 1 15 22 11 80 26 19 130

−476 −476 −301 −559 15 23 22 638 316 615 964

79 79 90 121 0 0 1 318 72 325 423

3 3 8 5 0 0 0 97 15 36 278

Table 5.4: Diagnostic statistics for variables (years of issue) from the third CA of the test data set.

Diagnostic statistics ‘for dummies’ 1. When first looking at the maps from a CA check the quality (qlt) for each object/variable. If it is low, say less than 100, it is poorly represented on the map. 2. To see which objects/variables have an important contribution to variation in the data set look at the inertia (inr). If it is higher than most of the others it will probably be important. Check, however, than it also is a reasonable size by looking at the mass (mas). 3. To see which objects/variables can be used to interpret one axis on the map, check the relative contributions for that axis. Thus in Table 5.3 we can see that both spn and ti2 have a good quality for the map (277 and 255 respectively), but that spn has a relative contribution of 275 for the first axis, but only 2 for the second compared to ti2 which has a relative contribution of 1 to the first axis but 255 to the second.

59

5. Comparing hoards — Correspondence Analysis

5.2.7

Testing for significance

It is possible to run a CA on a set of random numbers and produce a map. It would make no sense to interpret such a map. We can check for both statistical and archaeological significance in a variety of ways. Archaeological significance is a subjective test but we should not ignore it. If the hoards on our CA maps, for example, form groups on the basis of geography, date, site type or some other outside variable we can argue that the the results are likely to be meaningful. We should be careful, however, not to translate our expectations on to the results. Greenacre (1993, p. 173) provides a formal method for testing the significance of a principal axis. As the inertia of a table is related to the χ2 statistic we can calculate a value for χ2 for each axis of inertia by multiplying the inertia by the number of items in the analysis, in our case coins. More formally: χ2 = λ × n In the case of the third axis of inertia in the third analysis discussed above this gives us: χ2 = λ × n = 0.091 × 1922 = 174.9 This value is then compared to a table of critical values (Pearson & Hartley 1976, Table 51) to assess if the value of χ2 is significant for the probability level we desire and the size of our table. In this case out table is 12 × 106 and so we look up the value for 11 by 105 (columns−1, rows−1), i.e., 11 by 105 degrees of freedom.13 In this case, the value for the third axis is not significant at the 0.05 level, although it is close to the critical figure for the 0.1 level.14 The problem with this approach is that the size of the sample is important and therefore analyses containing huge hoards such as Mesagne (mes) will often have many statistically significant axes. We need to make a judgment between practical significance and statistical significance (Wang 1993). A further method would be to use some form of computationally heavy technique such as resampling or bootstrapping where the analysis is run multiple times to assess how much changes in the data would influence the results (Greenacre 1993, Chapter 20). I have not followed this route in the analyses presented here, but have frequently omitted single objects or variables to see how much they influence the results, and these have been commented on below.

5.2.8

Principal Components Analysis and CA — an empirical comparison (technical discussion)

Correspondence Analysis is only one multivariate technique which could be used for the analysis of this type of data. Before proceeding with the detailed analyses 13

Unfortunately, this table only gives values for a maximum table size of 10 by 200. However, reasonable estimates can be made by graphing the values given and extrapolating. 14 The 0.05 (i.e., 1 in 20, or 5%) level is often used in significance tests. There is nothing magic about this level. Strict adherence to the idea that somehow less than one chance in twenty is significant is common, despite most text books noting that this level is little more than convention. See Thomas (1978), Castleford (1991). It has led to the rejection of otherwise important results, e.g., Williams (1993), cf. Kvamme (1990).

60

+1.5

5.2. CA and the analysis of coin hoards and site assemblages

Italy Spain Portugal Romania Bulgaria Hungary former Yugoslavia

erd

el2

jae

sen

vas

-0.9

+0.9 ti2 fdc

cro sur

-0.5

gul

dra

spn mor

Figure 5.14: Object loading map from PCA of 13 hoards closing in 46 bc, see Table 5.1. Data points are coin hoards. First (horizontal) and second principal components.

I wish to consider one alternative method — Principal Components Analysis (PCA; Shennan 1997, pp. 269–303, Baxter 1994, pp. 48–99). In a review of the package Mv-Arch (Wright 1989) posted to the Archaeological Information Exchange 15 I noted that it was difficult to remove units or variables from an analysis. Wright (pers. comm.) replied: In the case of Correspondence Analysis, I have to say that I find it worrying that so much effort is spent on circumventing the results given by the straightforward version. . . . . . The essence of the method is dual scaling. . . [it] highlights two types of structure in a matrix: (a) the large occurrence of an (elsewhere generally rare) attribute at a site that has very few attributes, and (b) the small occurrence of an (elsewhere generally common) attribute at a site that is rich in attributes. . . It would be better to use Principal Components Analysis of the square roots of percentage frequencies, extracting the eigenvectors 15

The AIE was an electronic mailing and discussion list run from Southampton, initially by S. P. Q. Rahtz, then by myself. It has now been replaced by the list Arch-L.

61

5. Comparing hoards — Correspondence Analysis

from a covariance matrix of unstandardised variables. In other words, I wonder why you choose CA when what you want to see is the structure of the main body of the data. . .

This raises a number of points: 1. The nature of the joint plots in CA is one of its attractions. PCA biplots (Baxter 1992), when there are many variables, are visually more difficult to interpret as the arrows make for a confusing diagram. 2. The suggested method introduces the problem of compositional data16 (Aitchison 1986) into the analysis when the data set is not inherently compositional. 3. Both the initial analyses (showing ‘oddities’) and those showing variation in the main body of the data are of interest. A comparison between Wright’smethod and CA was undertaken using the same data as section 5.2.5. As before, canoco was used. The eigenvalues etc. for analyses are given in Table 5.2. Initially the 13 hoards closing in 46 bc were analysed in order to see how the method suggested by Wright would work on a data set with an outlier. Figure 5.14 shows the map of the object loadings. In this map the general pattern is similar to Figure 5.7 in that the hoard from Érd (erd) dominates the second axis, although less strongly. More interpretable patterning of the remaining points, similar to Figure 5.9, also exists. This is because the square roots of the values input into the analysis have been used. Such an option is also available in CA and would have had a similar effect (cf. Lockyear 2000b). However, the positions of the hoards in Figure 5.14 are still compressed on the second axis masking variation in the hoards. Re-running the analysis omitting Érd results in Figures 5.15–5.16. The object loading map (Fig. 5.15) is now very similar to Figure 5.9 if somewhat more spread out. However, the variable plot (Fig. 5.16), presented in the standard style, is much less interpretable than Figure 5.8. In PCA biplots the length of the arrow represents the approximate variances of the variables; the cosines of the angle between arrows their approximate correlation. Thus, two variables represented by arrows with an acute angle are highly correlated (Baxter 1992). For many variables the result is very confused. The order for hoards on the first axis is identical to CA apart from the Italian hoards although these are still tightly grouped on this axis. The order for years is also very similar. The order on the second axis is quite different although hoards above and below zero remain similar. The broad patterning of years is equally similar. In conclusion, the results of the PCA are not substantially different from those gained by CA. This is not surprising as CA can be viewed as a PCA of transformed data (Baxter 1991). Indeed, it would be worrying if the results were very different, suggesting unstructured data. The down-weighting of unusually abundant variables by using square roots of percentage frequencies could be achieved in CA. However, we wish to know about odd hoards and/or years as 16

Compositional data are data with a constrained total, e.g., percentages or proportions.

62

5.2. CA and the analysis of coin hoards and site assemblages

+0.7

vas

el2

jae

sen

dra

-1.0

+1.0

sur cro spn fdc

Italy Spain Portugal Romania Bulgaria former Yugoslavia

mor

-0.6

ti2

+0.7

Figure 5.15: Object loading map from PCA of hoards closing in 46 bc omitting Érd, see Table 5.1. First (horizontal) and second principal components. Data points are coin hoards.

123 76

151 135 134

46

115 75 142 154 127

100

78

147

47

114

-1.0

104 117 121 85

82

141 77

122

118 153 89 71 136

49

62

80 133

48 79

61 103

96

128

132

91

88

81 54

56

69 58

86 90

139 92 59 70 60 57 102 140 68 137 130 112 106 74 109 83 64 87

67 63 55 108 66

-0.6

113

+1.0

110

Figure 5.16: Variable loading map from PCA of hoards closing in 46 bc omitting Érd, see Table 5.1. First (horizontal) and second principal components. Each arrow represents a variable vector — see text for explanation.

63

5. Comparing hoards — Correspondence Analysis

this is the prime method by which ‘odd’ hoards can be detected. The iterative use of CA or PCA is preferable as each analysis illustrates different aspects of the data. The CA species maps are easier to interpret when there are many variables. Working with untransformed data seems generally preferable in this case to transformation to square root percentage frequencies. Although transformation of data is essential in many multivariate analyses, especially those on chemical compositions, it is not necessary here. I can see no inherent advantage in using PCA over CA in this case especially as CA is a technique specifically designed to analyse contingency tables.

5.3

A usable methodology

We have now defined a usable methodology for the examination of the data available. This is: • Select a sub-set of hoards which have as limited a range of closing dates as possible, one year if enough hoards are available. • Analyse the data using CA. For each group the analysis will be run at least twice, once using dates as variables, once using RRC issues. When necessary, re-run the analyses omitting samples or species. • Compare the analyses using Procrustes Analysis. • Plot the hoards as cumulative percentage curves. Use the colour and ‘plot series’ features to explore possible patterns revealed by the CAs on screen. • Describe the structure of hoards from the resultant maps in conjunction with the original data, diagnostic statistics and/or cumulative percentage curve graphs. In the following section I will not present the results from every single analysis particularly where the results from using years or issues as the variables are essentially identical, or where an analysis is so dominated by a single hoard or variable that it was subsequently omitted. The final step, of course, will be to attempt to interpret the patterns revealed in numismatic and archaeological terms. A summary of the results and conclusions is given in section 5.5.

5.4 5.4.1

The Analyses Introduction

The 300 hoards available for analysis were divided into 22 groups. In some cases groups overlapped for comparison. In the following sections the results will be described and discussed, but more concisely than the example above. For each analysis, appropriate tables and figures are given. The tables give the number of denarii in each hoard used in the analysis. Only well identified denarii, i.e., those with an accuracy code of 1, 5 or 6 (see Table 2.1), were used and as a shorthand these are called ‘good denarii’. The closing date was determined using any well identified coin with those accuracy codes. Hoard names in inverted commas, 64

5.4.2. Hoards closing 147–118 bc

Eigenvalues section

period

analysis

5.4.2 5.4.3 5.4.4 5.4.5 5.4.6 5.4.7

147–118 bc 118–108 bc 105–97 bc 118–97 bc 92–87 bc 87–81 bc

5.4.8 5.4.9 5.4.10 5.4.11 5.4.12 5.4.13 5.4.14 5.4.15 5.4.16 5.4.17 5.4.18 5.4.19 5.4.20 5.4.21 5.4.22 5.4.23 5.4.24

80–79 bc 78–75 bc 74 bc 73–69 bc 63–56 bc 56–54 bc 51–47 bc 46 bc 45–43 bc 42 bc 41–40 bc 40–36 bc 32–29 bc 29–28 bc 19–15 bc 15–11 bc 8–2 bc

1 1† 1† 2 1 1 2 1 1 1 1 1 1 1 1 1 1 1 5 1 3 1 1 1

Cum. perc. variance expl.

1

2

3

4

0.234 0.250 0.132 0.294 0.272 0.208 0.326 0.382 0.263 0.124 0.107 0.208 0.254 0.198 0.339 0.430 0.172 0.407 0.143 0.837 0.173 0.211 0.270 0.259

0.172 0.118 0.083 0.110 0.192 0.114 0.197 0.295 0.225 0.085 0.096 0.088 0.157 0.145 0.213 0.126 0.082 0.313 0.086 0.495 0.107 0.196 0.231 0.209

0.119 0.082 0.042 0.077 0.076 0.090 0.136 0.251 0.192 0.054 0.084 0.073 0.129 0.127 0.170 0.100 0.068 0.275 0.054 0.301 0.080 0.174 0.206 0.178

0.082 0.058 0.035 0.072 0.066 0.085 0.118 0.217 0.186 0.052 0.074 0.066 0.113 0.116 0.133 0.092 0.064 0.239 0.048 0.265 0.068 0.167 0.197 0.170

tot. in. 1.127 0.790 0.510 1.225 0.918 0.901 1.389 1.594 2.045 0.801 0.597 0.819 1.353 1.370 2.026 1.188 0.924 2.126 0.459 2.720 0.909 1.698 1.572 1.487

1

2

3

20.7 31.6 25.9 24.0 29.6 23.1 23.5 24.0 12.9 15.4 18.0 25.4 18.8 14.5 16.7 36.2 18.7 19.1 31.1 30.8 19.0 12.4 17.2 17.4

36.0 46.6 42.1 33.0 50.5 35.8 37.7 42.5 23.9 26.1 34.1 36.2 30.3 25.1 27.3 46.8 27.5 33.9 49.9 49.0 30.8 23.9 31.9 31.5

46.6 57.0 50.4 39.2 58.8 45.7 47.5 58.2 33.2 32.8 48.1 45.1 39.9 34.4 35.6 55.2 34.9 46.8 61.6 60.0 39.6 34.2 44.9 43.5

4 53.9 64.3 57.3 45.1 66.1 55.2 55.9 71.8 42.3 39.2 60.5 53.1 48.2 42.8 42.2 63.0 41.9 58.1 72.0 69.8 47.1 44.0 57.5 54.9

Table 5.5: Eigenvalues and cumulative percentage variance explained for the first four axes of inertia from CAs analysed in section 5.4 omitting analyses not fully discussed. † Analysis by years of issue. All others by issues. N.B. ‘tot. in.’ is the total inertia for the data set. For a discussion of this table see page 176.

such as ‘Hoffman’, are hoards with an inexact or unknown provenance. In the cumulative percentage graphs clarity has been attempted using line styles and labels. The interactive use of colour on screen allows patterning to be observed with greater ease. The eigenvalues and percentage variance explained figures from canoco for all these analyses are given in Table 5.5. In all cases the third, and sometimes the fourth axes of inertia were examined. Where these did not not reveal any significant information they have not been discussed (see also the discussion on page 175). Please note that rather than continually repeat ‘dated by Crawford to. . . ’ or ‘thought by Crawford to be minted in. . . ’ I have usually simply stated ‘dated to’ and ‘struck in’. I am fully aware that there are on-going debates concerning these problems.

5.4.2

Hoards closing 147–118 bc

The data set contained 18 hoards closing between 147–118 bc (Table 5.6). They contained 5,814 denarii. Issues from 211–158 bc formed 8.5% of the data set. Figure 5.17 is the cumulative percentage graph of these data; Figures 5.19–5.18 are the maps from CA by issues. This data set covers a rather large period of time due to the small number of early hoards currently in the chrr database and the CAs of this data set are, as a result, dominated by the time sequence. Both the CA using issues and using years as variables were very similar and the Procrustes Analysis showed that the 65

5. Comparing hoards — Correspondence Analysis

code

name

ban can fos ger jes lcl mas pac pet pta ric rom s01 sgi sto sy2 tds zas

Banzi Cani Islands Fossombrone Gerenzago † Jesi † Lucoli Maserà Pachino Petacciato Patrica Riccia Rome ‘West Sicily’ San Giovanni Incarico Stobi Syracuse Terranova di Sicilia † Zasiok

rrch 157 132 — 167 — 164 162 151 149 — 161 131 135 163 — 154 168 166

country

closing date

‘good total’

130 146 121 118 118 124 125 138 141 119 126 147 146 125 125 136 118 120

124 95 66 49 67 59 1016 30 224 100 2866 113 36 180 497 59 71 162

Italy Tunisia Italy Italy Italy Italy Italy Sicily Italy Italy Italy Italy Sicily Italy Fmr Yugoslavia Sicily Sicily Fmr Yugoslavia

Table 5.6: Hoards closing 147–118 bc used in CA discussed in section 5.4.2. † Also occurs in the next data set (see Table 5.7).

100 90 80

cumulative percentage

70 60 50 40 30 20 10 0 211

201

191

181

171

161

151

141

131

121

year BC

Figure 5.17: Cumulative percentage graph of hoards closing 147–118 bc. Dashed lines: hoards closing 125 bc; solid lines: hoards closing 118 bc; dotted lines: all others.

66

+5.5

5.4.2. Hoards closing 147–118 bc 62

157

152 156 153

110

209

134

198

-1.0

-1.0

146 129 138 89 163 58 140 162 125 197 203200 206 117 202 217 130 139 208 154 114 164 199 215 53 113 167 182 218 88 141 5072 44 147 55 106 54 220 60 224 221 265 57169 115 112 122 161 227 136 78 228 241 155 256 133 229 226 126 132 225

79

278 271

135

280

158 116

270

276

277 68

279

273

282 281

274

283

275

+8.0

262 80 46

Figure 5.18: Variable map from CA of hoards closing 147–118 bc. First and second axes of inertia, data points are RRC issue numbers.

greatest difference occurred in the order of the four earliest hoards. The analysis by issues is presented here. Despite the dominance of the time sequence a number of interesting points can be made. Firstly, the three hoards closing in 125 bc all seem very similar indeed. They are plotted very close together in Figures 5.17 and 5.19. The three hoards from 118 bc seem equally similar in Figure 5.17 but are more widely spaced in Figure 5.19. If we compare each group of three using the Kolmogorov Smirnov test (Shennan 1997, pp. 57–61) we find that none of the hoards from 118 bc are significantly different from each other at the 0.05 level, whereas the Maserà hoard (mas) is significantly different from Stobi and San Giovanni Incarico (sto & sgi).17 The following points help to explain these results: 1. Hoard size. (a) The three hoards from 118 bc are much smaller than those from 125 bc. (b) Maserà is the largest hoard. The size of samples compared using significance tests has a major influence 17

ger v. tds: Dmax = 15.51, Dmax0.05 = 25.2; ger v. jes: Dmax = 14.05, Dmax0.05 = 25.6; tds v. jes: Dmax = 10.56, Dmax0.05 = 23.2; sgi v. sto: Dmax = 6.26, Dmax0.05 = 11.8; mas v. sto: Dmax = 9.43, Dmax0.05 = 7.4; mas v. sgi: Dmax = 12.83, Dmax0.05 = 11.0.

67

5. Comparing hoards — Correspondence Analysis

+5.5

Italy Sicily former Yugoslavia Tunisia

s01 146 rom 147 can 146 pet 141

pac 138 ger 118 tds 118

lcl 124 fos 121

ban 130

-1.0

pta 119

+8.0

jes 118

zas 120

-1.0

sto 125 mas 125 sgi 125 sy2 136 ric 126

Figure 5.19: Sample map from CA of hoards closing in 147–118 bc. First and second axes of inertia, data points are hoards. The closing dates of the hoards are also given.

on the result (Shennan 1997, pp. 113–5). The larger the hoard, the more likely one is to have a significant result. 2. New years are added to the pool faster than old years fall out of circulation resulting in the hoards from 118 bc having coins from 124–118 bc in addition to the earlier issues. Note that: (a) Coin issues RRC 273–283 (124–118 bc) form only 1.3% of the total data set (74/5,814). (b) Coins issues RRC 273–283 form 25.7% of the hoards closing in 118 bc (48/187). This results in the CA giving more ‘weight’ to the variation in these few issues compared to variation in issues occurring in all hoards. The pattern of the latest hoards being quite spread out on one extreme of a curve is a common feature of CA of hoard data from a wide spread of years (see Lockyear 2000a, and page 176 below). The pattern was exaggerated here due to the factors listed above. Figure 5.17 also exhibits a very stepped profile up to 157 bc. This is a consequence of the comparatively poorly known dating of these early issues, which are assigned to wide date brackets. The steep rise in coinage levels after 68

5.4.3. Hoards closing 118–108 bc

157 bc is also notable, although it is possible that this is partly due to Crawford’s dating scheme. ‘West Sicily’ (s01) has a large proportion of early coins compared to Rome, Cani or Petacciato (rom, can & pet). The small hoard from Pachino (pac) also has a large proportion of early coins. From such a small sample it is difficult to draw definite conclusions but it would seem possible that early coins are better represented in Sicilian hoards than mainland Italy. A number of early issues were minted in Sicily (e.g., RRC 68–81; 211–209 bc) whereas all issues from 157–118 bc bar one18 were, according to Crawford, minted in Rome. Of the hoards from 118 bc the Jesi hoard (jes) has the most early coins. Otherwise, the three hoards are remarkably similar as noted above.19 Likewise, the similarities between the hoards from 125 bc is also marked despite, for example, Maserà being in northern Italy, San Giovanni Incarico being south of Rome, and Stobi being in the former Yugoslavia. Unfortunately, archaeological interpretation of cross-period levels of similarity is extremely difficult (see Chapter 7 and Lockyear 1993). Finally, the Yugoslav hoards from Zasiok and Stobi (zas & sto) do not appear particularly different from the Italian material.

5.4.3

Hoards closing 118–108 bc

The data set contained 16 hoards closing between 118–108 bc (Table 5.7). They contained 2,423 denarii. Issues from 211–158 bc formed 4.6% of the data set and were omitted from the CAs. Figure 5.20 is the cumulative percentage graph of these data; Figures 5.22–5.21 are the maps from CA by years of issue. About half of this data set comes from Spain, the remainder from Italy and Sicily. The sample map (Fig. 5.22) shows most of the Spanish hoards on the right of the plot, the majority of the Italian hoards on the left. Pozoblanco (pzl) is isolated at the top of the second axis. Figure 5.20 also shows that the Spanish hoards mainly have a ‘modern’ profile compared to Italy which is a reflection of their closing dates as shown below: closing date Italy Spain

118

117

116

3

1

1

115

114

113

112

1

1 3

1 1

111

110

109

108

1

2

1

As a result, in all hoards from Spain, except Baix Llobregat (llo), coins from 118–108 bc account for 14–35% of the hoard. Of the Italian hoards, Taranto (tr1) has 15%, but the others have only 0.5–5%.20 It is unlikely that this pattern is merely fortuitous and it would seem that coins of this period are particularly associated with Spain. Even the latest of the Italian hoards, Borgonuovo (brg) has only 0.5% of its coins from 118–108 bc. Within the Spanish material, their order on the first axis is not determined by their closing date and so their geographical distribution was examined (Lockyear 1996b, Fig. 8.17). The hoards fall into two groups, the first come from Andalucía, the second from Cataluña but this in not reflected in the CAs. 18

The issue from Narbo (Narbonne), RRC 282, 118 bc. For years 211–158: jes: 7 coins, 10.5%; tds: 3 coins, 4.2%; ger: no coins. 20 Coins from 118–108 bc in Spanish hoards: co1 35%; seg 35%; sar 25%; adr 24%; lab 20%; el1 18%; pz1 20%; co2 14% llo 5%. In Italian hoards: tr1 15%; mad 5%; bev 4%; ger 4%; tds 4%; jes 3%; brg 0.5%. 19

69

5. Comparing hoards — Correspondence Analysis

code

name

adr bev brg co1 co2 el1 ger jes lab llo mad pz1 sar seg tds tr1

Alcalá del Río Bevagna Borgonuovo Villanueva de Córdoba Córdoba El Centenillo Gerenzago † Jesi † La Barroca Baix Llobregat Maddaloni Pozoblanco Sarrià Segaró Terranova di Sicilia † Taranto

country

rrch — 171 — — 184 181 167 — 178 — 172 174 — 180 168 176

closing date

‘good total’

112 117 112 113 109 110 118 118 112 109 116 115 108 112 118 114

158 721 215 127 214 71 49 67 69 112 283 79 48 43 71 96

Spain Italy Italy Spain Spain Spain Italy Italy Spain Spain Italy Spain Spain Spain Sicily Italy

Table 5.7: Hoards closing 118–108 bc used in CA discussed in section 5.4.3. † Also included in the previous data set (see Table 5.6).

100 90 80 llo

cumulative percentage

70 60 50

co1

40 30 20 10 0 157

152

147

142

137

132

127

122

117

112

year BC

Figure 5.20: Cumulative percentage graph of hoards closing 118–108 bc Solid lines: hoards from Italy and Sicily; dashed lines: hoards from Spain; dash-dot line: the Pozoblanco hoard (pz1).

70

71

124

207

154

208

109

110

111

122

121

119

108

112

115

114

116

113

+3.0

Figure 5.21: Variable map from CA of hoards closing in 118–108 bc. First and second axes of inertia, data points are years of issue.

199

-1.5

128

120

209 138 134 206 147 151 137 189 143 126 125 136 155 123 157 141 149 135 153 152 211 179 169 194 150 145 132117

142 139 140 130

148

133

127

144

118

llo 109

brg 112

co2 109

el1 110

sar 108

tr1 114

mad 116

pz1 115

lab 112 adr 112

seg 112

co1 113

+3.0

Figure 5.22: Sample map from CA of hoards closing in 118–108 bc. First and second axes of inertia, data points are hoards. Labels include the closing date for each hoard.

-1.5

ger 118 tds 118 jes 118 bev 117

+3.0 -2.5

+3.0

-2.5

Italy Sicily Spain

5.4.3. Hoards closing 118–108 bc

5. Comparing hoards — Correspondence Analysis

The position of Baix Llobregat (llo) in both Figures 5.20 and 5.21 makes it appear similar to Italian hoards, apart from a relative lack of coins of 118 bc, which characterises most Spanish hoards, — see below. This similarity is mainly because all have very few coins of 116–109 bc. The second axis is dominated by the Pozoblanco hoard (pz1). Examination of the variable map and the cumulative percentage graph (Figs. 5.21 & 5.20) shows this hoard to have high numbers of coins of 119 and 118 bc.21 The coins from 118 bc in that hoard were minted at Narbo — a Roman colony in SW France. However, an examination of Figure 5.21 shows that apart from this hoard, coins from 118 bc are more associated with Italian hoards than Spanish material.22 Pozoblanco lies in south central Spain whereas other hoards, e.g., Segaró, lie much closer to Narbo. Pozoblanco also has the highest proportion of the other four years which stand out on the second axis (144, 133, 127 & 120). This hoard is clearly unlike others from either Spain or Italy in some details of its profile.

5.4.4

Hoards closing 105–97 bc

The data set contained 25 hoards closing between 105–97 bc (Table 5.8). They contained 6,181 denarii. Issues from 211–158 bc formed 3.3% of the data set and were made omitted from the CA. Figure 5.23 is the cumulative percentage graph of these data; Figures 5.25–5.24 are the maps from CA by years of issue. There are no hoards with over 30 ‘good’ denarii closing 107–6 bc; hence the gap between this data set and the previous one. No coin issues were struck in 107 bc and no hoards close with RRC 311–313. Two hoards close in 97 bc as they contain quinarii of this date but the quinarii have been excluded from the analysis. Only two issues of denarii are dated by Crawford to the period 99–93 bc.23 It is always possible, of course, that a hoard was concealed or lost at some unknowable period after the ‘closing’ date of the hoard, and in this data set there is an increased likelihood of this being true for the eleven hoards closing in 100–97 bc. The hoard from Torre de Juan Abad (jua) raises the problem of ‘extraneous’ coins. In RRCH Crawford lists this hoard (RRCH 189) as having 480 coins and closing with the issue of L. Thorius Balbus (RRC 316/1). The hoard had been catalogued by Crawford and was used in RRCH, Table XI. No mention was made of extraneous coins (cf. Córdoba, RRCH 184). When the hoard was published (Vidal Bardán 1982) two of the 478 coins listed were of a later date.24 These were said to be ‘logically’ not part of the hoard (Vidal Bardán 1982, p. 80),25 i.e., they are thought to be extraneous. As can be seen from Figure 5.25 this hoard, excluding those two later coins, does not appear as unusual and occurs 21

119 bc: 16 coins, 20.25%; 118 bc: 14 coins, 17.7%. Coins of 118 bc in Spanish hoards: co2, llo & seg 0 coins; adr 2 coins, 1.3%; el1 1 coin, 1.4%; lab 1 coin, 1.4%; co1 2 coins, 1.6%; sar 2 coins, 4.2%; pz1 14 coins, 17.7%. Coins of 118 bc in Italian and Sicilian hoards: brg 0 coins; mad 5 coins, 1.8%; tr1 2 coins, 2.0%; bev 20 coins, 2.8%; jes 2 coins, 3.0%; ger 2 coins, 4.0%; tds 3 coins, 4.2%. 23 RRC 334/1, ?97 bc; RRC 335/1a–10b, ?96 bc. 24 Nos. 477 & 478, RRC 366/1a, 82–1 bc, and 383/1, 79 bc. 25 “Los denarios números 477 y 478 datados en los años 82–81 y 79 aC., respectivamente, quedan aislados y separados de los denarios anteriores, y lógicamente son una intromisión no aceptable en el tesorillo.” 22

72

5.4.4. Hoards closing 105–97 bc

code

name

azn blg bvg cac cg2 clg cog crg csl cvg ele gdm iav imo jua lor mnf olm orc pat pnh pue rcn rio sel

Aznalcóllar Bologna Bevagna Cachapets Cerignola Castillo de las Guardas Cogollos de Guadix Crognaleto † Cástulo Carovigno Elena Gioia dei Marsi † Idanha-a-Velha Imola Torre de Juan Abad San Lorenzo del Vallo Manfria Olmeneta Orce Paterno Penhagarcía Puebla de los Infantes Ricina Rio Tinto Santa Elena

rrch — — — — — — — 212 — 208 199 213 — 210 189 195 198 203 211 207 191 — 201 194 193

country

closing date

‘good total’

104 100 100 101 100 105 104 97 101 100 101 97 100 100 105 102 103 100 100 100 104 103 101 102 101

35 92 227 262 96 113 83 137 47 459 59 220 1340 500 476 299 32 397 72 149 103 131 271 44 537

Spain Italy Italy Spain Italy Spain Spain Italy Spain Italy Italy Italy Portugal Italy Spain Italy Sicily Italy Spain Sicily Portugal Spain Italy Spain Spain

Table 5.8: Hoards closing 105–97 bc used in CA discussed in section 5.4.4. † Hoards close in 97 bc with a quinarius issue not included in the analyses.

within a group of Iberian peninsula hoards. The problem of extraneous coins will be further discussed below (page 173). Of the 25 hoards 12 come from the Iberian peninsula, 13 from Italy and Sicily. As with the previous data set, the distribution over time is not even: closing date Italy/Sicily Spain/Portugal

105

104

103

102

101

100

97

3

1 1

1 1

2 3

7 2

2

2

The previous data set had a concentration of Spanish hoards in the second half of its date range, this set in the first half. As might be expected given this, the Italian and Sicilian hoards have, generally, a more modern age profile (Fig. 5.23). An examination of the variable map (Fig. 5.24) reveals a concentration of late years at the positive end of the first axis. The majority of years prior to 117 bc have negative scores on the first axis. The date distribution of hoards by country is reflected in the sample map (Fig. 5.25) where the Italian hoards mainly to the right of the map and the Iberian peninsula hoards to the left. Manfria (mnf) appears with the Iberian peninsula hoards partly due to its early closing date; Orce (orc) lies within the Italian group due to its late closing date. Although six of seven hoards closing 105–3 bc appear at the negative end of the first axis, patterning by closing date in the rest of the sequence is quite mixed. The position of Idanha-a-Velha (iav) on the sample map (Fig. 5.25) is due to coins from 103 bc which account for 11.3% of this hoard. No other hoard has 73

5. Comparing hoards — Correspondence Analysis

100

pue

90 80 70 cumulative percentage

iav 60 orc

50 40 30 20 10 0 157

152

147

142

137

132

127

122

117

112

107

102

year BC

(a) Iberian peninsula plus Manfria (mnf, bold line) 100 90 80

cumulative percentage

70 60 50 40 rcn

30 20 10 0 157

152

147

142

137

132

127

122

117

112

107

102

year BC

(b) Italy and Sicily, Manfria in bold.

Figure 5.23: Cumulative percentage graphs of hoards closing in 105–97 bc. Solid lines: hoards from Italy and Sicily; dashed lines: hoards from Spain; bold line: Manfria.

74

112

117

109

103

106 108

101

102

104

111 105

+1.5

100

-2.0

pue

pnh

cog

mnf

jua

csl

clg

sel

azn

olm

rio

cac

lor

+2.5 rcn

cvg

gdm

iav

orc

cg2

crg

imo

ele

bvg

blg

pat

+1.5

Figure 5.24: Variable map from CA of hoards closing in 105–97 bc. First Figure 5.25: Sample map from CA of hoards closing in 105–97 bc. First and second axes of inertia, data points are years of issue. and second axes of inertia, data points are hoards.

-2.0

155

138 140 135 144 137 142 136 134 151 127 132 153 146 150 154 128 126 125 129 147 141 130 149 131 152 116 139 145 122 113 124 119 115 133 120 110 123 118 114 143

148

157

+2.5 -1.5

75 -1.5

Italy Sicily Spain Portugal

5.4.4. Hoards closing 105–97 bc

+2.0

5. Comparing hoards — Correspondence Analysis

Italy Sicily Spain Portugal

pz1

bev jes pue

mnf

tds ger

imo rcn

mad

pat olm cvg

sar

azn

-1.5

blg crg

brg

cog

el1

+1.5

pnh tr1

bvg ele

rio

lor cac

orc

lab

llo

csl cg2

clg

gdm

co2 jua co1

sel

adr

-1.0

seg

Figure 5.26: Map from CA of hoards of 40 hoards closing 118–110 bc analysed by issues.

more than 3.5%.26 In Figure 5.23 the line for Idanha-a-Velha cuts across many of the other lines at this date, emphasising its uniqueness. The diagnostic statistics show that this hoard has an extremely high relative contribution to the second axis of 938 and 103 bc has a relative contribution of 827. This nicely illustrates how the diagnostic statistics help interpret the CA maps. The hoard from Ricina (rcn) lies almost at the origin of the map. This is often a sign that an object is poorly represented by the map, and indeed looking at the diagnostic statistics shows its score for ‘quality’ is very low at 7. Looking at Figure 5.23b we can see that it does indeed have a very strange profile with very large numbers of early coins and none of the usual peaks. One could speculate that the hoard had been cherry-picked for good examples of types before being catalogued leading to a smoother curve than usual. 26

Coins of 103 bc: azn, clg, cog, csl, jua, lor, pnh & rio: no coins; cac: 1 coin, 0.4%; sel: 2 coins, 0.4%; pue: 1 coin, 0.8%; cg2: 1 coin, 1.0%; blg: 1 coin, 1.1%; cvg: 6 coins, 1.3%; ele: 1 coin, 1.7%; rcn: 5 coins, 1.8%; crg: 3 coins, 2.2%; imo: 13 coins, 2.6%; gdm: 6 coins, 2.7%; orc: 2 coins, 2.8%; olm: 12 coins, 3.0%; mnf: 1 coin, 3.1%; pat: 5 coins, 3.4%; bvg: 8 coins, 3.5%; iav: 151 coins, 11.3%.

76

5.4.5. Re-examining hoards from 118–97 bc by issues +2.0

to 121 BC 120 to 116 BC 115 to 111 BC 110 to 106 BC 105 to 100 BC

282

327

281

329

280 330

-1.5

+1.5

284 283

328

-1.0

285

286

Figure 5.27: Map from CA of issues from 40 hoards closing 118–110 bc.

5.4.5

Re-examining hoards from 118–97 bc by issues

At the time of the original analyses the chronological differences between the Italian and Iberian peninsula hoards described in the previous two sections was observed, but a note of caution was sounded because the hoards had been analysed by Crawford’s dating scheme, and there was the possibility that the pattern revealed was an artefact of that scheme (Lockyear 1996b, sections 8.3.3–8.3.4). In order to examine this problem a larger data set was extracted from the database including all the hoards closing 118–97 bc but this time by RRC issues. The 41 hoards included are those listed in Tables 5.7 and 5.8. For this particular analysis, however, including all issues might confuse the pattern in the most recent ones in which we are interested. I decided, therefore, to omit all the issues prior to RRC 260 which is dated by Crawford to 128 bc, i.e., ten years before the earliest hoard in the data set. As a result some hoards were much smaller than the 30 coin minimum size usually employed. The initial analysis showed some promise but the hoard from Idanha-a-Velha (iav) had a strong influence on the result due to having 145 examples of RRC 319, over 80% of the coins of that type in the data set. Re-running the analysis without Idanha-a-Velha produces clear maps with a marked horseshoe curve 77

No examples of 286 232.3, the critical value for p = 10; ν = 140; α = 0.05.

92

5.4.9. Hoards closing 78–75 bc

0

500

1000

1500

●

●

●

●

● ● ●

−500

●

−500

0

500

1000

1500

Figure 5.41: Procrustes Analysis comparing object maps from CA of hoards closing 80–79bc analysed by years of issue v. issues.

code

name

adm alx cor inu ion ker mal man mbr ner noy ran ste zat

Alba di Massa Alexandria Cornetu (Căpreni) Inuri Montalbano Ionico Kerassia Maluenda San Mango sul Calore Mihai Bravu Neresine, Lussino Island Noyer Randazzo Stejeriş Zătreni

rrch 289 295 296 — 297 283 282 294 — — — 287 — —

country Italy Romania Romania Romania Italy Greece Spain Italy Romania Fmr Yugoslavia France Sicily Romania Romania

closing date

‘good total’

77 77 75 77 75 78 78 75 75 78 78 77 75 75

82 32 128 37 45 47 32 81 56 42 51 30 200 41

Table 5.12: Hoards closing 78–75 bc used in CA discussed in section 5.4.9.

93

5. Comparing hoards — Correspondence Analysis

MNT SPO

PL1 AMA

FRA

Figure 5.42: Map of Italian hoards closing 80–79 bc.

94

5.4.9. Hoards closing 78–75 bc 100 90 80 ran

cumulative percentage

70 60 50 40 30

noy

20

ner

10 0 157

152

147

142

137

132

127

122

117

112

107

102

97

92

87

82

77

year BC

+3.0

Figure 5.43: Cumulative percentage graph of hoards closing 78–75 bc. Solid lines: Italy; dashed lines: Romania; dash-dot lines: other. 201 202

359

297 360 371292 311 266 351 223 329 248 219 317 263 203 312 255 291 308 378 281 357 276 287 290 380 387 244 384 305 215 285 362 231 316 208 340 326 388 300 250 364 386 372 296 295 314 239 392 334 318 352 341 279 350 379 366 354 270 344 342 302 278 273 301 232 328 337 260 204 286 274 385 298 259 275 383 335 361 237 363 345 346 261 353 249 322 382 282 367 262 304 307 245 252 327 325 299 210 374 280 348 217 391 277 236 220 289 235 247 267 205 214 324

-2.0

303 323

234

319 306

+2.5 228 238 284

256224233 218257216

349

243 320

-2.0

313

Figure 5.44: Variable map from CA of hoards closing 78–75 bc. First and second axes of inertia, data points are RRC issue numbers.

95

+3.0

5. Comparing hoards — Correspondence Analysis

Italy Sicily Spain Greece Romania former Yugoslavia France

cor

noy ran

zat

mal

mbr

-2.0 ker

+2.5

alx

adm ner

ste

inu ion

-2.0

man

Figure 5.45: Sample map from CA of hoards closing in 78–75 bc. First and second axes of inertia, data points are hoards.

a very sparse data set as grouping issues by years reduces the percentage of zero entries in the data to 53%.

5.4.10

Hoards closing in 74 bc

The data set contained 23 hoards closing in 74 bc (Table 5.13). They contained 7,371 denarii. Issues from 211–158 bc formed 1.7% of the data set and were omitted from the CAs. Figure 5.46 is the cumulative percentage graph of these data; Figures 5.47–5.50 present the maps from the CAs. The first question that arises is why are there so many hoards closing at this date? Crawford assigns four issues to 74 bc, RRC 394–397. All but two of the hoards include examples of the common issue RRC 394. Crawford’s die estimates make this issue much the same size as the previous one (RRC 393) which he dates to 76–75 bc. The admittedly small previous data set, however, contains no examples of that coin, this data set only contains eight examples, all from the Pontecorvo hoard (pon), the next data set only has two examples from Alt Empordà (emp). Hoards from 63–54 bc (section 5.4.12) contain 146 examples of this coin but 139 of those are from the huge Mesagne hoard (mes). The dating of issues, particularly between the Social War and the Civil War, is

96

5.4.10. Hoards closing in 74 bc

code

name

bdr cab cdr cos ctr hn4 it2 jdi lic mac mar mtr ori pey pic pl2 pon poo rig smb sp2 suc tuf

Barranco de Romero Cabeça de Corte Castro de Romariz Cosa Canturato Hunedoara IV ‘Italy’ Jdioara Licodia Maccarese Rio Marina Cergnago (Mortara) Oristà Peyriac-sur-Mer Potenza Picena Palestrina Pontecorvo Poio Rignano Las Somblancas ‘Spain’ Sučurac Tufara

rrch — 300 — 313 301 303 — — 308 309 306 286 — 304 312 315 311 305 564 — 307 310 —

country Spain Portugal Portugal Italy Italy Romania Italy Romania Sicily Italy Elba Italy Spain France Italy Italy Italy Portugal Italy Spain Spain Fmr Yugoslavia Italy

closing date

‘good total’

74 74 74 74 74 74 74 74 74 74 74 74 74 74 74 74 74 74 74 74 74 74 74

65 158 70 1999 50 42 47 67 120 1212 43 654 58 92 439 357 942 211 92 84 246 165 158

Table 5.13: Hoards closing in 74 bc used in CA discussed in section 5.4.10.

a matter of quite some debate (Harlan 1995; Hersh & Walker 1984; Mattingly 2004, see section 5.5.2;). Mattingly (2004, p. 284–5, 287) argues that the two variants of this issue, RRC 393/1a and 393/1b are in fact two different issues and that 393/1b dates to 58 bc.30 RRC 393/1a is, however, still the more common issue and only occurs in four hoards closing prior to that date — the Mihăeşti, Pontecorvo, Alt Empordà and San Gregorio di Sassola hoards (mha, pon, emp & gre) — all of which have examples of RRC 394/1a–b. It would seem likely, therefore, that RRC 393/1a should also be dated somewhat later given that eventually it is as common as RRC 394/1a: the former has 379 examples in the chrr database, the latter 421. The two hoards that do not close with RRC 394 — Castro de Romariz and Las Somblancas (cdr & smb) — both end with a single example of RRC 395. Perhaps this issue is also out of sequence? The block of hoards closing with RRC 394 does, however, indicate that it was a large issue followed by a series of smaller issues. The cumulative percentage curve graph shows that the majority of the hoards have a very similar profile to each other with the exception of the ‘Italy’ and Canturato hoards (it2 & ctr) which are very modern in their profiles, and the ‘Spain’ hoard (sp2) which is archaic. There are only two Romanian hoards closing at this date, neither of which seem notably archaic in profile. The first two axes of the CA only ‘explained’ 26.1% of the variation in the data, only marginally more than the previous analysis. Unsurprisingly the sample map (Fig. 5.48) highlights the modern profile hoard ‘Italy’ on the first axis and ‘Spain’ on the second. A second analysis omitting those two hoards (not 30

This is a rare (unique?) case where one of Crawford’s issues has been split in this fashion.

97

5. Comparing hoards — Correspondence Analysis

100 90 80 sp2

cumulative percentage

70 60

hn4

50 40 30 20 10 0 157

152

147

142

137

132

127

122

117

112

107

102

97

92

87

82

77

year BC

(a) Thick solid line: Cosa hoard; dotted lines Iberian peninsula (except sp2); thin solid line: sp2; dashed lines: Romania. 100 90 80

cumulative percentage

70

pon

60 50 40

it2

30 ctr 20 10 0 157

152

147

142

137

132

127

122

117

112

107

102

97

92

87

82

77

year BC

(b) Thick solid line: Cosa hoard; solid lines Italy (except ctr); dashed line: ctr; dotted lines: France, Elba and Yugoslavia.

Figure 5.46: Cumulative percentage graphs of hoards closing in 74 bc.

98

+3.5

5.4.10. Hoards closing in 74 bc

295

322

366 296

202

282 327

326 297

325

316 328 324

371 290 334 287 247 234 198 301 386 305 323 279 260 293 267 304 362 385 218 242 277 298 379 372 340 255 364 225 361 239 294 330 257 312 207 238 217 280 226 363 391 360 252 253 357 314 341 250 249 203 197 256 201 199 206 233 208 248 224 222 221 210 235 232 231 244 204 200 263 205 214 318

265

-2.5

230

359

376 229

375

380

+3.0

-3.0

241 370 227

240

Figure 5.47: Variable map from CA of hoards closing in 74 bc. First and second axes of inertia, data points are RRC issue numbers.

presented) was no more successful: its first two axes ‘explained’ 24.2% of the variation. The ‘Italy’ hoard is one of the smaller ones in this data set. Looking at the map of issues (Fig. 5.47) and the diagnostic statistics we can see that this hoard has very high numbers of RRC 378–383, issues dated by Crawford to 81–79 bc but the we can also see that the Hunedoara IV, Maccarese, Canturato and Barranco de Romero hoards have much of this coinage, 1 10% in each case. Picking out the issues with a relative contribution of over 300 to the first axis and a negative score we can see that most of them date to 138–101 bc.31 The four hoards at the negative end of the first axis, ‘Spain’, Poio, Jdioara and Pontecorvo (sp2, poo, jdi & pon) all have 1 40% of their assemblages dating to this period. The first axis is thus showing the contrast between those two blocks of issues. 31

Issues with a negative score on the first axis and a relative contribution 1 300 are RRC 200, 233, 235, 247, 265, 274, 318 and 325.

99

Italy Sicily Spain Portugal Romania former Yugoslavia Elba France

+3.5

5. Comparing hoards — Correspondence Analysis

sp2

cab

ori

-2.5

mtr pl2 smb lic rig pey pic mar cos cdr suc

poo jdi pon

hn4

it2 mac ctr

+3.0

tuf

-3.0

bdr

Figure 5.48: Sample map from CA of hoards closing in 74 bc. First and second axes of inertia, data points are hoards.

The second axis only has a single issue with a high relative contribution at the negative end of the axis (RRC 274) and a group of eleven issues at the positive end.32 The fact that there are blocks of issues creates confidence in the result. The ‘Spain’ hoard has 31.7% of its total consisting of those issues, and Cabeça de Corte and Oristà (cab and ori) both have more than 10%. It is interesting to note that all three of these hoards are from the Iberian peninsula, and that these issues were all involved in the earlier analysis comparing Italy and Spain (see section 5.4.5). The third axis (Figs. 5.49–5.50) only ‘explains’ 6.7% of the variation in the data. Only five hoards have a relative contribution of more than 100, and only Cergnago (mtr) has a score more than 200. The axis picks out a few issues such as RRC 342 of which ‘Italy’ unusually has none, Cergnago 7.3% and Jdioara 7.5% but most of the hoards do not fit this pattern. 32 Issues with a positive score on the second axis and a relative contribution 1 300 are RRC 282, 295–297, 316, 322, 325–328 and 366.

100

101

241 370 389 227

225

377

198

359 266 388 247 200 203 301 318371 330 210 305 366 207 235 236 324 229 321 242 265 326 199 290 231 218 259 204 325 282 296 287 260 219 298 363 316 297327 201 293 322 214 232 228 216 353 365 328 202 277 221 234 206 215 302 323 197 205 217 255 281 280 304 224 356 257 220 295

+3.5

Figure 5.49: Variable map from CA of hoards closing in 74 bc. Second and third axes of inertia, data points are RRC issue numbers.

-3.0

240

253

208 252 336262 249 244 243 314

230

226

375

jdi

hn4

mtr

smb

ctr

pl2 lic rig mac

it2

cab ori

sp2

+3.5

Figure 5.50: Sample map from CA of hoards closing in 74 bc. Second and third axes of inertia, data points are hoards.

-3.0

cos

pon suc mar pic cdr tuf

bdr

pey

poo

+4.5 -4.5

+4.5 -4.5

Italy Sicily Spain Portugal Romania former Yugoslavia Elba France

5.4.10. Hoards closing in 74 bc

5. Comparing hoards — Correspondence Analysis

To summarise, the hoards on the whole are remarkably similar. The CA has shown however, that there is a core of extremely similar hoards mainly from Italy, and then a contrast between some hoards with coins from the 130s and 120s v. coins of 81–79, and some hoards which have above average numbers of coins of the period 118–100 bc.

5.4.11

Hoards closing 73–69 bc

The data set contained nine hoards closing in 73–69 bc (Table 5.14). They contained 3,286 denarii. Issues from 211–158 bc formed 2.0% of the data set and were omitted from the CAs. Figure 5.51 is the cumulative percentage graph of these data; Figures 5.52–5.53 present the maps from the CA by issues. Figure 5.51 shows a remarkably homogeneous set of curves for this data set. Only the Tolfa hoard (tol) stands out as having a comparatively ‘modern’ age profile despite closing in 72 bc. Policoro (plc), the most archaic hoard, is also Italian. The Romanian and Spanish hoards are in the generally in the middle ground. The first two axes of inertia from the CA ‘explained’ 34.1% of the variation on the data which is higher than for the previous two sections. On the other hand, this data set is somewhat smaller than the previous one and the total inertia for this table is very low (Table 5.5). The decrease in ‘percentage explained’ from one axis to the next is steady. I decided, therefore, it would be useful to test the significance each axis (see page 5.2.7; Greenacre 1993, p. 173) and found that the the first two axes were significant at the 0.05 level.33 Unsurprisingly, the sample map (Fig. 5.53) picks out the Tolfa hoard (tol) on the first axis. The variable map picks out RRC 396, 397 and 399 with which the Tolfa hoard closes. The diagnostics show that the Alt Empordà hoard also has a high relative contribution to the first axis and it closes with RRC 400 and 401. This is not, however, a simple matter of date as Tolfa has neither of those issues and Alt Empordà has none of the issues with which the Tolfa hoard closes. These small differences are insufficient, however to make Tolfa stand out on the cumulative percentage curve graph. To understand this we can pick out issues with a high values for mass, a high relative contribution to the first axis and a positive score. From this we can see that issues RRC 378–384 are important and indeed those issues form 15.5% of the Tolfa hoard compared to 4.4% in the Alt Empordà hoard and 6.7% in the Tunşi hoard. A wide selection of issues contribute highly to the negative end of the first axis but these are mainly in the range RRC 203–348 and so we must conclude that Tolfa has lower numbers of early issues, generally prior to the Social War. On the whole, the large issues of the Social War do not contribute highly to this axis suggesting that some 20 years after the war the distribution of these coins was now relatively even. The second axis is mainly highlighting the differences between Sfinţeşti and Castelnovo (sfi & cst) in a variety of issues, e.g., RRC 344 v. RRC 372.34 At 33 2 χ = λ × n where λ is the eigenvalue for that axis and n is the number of coins in the analysis. Thus for the first axis λ = 0.107 and n = 3220 (after removal of the early issues) giving 0.107 × 3220 = 344.54. For axes 2–4 χ2 = 309.12, 270.48, and238.28 respectively. The critical value for a 10 × 180 table at the 0.05 level is 283.0 so we reject H0 for axes 1 and 2 and accept it for axes 3 and 4. 34 sfi: RRC 344: 1 coin, 1.1%, RRC 372: 5 coins, 5.5%; cst: RRC 344: 23 coins 5.9%, RRC 372: 2 coins, 0.5%.

102

5.4.11. Hoards closing 73–69 bc

code

name

cst emp oss plc sfi tin tol tun vpt

Castelnovo Alt Empordà Ossero Policoro Sfinţeşti Tincova Tolfa Tunşi Villa Potenza

country

rrch — — 316 — 320 — 317 — 319

closing date

‘good total’

71 71 72 72 71 69 72 73 71

391 1122 465 302 91 135 238 131 411

Italy Spain Italy Italy Romania Romania Italy Romania Italy

Table 5.14: Hoards closing 73–69 bc used in CA discussed in section 5.4.11.

100 90 80

cumulative percentage

70 60 50 40 tol 30 plc 20 10 0 157

152

147

142

137

132

127

122

117

112

107

102

97

92

87

82

77

year BC

Figure 5.51: Cumulative percentage graph of hoards closing 73–69 bc.

103

72

+4.0

5. Comparing hoards — Correspondence Analysis

73 BC 72 BC 71 BC 69 BC all other issues

371 229 199 292 287

405 234 290

392 401 286

249 239 226 241

-2.0

291 301

372 314 298

263 245 218 302 391 328 279 235 326 233 305 217 253 208 325 367 206 275 210 317 311 384 197 207 362 318 238 222 380 387 285 205 346 216 232 243 250 236 261 271 334 274 220 242 336 323 224 360 223 244369 231 215 247 310 202 266 219 248 293 289 203 375 201 297 398 259 256 252 345 204 230 214 254 295 262 228 260 400 330 237

277

399

397

200

+4.5 396

378

269

303

267

-3.0

240

209

Figure 5.52: Variable map from CA of hoards closing 73–69 bc. First and second axes of inertia, data points are RRC issue numbers.

this point we must ask, as Wang (1993) argues, whether statistical significance is of any practical significance in this case.

5.4.12

Hoards closing 63–56 bc

The data set contained fifteen hoards closing in 63–56 bc (Table 5.15). They contained 7,875 denarii. Issues from 211–158 bc formed 0.5% of the data set and were omitted from the CAs. Figure 5.54 is the cumulative percentage graph of these data; Figures 5.55–5.56 present the maps from the CA. This group of hoards is dominated by the extremely large and important hoard from Mesagne (mes; Hersh & Walker 1984). This hoard was discovered after Crawford published RRC and has resulted in several alternative dating schemes (Harlan 1995, pp. 185–6; Hersh & Walker 1984, Table 2, Mattingly 2004, pp. 280–8). Using the Hersh–Walker dating scheme only changes the closing date of seven of these hoards, and none by more than three years. In the analyses we

104

+4.0

5.4.12. Hoards closing 63–56 bc

Italy Spain Romania

sfi

tin

emp

tol

oss

-2.0

+4.5

vpt plc

tun

-3.0

cst

Figure 5.53: Sample map from CA of hoards closing 73–69 bc. First and second axes of inertia, data points are hoards.

also need to be aware that Mesagne accounts for over 75% of the hoards in this data set, and that five of the Romanian hoards are very small with less than 36 coins in each. From the cumulative percentage curves (Fig. 5.54a) we can see that Mesagne has a modern profile compared to most of the other hoards, but we should take into account that those hoards are mainly from Romania. Mesagne appears fairly similar to the small Sustinenza hoard and the Greek hoard from Kavalla (sus & kav) but the Rutigliano and San Gregorio di Sassola hoards (rut & gre) appear more archaic. The first two axes from the CA ‘explain’ 36.2% of the variation in the data. On the sample map (Fig. 5.56) Mesagne is plotted at the negative extreme of the first axis with which it has an extremely high relative contribution of 965. The Kavalla and Rutigliano hoards (kav & rut) have extremely low relative contributions of less than 10. Amongst the issues (Fig. 5.55) a large number have high relative contributions of greater than 300 to the first axis including a

105

5. Comparing hoards — Correspondence Analysis

100 90 gre 80

rut

cumulative percentage

70 stn 60 50 40 30 20 10 0 157

147

137

127

117

107

97

87

77

67

57

year BC

(a) All hoards. Solid lines: Italy; dot-dash line: Greece; dashed lines: Romania; thick line: Mesagne.

100 90 80

cumulative percentage

70 stn 60 50 40 30 lcr 20 10 0 157

147

137

127

117

107

97

87

77

67

57

year BC

(b) Romanian hoards. Solid lines: stn, amn, dun, baz, smc & icn; dot-dash lines: lcr, aln, fnd & bon.

Figure 5.54: Cumulative percentage graph of hoards closing 63–56 bc.

106

5.4.13. Hoards closing 56–54 bc

code

name

aln amn baz bon dun fnd gre icn kav lcr mes rut smc stn sus

Alungeni Amnaş † Baziaş (Lunca Deal) Bonţeşti Dunăreni † Frauendorf (Axente Sever) † San Gregorio di Sassola Icland † Kavalla Licuriciu Mesagne Rutigliano Someşul Cald † Stăncuţa Sustinenza †

rrch 335 338 293 — — 341 337 — 336 332 — — 321 331 339

country Romania Romania Romania Romania Romania Romania Italy Romania Greece Romania Italy Italy Romania Romania Italy

closing date

‘good total’

59 56 63 62 56 56 58 56 58 62 58 58 56 63 56

32 155 36 36 128 563 529 33 59 63 5940 89 115 34 63

Table 5.15: Hoards closing 63–56 bc used in CA discussed in section 5.4.12. † Also used in the next section (see Table 5.16).

group of issues which only occur in the Mesagne hoard.35 The positive end of the first axis, therefore, is associated with a selection of issues not present in the Mesagne hoard.36 The early rare issues are missing obviously by chance but the later issues are significant as their absence from Mesagne is important dating evidence and Hersh & Walker (1984, Table 2) do indeed move many of those issues to 58–56 bc.37 We would hope for a nice clear horseshoe curve in this sort of situation but here we have the contradiction that nearly all of the hoards have a more archaic structure than Mesagne, and in the case of the ten Romanian hoards — four of which post-date Mesagne on this evidence — this can be quite marked. Thus the CA is showing both the issues not in the Mesagne hoard and the multitude of early issues which are much more highly associated with the archaic, usually Romanian hoards. This has led to the rather messy cloud of data points in Figure 5.55. The second axis appears to split the Romanian hoards into two groups. Figure 5.54b plots these two groups and although we can see that the Stăncuţa hoard (stn) is the most archaic, and the Licuriciu (lcr) is the most modern, the two groups from the CA are not perfectly reflected on that graph apart from around about 86 bc. From about 78 bc the hoards are all remarkably similar and it is a contrast between the the quantities of Social War coinage and earlier coinage that makes the most difference.

5.4.13

Hoards closing 56–54 bc

The data set contained fourteen hoards closing in 56–54 bc (Table 5.16). They contained 2,613 denarii. Issues from 211–158 bc formed 1.2% of the data set and were omitted from the CAs. Figure 5.57 is the cumulative percentage graph of these data; Figures 5.58–5.59 present the maps from the CA. 35

RRC 198, 201, 207, 222, 223, 226, 253, 309, 369, 370, 371. RRC 210, 218, 229, 310, 396, 397, 405, 410, 411, 418, 421, 424, 425, 426, 427. 37 RRC 405: 57 bc; 410: 56 bc; 411: 58 bc; 418: 58 bc; 421: 57 bc; 424: 56 bc; 425: 57 bc; 426: 56 bc; 427 56 bc. 36

107

Italy Romania Greece

421

+4.5

63 BC 62 BC 61 BC 59 BC 58 BC 57 BC 56 BC all other issues

+4.5

5. Comparing hoards — Correspondence Analysis

426, 427 241 224 260 240 254 216 271 354

247

lcr

353 290 242 414, 415, 256 252 416, 422 235 287 238 249 298 280 281 198 244 257 318 279 197 239 325 250 206 413 -1.0 217 200 231

199 417

303

aln

233

fnd

236

228 205 275 245

215 237

sus mes

405

+4.5 219

-1.0

rut

bon

+4.5

kav

424

baz

218 311 273 208 292 248 278 419 321 270 274 314 312 400 425 276 265 330 277 232 214 310 269 399 266 204 304

gre

smc icn dun

amn

stn

410 229 396

-4.5

-4.5

210 283 397, 418

Figure 5.55: Variable map from CA of hoards clos- Figure 5.56: Sample map from CA of hoards ing 63–56 bc. First and second axes of inertia, data closing 63–56 bc. First and second axes of inertia, points are RRC issue numbers. data points are hoards.

The Ancona hoard (an1) has no coinage of before 109 bc which gives it a very modern structure. However, its profile is quite odd. As noted above (page 82), this hoard has data quality problems and it has been omitted from Figure 5.57. Removing the hoard from the CAs had little effect on the overall results. The cumulative percentage curve graph (Fig. 5.57) shows the Romanian and Bulgarian hoards with very similar if somewhat archaic profiles and the three Italian and the remaining Greek hoard with more modern profiles. It is clear that over time the Romanian hoards are becoming more and more archaic compared to the Italian hoards and are rarely receiving much recent coinage. The CA maps (Figs. 5.58–5.59) ‘explain’ 30.3% of the variation in the data. We can see from the sample map that the Thessalonica hoard (ths) is somewhat different from the other hoards standing clear on both the first and second axes. This hoard has a reasonable relative contribution to both axes (236 & 554). The majority of the newer issues from RRC 420 onwards are at the positive end of the first axis — RRC 420–2, 425, and 427–9 all have good relative contributions to it — reflecting their association with the more modern profiled hoards. The exceptions are RRC 424 and 426 which are plotted at the negative end of the first axis because they are rare and occur in the Frauendorf hoard (fnd) although

108

5.4.13. Hoards closing 56–54 bc

code

name

amn an1 buz cln com dun fnd gra icn kar sds smc sus ths

Amnaş † Ancona Buzău Călineşti Compito Dunăreni † Frauendorf (Axente Sever) † Grazzanise Icland † Karavelovo Sălaşul de Sus Someşul Cald † Sustinenza † Thessalonica

rrch 338 344 346 347 345 — 341 349 — — 348 321 339 —

country

closing date

‘good total’

56 55 54 54 55 56 56 54 56 54 54 56 56 54

155 42 48 92 929 128 563 256 33 35 92 115 63 51

Romania Italy Romania Romania Italy Romania Romania Italy Romania Bulgaria Romania Romania Italy Greece

Table 5.16: Hoards closing 56–54 bc used in CA discussed in section 5.4.13. † Also used in the previous section (see Table 5.15).

100 90 80

cumulative percentage

70 60 gra

sus

ths

77

67

50 40

com

30 20 10 0 157

147

137

127

117

107

97

87

57

year BC

Figure 5.57: Cumulative percentage graph of hoards closing 56–54 bc omitting Ancona (an1).

109

+4.8

5. Comparing hoards — Correspondence Analysis

59 BC 58 BC 57 BC 56 BC 55 BC 54 BC all other issues

423

432

434 410 433

418

408

391

388 380

429 428

403 311

280

271

-2.5

301 431 394 263 322 256 366 405 223 417 330 233 414 336 248 351 389 317 229 294 312 386 243 430 293 425 295 426 307 228 413 385 325 252 382 234 215 337 200245 259 363 217 238 275 412 206 208 327 235 236 199 237367 260 219 422 210 303 221 346 240 393 204 205 321375 216261 218 197 345 427 244 267 349 424 242 415 296232 283 262 406 254 214 292 343 348 266 289 421 420 220

+4.0

-2.0

201, 207, 222, 231, 334, 369

Figure 5.58: Variable map from CA of hoards closing 56–54 bc. First and second axes of inertia, data points are RRC issue numbers.

they have very poor relative contributions to the first axis. As well as these newest issues, the positive end of the first axis is also associated with a selection of slightly earlier issues particularly in the range RRC 348–416. Picking on just one as an example RRC 408 has a particularly high inertia, indicating that it contributes significantly to the variation in the data, is well represented on the map (Fig. 5.58). Checking the data we find that none of the hoards to the left of the first axis have an example of this relatively common coin of C. Piso L.f. Frugi, i.e., no Romanian hoard has an example of this coin. Hersh & Walker (1984, Table 2) move the date of this issue down from 67 bc to 61 bc. The lack of this coin in Romanian hoards may well be significant in the debate regarding the date of the supply of coinage to Iron Age Dacia. The negative end of the first axis is associated with a number of early issues with types in the range RRC 215–364 having a good relative contribution. For example, issues RRC 352–354 have a high relative contribution (725, 483 & 408) as well has a high mass (i.e., are common issues) and a high inertia (7.27, 11.45

110

+4.8

5.4.13. Hoards closing 56–54 bc

Italy Romania Bulgaria Greece

ths

an1

buz

kar cln

fnd amn

-2.5

dun smc

gra sus

+4.0

sds icn

-2.0

com

Figure 5.59: Sample map from CA of hoards closing 56–54 bc. First and second axes of inertia, data points are hoards.

& 13.02 respectively38 ). Going back to the data we can see that Frauendorf, Sălaşul de Sus and Călineşti have the highest percentages of these issues (fnd, sds & cln) Thus we can see that the first axis is contrasting archaic v. modern profiled hoards, it indicates where that difference is occurring (coins of the 80s bc v. coins of the 60s bc) and is also picking out some specific and potentially archaeologically significant details. From the sample map (Fig. 5.59) we can see that the second axis is largely picking out differences between the five more modern profiled hoards to the right of the plot, although the Sustinenza hoard (sus) has a poor relative contribution to the axis and Buzău (buz) has a moderate one. The list of issues at the bottom of the second axis (RRC 201 etc.) only occur in the Compito hoard (com) probably by chance as that is the largest hoard in the group. Ignoring those 38

These inertias are ‘high’ in the context of these analyses which contain large numbers of variables, in this case 195 issues with a median inertia of 4.5. For hoards, where the numbers are much lower, the inertias are much higher, in this case having a median of 70.

111

5. Comparing hoards — Correspondence Analysis

code

name

ath ben bhr bra bro cas cr1 cuc grj loc mig ods p06 roa sat tr2 trn

Athens Beneventum ‘Bahrfeldt’ Brandosa Broni Casaleone † Carbonara Cuceu La Grajuela Locusteni Mignano Orbeasca de Sus Padova Roata de Jos Satu Nou Taranto ‘Transylvania’

rrch — 366 — 352 350 351 362 — — 367 355 — 364 356 368 — 369

country Greece Italy — Italy Italy Italy Italy Romania Spain Romania Italy Romania Italy Romania Romania Italy Romania

closing date

‘good total’

49 48 49 49 51 51 48 48 51 48 49 48 48 49 49 49 47

47 215 426 415 81 712 383 485 523 88 33 139 54 35 125 52 36

Table 5.17: Hoards closing 51–47 bc used in CA discussed in section 5.4.14. † Listed as Sustinenza 1901 by Backendorf (1998, p. 122, 446–9).

issues, RRC 341 and 348 (90 & 87 bc) seem especially important for the second axis. The positive end indicates quite strongly that many of the newest issues are associated with the Thessalonica hoard (ths)39 but other issues also contribute quite strongly especially RRC 408.40 Thus the second axis is contrasting the newest coins along with some other large issues such as RRC 408 against a selection of Social War period issues.

5.4.14

Hoards closing 51–47 bc

The data set contained seventeen hoards closing 51–47 bc (Table 5.17). They contained 3,849 denarii. Issues from 211–158 bc formed 1.3% of the data set and were omitted from the CAs. Figure 5.60 presents the cumulative percentage graphs of these data; Figures 5.61–5.62 present the maps from the CA. The cumulative frequency graph (Fig. 5.60a) shows that the Beneventum hoard (ben) is extremely archaic in structure and that two hoards, Athens and Padova (ath & p06) are extremely modern although careful examination of the upper ends of the curve show that they cross some of the other hoards. The first axis of inertia (Fig. 5.62) shows this gradient with the most modern hoards at the negative end of the axis and the most archaic at the positive end, and the map of issues (Fig. 5.61) naturally reflects this. One exception to this distribution is RRC 452 which was struck in 48 bc but occurs with the older issues. There are only three examples of this issue in the data set and curiously they all occur in hoards with an archaic profile.41 The Beneventum hoard is unusually archaic for an Italian hoard and one questions the quality of the data. Four of the Romanian hoards are closely grouped on the map reflecting their 39

27.4% of ths consists of RRC 420–434. ths: 4 coins, 7.8%; an1: 1 coin, 2.4%; gra: 3 coins, 1.2%; sus: 2 coins, 3.2%; com: 5 coins, 0.5%. 41 One each in Beneventum, Cuceu and Locusteni (ben, cuc & loc). 40

112

5.4.14. Hoards closing 51–47 bc

100 90 80 70 cumulative percentage

ben 60 50

ath

40

p06

30 20 10 0 157

147

137

127

117

107

97

87

77

67

57

47

year BC

(a) All hoards: solid lines: Italy; dotted line: Spain; dashed lines: Romania; dot-dashed line: Greece. 100 90 80 cas

70 cumulative percentage

ben

sat

grj

cr1

60 50 40

p06

30 20 10 0 157

147

137

127

117

107

97

87

77

67

57

year BC

(b) Selected hoards: lines as marked.

Figure 5.60: Cumulative percentage graphs of hoards hoards closing 51–47 bc.

113

47

+4.0

5. Comparing hoards — Correspondence Analysis

269

53 BC 51 BC 50 BC 49 BC 48 BC 47 BC all other issues

309 448

442

201

443

455

404 230

449 222 329

248 295

259

359

-2.0

450

255 253 296 398 351 374 314 262 348 408 412 313 236 207 256 437 432 205 200 254 217 321 284 221 239 413 317245 263 378 444 244 345 204 243 287 250 428 440 360 364 231 280 267 242 439 337 241 233 274 318 208 281 275 379 203 286 291 434 409 390 261 206 238 265 305 273 416 366 219 257 249 420 197 425 415 279 363 308 344 285 422 199 216 435 312 322 414 237 210 298 215 293 405 303 278 228 406 430 341 270 218 283 276 271 325 240 220 292407 234 356 370 224 386 235 388 266 304 258 391

324 232 247

252

214 223

260 452 323

+5.0

307

396, 438

377

-1.5

411 198 294

Figure 5.61: Variable map from CA of hoards closing 51–47 bc. First (horizontal) and second axes of inertia.

similar archaic profiles (cuc, loc, ods & sat). The ‘Transylvania’ hoard (trn) is plotted away from this group on the second axis partly due to the only example of RRC 455, the newest coin issue in the data set, coming from that hoard. Again, it is notable that this coin comes from an archaic profiled Romanian hoard. At first sight the second axis of inertia appears to be representing the slight differences in closing date between the hoards: the three hoards closing in 51 bc (bro, cas & grj) are all closely grouped near the origin of the map. There is, however, no patterning in the distribution of hoards closing in 49 or 48 bc. A comparison using Procrustes Analysis between this analysis by coin issues and one by years of issue (CA not presented, see Figure 5.63 for the comparison) showed some large differences between the two. This is because the second axis of inertia in this analysis splits up the issues of 49 and 48 bc: RRC 440, 444, 450 and 452 are towards the bottom of the map and RRC 442, 443, 448, 449 and 455 are towards the top. With the start of the Civil Wars we start seeing patterns in the coinage pool created by the various fractions minting coins to pay their supporters. For example, RRC 443 was struck by a mint moving with Caesar and RRC 444 by a mint which was moving with his rival Pompey the Great. RRC 440 and 444 were struck by the moneyer Q. Sicinius, and both these issues are in a similar position on the map.

114

+4.0

5.4.14. Hoards closing 51–47 bc

Italy Spain Romania Greece unknown

tr2 cr1 trn ben

ath

bhr mig

roa

loc

-2.0

bra cas

bro grj

+5.0

ods cuc

-1.5

p06

sat

Figure 5.62: Object map from CA of hoards closing 51–47 bc. First (horizontal) and second axes of inertia.

The Taranto and Carbonara hoards (tr2 & cr1) are not only close on the CA map, but also geographically close in the south of Italy, whereas the other Italian hoards — Padova, Brandosa, Casaleone and Mignano (p06, bra, cas & mig) are all to the north of Rome. Thus the CA, particularly the second axis, is picking up subtle variation in the Italian coinage pool. The position of the Athens hoard (ath) is explained by it containing an example of RRC 442 (with which it closes) but overall having a modern profile closest to Padova (p06). The Roata de Jos hoard (roa) has the most modern profile of the Romanian hoards in this group, close in profile to some of the Italian hoards. To summarise, the first axis is showing the overall pattern of archaic v. modern profiles in the data set whereas the second axis is beginning to tease out differences in the newest coins in the data set. These differences appear to be related to who was responsible for striking those issues and the events of the Civil War.

115

5. Comparing hoards — Correspondence Analysis

5.4.15

Hoards closing in 46 bc

The data set contained twenty hoards closing in 46 bc (Table 5.18). They contained 2,488 denarii. Issues from 211–158 bc formed 1.9% of the data set and were omitted from the CAs. Figures 5.64a–5.64b are the cumulative percentage graphs of these data; Figures 5.65–5.66 are the maps from the CA. This data set is an expanded version of that used as a test case in section 5.2.5. In those analyses it was found that the hoard from Érd (erd) was very unusual in structure (see page 48) and I have therefore also omitted that hoard from this analysis. A comparison of the CA maps for this analysis (Figs. 5.65–5.66) with that for the test data set (Figs. 5.11–5.12) shows a high level of similarity indicating that we have strong structure in our data and thus very stable maps. The first axis in the maps shows the usual archaic to modern profile with the archaic hoards at the negative end of the axis and the modern hoards at the positive end (cf. Fig. 5.64 with Fig. 5.66). Coins of 46 bc form 21% and 28% of the two Iberian peninsula hoards (el2 & sen) but only 0.2% of the Gulgancy hoard (gul). The Italian hoards split into two groups with four hoards having very similar profiles and two being more modern in profile (mor, sur, cro, spn cf. ecl & pli; Figure 5.64a) The Romanian hoards as usual are archaic but they also split into two groups, four with profiles closer to Italian hoards and two more archaic (ti2, rmv, ili, spr cf. vas & sin; Fig. 5.64b). The second axis splits up the issues from the last few years in the data set (Fig 5.65). There are 36 examples of the two Spanish issues, RRC 468 and 469, all but two of which occur in Iberian peninsula hoards, principally El Centenillo and Sendinho da Senhora (el2 & sen). Of the three African issues, RRC 459 and 461 are still very rare with three of four examples in the Iberian hoards, and one example of RRC 459 in the Sînvăsii (sin) hoard. This contrasts with RRC 458 which was also minted in Africa but is much more associated with Italian hoards: 18 of the 22 examples occur in Italian hoards compared to 3 in the Iberian hoards. This pattern can be explained by looking at not only where the coins were minted, but by whom. RRC 459 and 461 were minted for Metellus Scipio who, with Cato the Younger, was defeated by Caesar at the Battle of Thapsus (modern Tunisia) in 46 bc, whereas RRC 458 was struck for Caesar. Caesar returned to Rome in July but some of his opponents fled, eventually, to Spain. Caesar arrived in Spain in early December. Both sides in the conflict issued coins: Pompey the Younger struck RRC 469 and Caesar struck RRC 468. Pompey the Younger and Sextus Pompeius were then defeated by Caesar at the Battle of Munda in 45 bc. Although it is tempting to try and suggest that El Centenillo and Sendinho da Senhora were hoards buried by opponents of Caesar and Fuente de Cantos and Jaén (fdc & jae) were buried his supporters, the data are insufficient to support the idea. RRC 443, an earlier issue which was also struck for Caesar, is pretty evenly split between Iberian and Italian hoards, although 13 of the 15 Iberian examples come from El Centenillo and Sendinho da Senhora. Issues RRC 427 and 453 are fairly evenly split between Iberia and Italy. The majority of the issues struck in Rome are strongly associated with Italian hoards, and the hoard from the former Yugoslavia, Dračevica (dra). The odd position of the Jaén hoard (Fig. 5.66) is due to it having a generally archaic profile but also closing with the Spanish issues 461, 468 and 469.

116

1000

5.4.15. Hoards closing in 46 bc

500

●

● ● ●

0

●

●

●

●

● ●● ●

● ●

●

−500

●

●

−1000

−500

0

500

1000

1500

Figure 5.63: Procrustes Analysis comparing object maps from CA of hoards closing 51–47 bc analysed by years of issue v. issues.

code

name

cro dra ecl el2 erd fdc gul ili iss jae mor pli rmv sen sin spn spr sur ti2 vas

Crotone Dračevica Mirabella Eclano El Centenillo Érd Fuente de Cantos Gulgancy Ilieni Puy D’Issolu Jaén Morrovalle Policoro Râmnicu Vâlcea Sendinho da Senhora Sînvăsii Spoiano Sprîncenata (Viespeşti) Surbo Tîrnava Văşad

rrch 383 379 — 385 373 — 377 — — 386 380 — — 388 — — — 381 — —

country Italy Fmr Yugoslavia Italy Spain Hungary Spain Bulgaria Romania France Spain Italy Italy Romania Portugal Romania Italy Romania Italy Romania Romania

closing date

‘good total’

46 46 46 46 46 46 46 46 46 46 46 46 46 46 46 46 46 46 46 46

86 109 85 57 51 387 459 108 39 65 125 42 43 76 43 264 110 138 148 53

Table 5.18: Hoards closing in 46 bc used in CA discussed in section 5.4.15.

117

5. Comparing hoards — Correspondence Analysis

100 90 80

cumulative percentage

70 60 dra

pli

50 40 30 20 10 0 157

147

137

127

117

107

97

87

77

67

57

47

year BC

(a) Thick solid line: Spoiano (spn); thin solid lines: Italy; dashed line: former Yugoslavia (dra). 100 90 80

cumulative percentage

70 60 50 gul

vas

jae

el2

sen

40 30 20 10 0 157

147

137

127

117

107

97

87

77

67

57

47

year BC

(b) Thick solid line: Spoiano (spn); thin solid lines: Spain and Portugal; dotted line: France; dashed lines: Romania and Bulgaria.

Figure 5.64: Cumulative percentage graphs of hoards closing in 46 bc.

118

Rome Spain Africa moving with Pompey moving with Caesar eastern mint uncertain before issue 425

+4.0

5.4.16. Hoards closing 45–43 bc

469

468

461

459

443 427

453

431 464

434

458 465 426 442

-2.5

+2.5 450

449

448 430

428 433 432 440

-1.5

454 429

463 452

444 467 455 425, 445

Figure 5.65: Variable map from CA of hoards closing in 46 bc. First and second axes of inertia, data points are RRC issue numbers.

To summarise, the first axis of the CA has shown the usual archaic to modern profile, but the second axis has revealed patterns in the issues relating to the Civil Wars.

5.4.16

Hoards closing 45–43 bc

The data set contained twelve hoards closing 45–43 bc (Table 5.19). They contained 4,487 denarii. Issues from 211–158 bc formed 0.6% of the data set and were omitted from the CAs. Figure 5.67 is the cumulative percentage graph of these data; Figures 5.68–5.69 present the maps from the CA by issues. Although the hoards in this data set close between 45 and 43 bc, coins of these years are not especially common and form only 1.1% of the data set whereas issues dating to 49–46 bc form 28.8% and therefore have a major influence on the hoard profiles. In the cumulative percentage graph (Fig. 5.67) the two hoards from Padova (p03 & p07) stand out clearly as being dramatically modern in 119

+4.0

5. Comparing hoards — Correspondence Analysis

Italy Spain Portugal Romania Bulgaria former Yugoslavia France

el2 sen

jae

gul pli vas

sin

-2.5

fdc rmv iss ti2 spr

sur cro spn

dra

-1.5

mor ili

+2.5

ecl

Figure 5.66: Sample map from CA of hoards closing in 46 bc. First and second axes of inertia, data points are hoards.

code

name

cat fir jeg osl p03 p07 pas pia pot sbs thr vll

Cataluña Firenze Jegălia Ossolaro Padova Padova ‘Pasquariello’ Piatra Roşie Potenza San Bartolomeo in Sassoforte ‘Thrace’ Villette

rrch — 399 — 390 — 391 398 — 400 401 402 393

country Spain Italy Romania Italy Italy Italy Italy Romania Italy Italy Greece France

closing date

‘good total’

44 43 43 45 43 45 43 43 43 43 43 45

89 149 453 1512 42 655 105 268 404 426 54 340

Table 5.19: Hoards closing 45–43 bc used in CA discussed in section 5.4.16.

120

5.4.16. Hoards closing 45–43 bc 100 90 80

cumulative percentage

70

jeg

pia

osl

60 50 40

cat

30

p03

p07

20 10 0 157

147

137

127

117

107

97

87

77

67

57

47

year BC

Figure 5.67: Cumulative percentage graph of hoards closing 45–43 bc. Solid lines: Italy; dashed lines: Romania; dotted lines: Greece, France and Spain.

their profile: issues from 46 bc account for over 47% in both cases. Figure 5.67 also shows the two Romanian hoards with very archaic profiles, a group of Italian hoards with very similar profiles and the remaining three hoards with profiles similar to the bulk of Italian hoards if a little more modern in structure. Given the strong structure shown in the cumulative frequency curves it is unsurprising that the first axis of inertia (Fig. 5.69) is representing the usual archaic to modern profile with the archaic Romanian hoards Jegălia and Piatra Roşie (jeg & pia) at the negative end of the axis and the two Padova hoards (p03 & p07) at the positive end. Examining the diagnostic statistics shows that the hoards from Potenza and San Bartolomeo in Sassoforte (pot & sbs) are very poorly represented in these maps. Checking the closing dates of the hoards showed no patterning. The variable map (Fig. 5.68) reveals some more interesting patterning. Issues RRC 435, 483, 488, and 504 at the extremes of the second axis are all unique issues are are of little interest.42 Of the remaining labelled issues we can see that there is a tendency for Italian issues to be in the top right quadrant of the map and the non-Italian issues to be in the bottom right quadrant indicating that the two Padova hoards tend to be more associated with issues struck outside of Italy. Of the four African issues, Padova (p07) is the only hoard to have all four issues.43 As we only have one hoard from Spain in this data set, it is difficult to confidently assess any association between these issues and the Spanish ones as noted previously. The three Spanish issues in this data set (RRC 468–470) are widely spread on the map and there seems little patterning although the Cataluña (cat) does have the highest percentage (3.4%) of Spanish issues. RRC 42

RRC 435 and 483 are from the ‘Pasquariello’ hoard (pas); RRC 488 is from the Piatra Roşie hoard (pia); RRC 504 is from the Padova hoard (p03). Rerunning the analysis without these issues made little appreciable difference. 43 RRC 462 did not occur in the previous group of hoards (see section 5.4.15).

121

+2.5

5. Comparing hoards — Correspondence Analysis

435, 483

Rome Spain Africa moving with Pompey moving with Caesar eastern mint uncertain moving with Brutus Gallia before issue 435

485 455 470

473 445 480

451 474

439 444 487

441 449 442 452 448 472

450

437

454 453 443

468

-1.5

438

486

469

+2.0

446

458 463

459 465

461

447

440

467 464

462

488

-3.0

504

Figure 5.68: Variable map from CA of hoards closing 45–43 bc. First and second axes of inertia, data points are RRC issue numbers.

443 has an interesting distribution with the Cataluña and Villette hoards (cat & vll) having the highest percentages (18% and 11%) closely followed by Padova (p07) but all other hoards having less than 5%. This issue of Caesar’s is often more associated with hoards away from Rome/central Italy. Issues RRC 463–5 struck in 46 bc are all large issues but only RRC 464 occurs strongly in both Padova hoards; curiously RRC 463 and 465 are absent from the smaller but later Padova hoard (p03). RRC 467 is also moderately strongly associated with the two Padova hoards. One curious facet of the CA is the poor ‘quality’ of the representation of the Potenza and San Bartolomeo in Sassoforte hoards (pot & sbs) as they do not appear unusual in Figure 5.67. The San Bartolomeo in Sassoforte hoard is better represented on the third axis of inertia, but the Potenza hoard remains poor. It is likely that a series of very detailed minor differences in the issues represented is the cause of this. To summarise, the first axis shows the usual archaic-modern gradient, but the combination of the first and second axis pulls out some differences between

122

+2.5

5.4.17. Hoards closing in 42 bc

Italy Spain Romania Greece France

pas

thr osl

cat

fir

pot

-1.5

sbs

+2.0

vll

p07

pia

p03

jeg

-3.0

Figure 5.69: Sample map from CA of hoards closing 45–43 bc. First and second axes of inertia, data points are hoards.

hoards which are partly related to where the coins are minted and by whom. The historical question posed by these analyses is ‘what happened at Padova?’ which resulted in the coinage pool there being dominated by recent issues often struck outside of Rome.

5.4.17

Hoards closing in 42 bc

The data set contained eighteen hoards closing in 42 bc (Table 5.20). They contained 6,940 denarii. Issues from 211–158 bc formed 0.4% of the data set and were omitted from the CAs. Figure 5.70 is the cumulative percentage graph of these data; Figures 5.71–5.73 present the maps from the CA by issues. This data set mainly consists of hoards from Italy and Romania. The cumulative percentage graph (Fig. 5.70) shows that these hoards largely fall into two groups: archaic hoards from Romania and hoards with more modern profiles from Italy. The exceptions to this are the Piedimonte d’Alife hoard (pie) which has an archaic profile similar to Romanian hoards and the Bran Poartă hoard which has more Civil War period coinage than its Romanian fellows. The sample 123

5. Comparing hoards — Correspondence Analysis

code

name

alv ana bor bpt chi fa1 fa2 hag isl lis men nag nb2 pie pqu pre rsn tni

Alvignano Santa Anna Borzano Bran Poartă Civitella in Val di Chiana Fărcaşele I Fărcaşele II Haggen Islaz Lissac Menoita Nagykágya (Cadea) Nicolae Bălcescu II Piedimonte d’Alife Pieve-Quinta Prejmer Rossano Terni

country

rrch 417 407 418 408 419 420 — 405 — 409 414 411 — 406 421 412 — 415

closing date

‘good total’

42 42 42 42 42 42 42 42 42 42 42 42 42 42 42 42 42 42

2335 1711 582 59 246 81 113 61 124 52 97 131 43 190 834 150 96 35

Italy Italy Italy Romania Italy Romania Romania Switzerland Romania France Portugal Romania Romania Italy Italy Romania Italy Italy

Table 5.20: Hoards closing in 42 bc used in CA discussed in section 5.4.17.

100 90 fa2 80

cumulative percentage

70 60 50 40

alv

30 20

tni

10 0 157

147

137

127

117

107

97

87

77

67

57

47

year BC

Figure 5.70: Cumulative percentage graph of hoards closing in 42 bc. Solid lines: Italy; dashed lines: Romania; dotted lines: France, Portugal and Switzerland.

124

506

+3.5

5.4.17. Hoards closing in 42 bc

49 BC 48 BC 47 BC 46 BC 45 BC 44 BC 43 BC 42 BC all other issues

497 508

460

483

265, 400 404, 446, 489, 485, 501, 511

487 486 440 472

-2.0

468 454 480

494

474 463 442 449 448 445 473 462 443 450 447 444 453 464455 452 458 465 457 467 459

+3.5 469 488

500

441

-2.0

451

461

Figure 5.71: Variable map from CA of hoards closing in 42 bc. First and second axes of inertia, data points are RRC issue numbers.

map (Fig. 5.73) reflects this division, principally on the first axis of inertia. The first axis also reflects how modern in profile the Italian hoards are. The diagnostic statistics show that hoard from Lissac (lis) is very poorly represented on this map. The variable map is strongly dominated by a number of rare or unique issues and I have therefore cropped the map somewhat to show the patterning in the other issues.44 Two versions of the variable map are presented, one which uses symbols to reflect the year of issue,45 one to show the locations of minting. The variable map (Fig. 5.71) shows issues dating to 43–42 bc widely spread on the map: RRC 497 and 508 at the top right reflecting their unique occurrence in the Fărcaşele II and Islaz hoards (fa2 & isl), RRC 506 at the top right showing that the only example comes from the Terni hoard (tni).46 The block 44

RRC 497, 506 and 508 lie off the map as indicated by the arrows (Figs. 5.71–5.72). Re-runs of this analysis omitting various unique examples in this data set, improved the scaling of the map but did not alter its basic configuration apart from Terni (tni; see footnote 46). 45 N.B. as always, where an issue spans a range of years the beginning of that span has been used, i.e., field date_from in the chrr database. 46 This is an excellent example of the disproportionate affect a rare coin in small hoard has

125

506

+3.5

5. Comparing hoards — Correspondence Analysis

Rome Sicily Spain Africa moving with Pompey moving with Caesar moving with Octavian moving with Sex. Pompeius moving with Cassius and/or Brutus Gallia eastern mint uncertain before issue 440

497 508

460

483

265, 400 404, 446, 485, 489, 501, 511

487 486 440 472

-2.0

468 454 480

494

474 463 442 449 448 445 473 462 443 450 447 444 453 464455 452 458 465 457 467 459

+3.5 469 488

500

441

-2.0

451

461

Figure 5.72: Variable map from CA of hoards closing in 42 bc. First and second axes of inertia, data points are RRC issue numbers.

of issues on the left of the map which includes RRC 501 and 511 all come from the Alvignano hoard (alv) and is in part a reflection of the large size of that hoard. Of the issues struck in 43–42 bc the three most common in the data set are RRC 486, 487 and 494, of which the largest is RRC 494. Its position on the map is partly due to being unusually common in the Lissac (lis) hoard although it occurs in all but the Fărcaşele II hoard. The rarer issues of this period are widely scattered around the map with a number coming from Romanian hoards and four issues only occurring in Alvignano (alv) which is the largest hoard. Of the issues discussed in the previous few sections, RRC 443 is now more widely distributed around the Italian hoards although it is fascinating to note that the highest percentage of this issue comes from the French hoard, Lissac (lis). The African issues RRC 458–462 are also now more widely spread around the Italian hoards, although all four examples of RRC 461 occur in the Santa Anna hoard (ana). Of the Spanish issues, RRC 468 has 36 examples in the on the results of a CA. Omitting all unique examples from the data set results in Terni being plotted much closer to Alvignano (alv).

126

+3.5

5.4.18. Hoards closing 41–40 bc

Italy Portugal Romania Switzerland France

tni

fa2

isl

nb2 hag

fa1

men

pre

pie

alv

bpt lis

nag

-2.0

+3.5 chi bor ana pqu

-2.0

rsn

Figure 5.73: Sample map from CA of hoards closing in 42 bc. First and second axes of inertia, data points are hoards.

data set of which 23 come from Alvignano hoard (1%; alv) but the highest percentage, unsurprisingly, is from the Menoita hoard (men: 5 examples, 5.1%). To summarise, this data set shows less variation between hoards than some of the other groups from the period of the Civil Wars, but the archaic–modern profile persists with Romanian hoards being typically archaic in profile. The newest issues are generally quite rare in this group and are widely scattered around the hoards, perhaps reflecting the wide distribution of places of minting. Some of the associations between specific issues and regions noted in previous sections persist into this data set.

5.4.18

Hoards closing 41–40 bc

The data set contained eleven hoards closing in 41–40 bc (Table 5.21). They contained 1,846 denarii. Issues from 211–158 bc formed 0.5% of the data set and were omitted from the CAs. Figure 5.74 is the cumulative percentage graph of these data; Figures 5.75–5.76 present the maps from the CA by issues. All the hoards in this group close in 41 bc except for ‘West Sicily’ (s03) which closes in the following year. The cumulative frequency graph (Fig. 5.74) shows 127

5. Comparing hoards — Correspondence Analysis

code

name

agn bod cis frn isa s03 sd2 stp tu3 val vis

Agnona Bodrum Cisterna di Latina Francin Işalniţa ‘West Sicily’ † Sadova II Stupini ‘Turkey’ San Pietro Vernotico (Valesio) Vişina

country

rrch 424 — 425 413 428 435 — — — — —

closing date

‘good total’

41 41 41 41 41 40 41 41 41 41 41

272 62 506 43 134 162 30 227 70 201 139

Italy Turkey Italy France Romania Sicily Romania Romania Turkey Italy Romania

Table 5.21: Hoards closing 41–40 bc used in CA discussed in section 5.4.18. † Also included in the next data set (Table 5.22).

100 90 80 isa

agn

cumulative percentage

70 frn 60 50

val

40

s03

30

tu3 bod

20 10 0 157

147

137

127

117

107

97

87

77

67

57

47

year BC

Figure 5.74: Cumulative percentage graph of hoards closing 41–40 bc. Solid lines: Italy; dashed lines: Romania; dotted line: France; dash-dot lines: Turkey.

128

+4.5

5.4.18. Hoards closing 41–40 bc

Rome Sicily Spain Africa moving with Mark Antony moving with Cn. Dom. Ahenobarbus moving with Octavian moving with Sex. Pompeius moving with L. Staius Murcus uncertain moving with Cassius and/or Brutus Gallia before issue 453

386, 483, 506

500 517

501

519

486

473 474 458 468 463 494

496

465 454 489, 507 464 518

-2.5

480

+4.0 459

453 467 469 512 487 472, 497

430, 462, 510, 511, 525

-1.0

455

Figure 5.75: Variable map from CA of hoards closing 41–40 bc. First and second axes of inertia, data points are RRC issue numbers.

very wide variation between hoards with the two hoards from Turkey having very modern profiles (bod & tu3), the Italian and French hoards in the middle and the four Romanian hoards with archaic profiles although the Işalniţa hoard (isa) is very odd in its profile. The original analysis by years was dominated by these gross differences (Lockyear 1996b, section 8.3.17). Although the analysis by issues presented here has the same basic configuration the two Turkish hoards are more widely spread and the Italian and French hoards are more closely grouped than previously (Fig. 5.76). Once more, however, the first axis is mainly showing the archaic–modern gradient. The variable map (Fig. 5.75) has issues from 43–40 bc widely spread on the positive half of the first axis and along the entire length of the second axis. Comparing the map with the original data shows some fascinating patterning. Working our way down the horseshoe curve from the top right we first find three issues — RRC 386, 483 & 506 — with a single example each, all from the Bodrum hoard (bod). Then there are 12 examples of RRC 500, eight from Bodrum and two from the ‘Turkey’ hoard (tu3) and one example of RRC 501 from each of these hoards. Issues RRC 500, 501 and 506 were struck in 43–42 bc by Cassius

129

+4.5

5. Comparing hoards — Correspondence Analysis

Italy Sicily Romania Turkey France

bod

isa

tu3

-2.5

+4.0

val stp agn vis

s03

-1.0

sd2

cis frn

Figure 5.76: Sample map from CA of hoards closing in 41–40 bc. First and second axes of inertia, data points are hoards.

and/or Brutus. Brutus spent part of 43–42 bc in Crete before his defeat at the Battle of Philippi in 42 bc and thus these issues must have been struck in the east. RRC 517 struck by Mark Antony in 41 bc also occurs mainly in the two Turkish hoards (bod: 11; tu3: 4), no examples occur in Italian hoards and one example was found in each of four Romanian hoards (isa, sd2, stp & vis). Crawford notes that this issue was also struck in the east (Crawford 1974, p. 100). Although RRC 519 only has three examples in this data set, two come from the ‘Turkey’ hoard and one from Işalniţa. We have, therefore, a strong correlation between the issues struck in the east and the hoards from Turkey, and to a lesser extent Romania. In contrast there are 33 examples of RRC 458, Caesar’s African issue, in the data set and they are widespread across the hoards apart from those from Romania from which they are entirely absent. The isolated group of issues in the bottom right quadrant of the variable map (Fig. 5.75) are all, as we could predict from the sample map (Fig. 5.76) from the ‘West Sicily’ hoard (s03). There are 23 examples of RRC 511 in the data set and a single example of the remaining four issues. RRC 511 was struck by Sextus Pompeius in Sicily.

130

5.4.19. Hoards closing 40–36 bc

Most of the remaining labelled issues are clustered near the origin of the map in the bottom right quadrant which is a reflection of their occurrence in the hoards with more ‘average profiles’, the older issues are spread out to the left of the map reflecting their association with the archaic hoards from Romania.

5.4.19

Hoards closing 40–36 bc

The data set contained nine hoards closing 40–36 bc (Table 5.22). They contained 8,444 denarii. Issues from 211–158 bc formed 0.7% of the data set. Figure 5.83 is the cumulative percentage graph of these data; Figures 5.78–5.79 are the maps from the CAs. Crawford only dates one small issue of denarii to the period 35–33 bc which explains the gap between this group of hoards and the next. It also means that it is likely that these hoards were collected any time up until the spread of the legionary denarius of Mark Antony (see section 5.4.20). The hoards in this group are generally very large with seven having more than 500 coins. We should expect, therefore, that they provide us with a good representation of the coinage pool. Despite this, the CAs were initially dominated a few issues and several iterations of removal and rerunning the analysis were performed, a method often called ‘peeling’. The first analysis omitted issues prior to RRC 197, as previously, but the result was dominated by RRC 510 of which a single coin was present in the ‘West Sicily’ hoard (s03). The second analysis omitted that coin but was in turn dominated by RRC 511. This issue is much more common with 41 examples in the data set but it is unevenly distributed: eight hoards have 0–0.6% but ‘West Sicily’ still dominates with 14.2%. This issue was struck between 42–40 bc in Sicily by Sextus Pompeius, youngest son of Pompey the Great who controlled the island from 43 bc until the defeat of his navy at the Battle of Naulochus in 36 bc. This fact combined with the early closing date of the hoard explains its high representation in that hoard (see section 5.4.18). The third and fourth analyses were dominated by three more unique issues: RRC 435 in the Arbanats hoard (arb) and RRC 309 and 377 in the Poroschia hoard (prs). The fifth analysis is presented here. Figures 5.78–5.79 are the maps from that CA. As can be seen the first axis is largely dominated by a small number of issues which are associated with the Arbanats hoard (arb). Only one example of issues RRC 404 and 485 are in this data set, both in the Arbanats hoard. There are four examples of RRC 525, three from Arbanats and one from the ‘West Sicily’ (s03) hoard which accounts for the position of that issue on the second axis. RRC 443 is a common issue in this group of hoards with 271 examples in the data set of which 132 examples come from Arbanats, some 14.2% of the hoard compared to 0–5.4% in all the other hoards. The association between Gaul and this common issue of Caesar’s is an interesting observation given that when it was struck in 49–48 bc his power base, in the form of the legions who had fought for him during the Gallic Wars, was still in Gaul. Examining the percentages of that issue in hoards we find that Meolo (me2) and Mornico Losana (los), both from northern Italy and Contigliano (ctg) from central Italy are the next best represented, with Avetrana and Carbonara (ave & cr2), both from the south of Italy, poorly represented. We can, therefore, see a gradient in the representation

131

5. Comparing hoards — Correspondence Analysis

code

name

arb ave cr2 ctg los me2 prs rac s03

Arbanats Avetrana Carbonara Contigliano Mornico Losana Meolo Poroschia Răcătău de Jos II ‘West Sicily’ †

rrch 430 440 443 432 442 437 436 — 435

country

closing date

‘good total’

39 38 36 39 38 39 39 39 40

929 1652 2372 634 1088 1013 541 53 162

France Italy Italy Italy Italy Italy Romania Romania Sicily

Table 5.22: Hoards closing 40–36 bc used in CA discussed in section 5.4.19. † Also included in the previous data set (Table 5.21).

100 90 80

cumulative percentage

70 60 50 40 30 20 10 0 157

147

137

127

117

107

97

87

77

67

57

47

year BC

Figure 5.77: Cumulative percentage graph of hoards closing 40–36 bc.

132

37

+4.0

5.4.19. Hoards closing 40–36 bc

Rome Italy Spain Africa moving with Mark Antony moving with Cn. Dom. Ahenobarbus moving with Octavian moving with Sex. Pompeius moving with C. Antonius uncertain moving with Cassius and/or Brutus Gallia before issue 453

379 404, 485 408

317

472

443 495

528 488

-2.0 223, 265, 457 501

460 507 517

533 534 226, 269, 314, 399, 447, 506 484, 513, 514, 540

350A 453 502 529

455

468

461 490 489 469

525

+4.0

459 496 505 494 473 518 474 486 464 497 463, 465, 523 480 467 454 516 458 462 487 512 526

-2.0

483 519 500

Figure 5.78: Variable map from fifth CA of hoards closing 40–36 bc. First (horizontal) and second axes of inertia.

of this issue from France through northern to southern Italy. This is shown in Figure 5.80 which plots these percentages as proportional circles on the CA map. RRC 468 has 70 examples in the data set of which 31 come from Arbanats forming 3.3% of that hoard, it forms between 0–1.9% of the other hoards. Again, this issue was struck by Caesar although this time in Spain in 46–45 bc and we can see a general north-south gradient in the percentages (Fig. 5.81). We can compare this pattern with that for RRC 494 which has 291 examples in the data set. This issue was minted in Rome in 42 bc and its distribution in Italy is very even but with lower percentages for hoards from France and Romania as one might expect (see Fig. 5.82). The second axis of inertia appears to be showing the general trend from archaic to modern profiled hoards with, as usual, the Romanian hoards showing as archaic at the top of the axis and ‘West Sicily’ showing as the most modern. Tying down specific issues responsible for this trend is difficult and is likely to be the cumulative effect of many issues as shown by the spread of issues on this axis.

133

+4.0

5. Comparing hoards — Correspondence Analysis

Italy Sicily Romania France

prs

rac

arb

-2.0

ave

los

cr2

+4.0

me2 ctg

-2.0

s03

Figure 5.79: Object map from fifth CA of hoards closing 40–36 bc. First (horizontal) and second axes of inertia.

From the above discussion we can see that the CA has shown us not only crude archaic v. modern pattern in hoard profiles, but also more subtle differences and trends in the distribution of coin issues. It would appear at this date that coins struck and issued in Italy quickly spread throughout the Italian coinage pool, but issues struck outside Italy (or released into the coinage pool outside Italy) took longer to spread to and through Italy and we can see a north-south gradient in the distribution.

5.4.20

Hoards closing 32–29 bc

The data set contained eleven hoards closing in 32 bc (Table 5.23). They contained 1,758 denarii. Issues from 211–158 bc formed 0.4% of the data set. The sparse nature of this data set led to all issues prior to RRC 215 being omitted from the CA. Three hoards, Cerriolo, Gajine and Maleo (crr, gaj & mlo), also contained early imperial quinarii dating to 29–27 bc and therefore actually close slightly later. Figure 5.83 is the cumulative percentage graph of these data; Figures 5.84–5.85 present the maps from CA by issues.

134

No examples of 443 2%

+4.0

5.4.20. Hoards closing 32–29 bc

4% 6% 14%

+4.0

-2.0

-2.0

No examples of 468 1%

+4.0

Figure 5.80: Percentage of RRC 443 in hoards shown as proportional circles, cf. Fig. 5.79.

2% 3%

4%

+4.0

-2.0

-2.0

Figure 5.81: Percentage of RRC 468 in hoards shown as proportional circles, cf. Fig. 5.79.

135

No examples of 494 1%

+4.0

5. Comparing hoards — Correspondence Analysis

2% 3% 4%

+4.0

-2.0

-2.0

Figure 5.82: Percentage of RRC 494 in hoards shown as proportional circles, cf. Fig. 5.79.

The denarii issues in the hoards included in this analysis end with the legionary issue of Mark Antony (RRC 544). This issue was minted in 32–31 bc prior to the Battle of Actium where the combined forces of Cleopatra and Mark Antony were defeated by Octavian (Scullard 1982, pp. 168–171). Each legion had a coin type minted in its honour. It was an extremely large issue estimated by Crawford at 864 dies (Crawford 1974, Table L, pp. 699–71). It was also debased, with only c. 85% silver (Crawford 1974, Table XLV, pp. 570–1) and became very common throughout the Empire and in large areas of barbaricum. The hoards in this data set are generally small. Of the total number of coins, 748 (42.5%) are legionary denarii. However, 604 of those come from the largest hoard, Delos (del). Figure 5.84 shows the variable map with all issues plotted. The first axis is dominated by RRC 544 as was expected. The sample map (Fig. 5.85) has three hoards plotted at the right-hand extreme of the first axis (act, bds & del), all of which can be seen to be extremely ‘modern’ in structure (Fig. 5.83) with RRC 544 forming at least 80% of all three hoards.47 The species map also shows that a number of other issues struck by Mark Antony are associated with these three hoards (RRC 539, 542–544). RRC 496, also struck by Mark Antony, is plotted closer to the main group but this issue was struck some ten years earlier than the others. The second axis has a wide spread of scores for issues. In general, earlier issues are at the positive end of the second axis and issues from the 40s bc are at the negative end of the axis. The second axis therefore contrasts extremely archaic hoards (obi, gur & gaj) against moderately archaic hoards (it6 & 47

act: 32 coins, 80%; del: 604 coins, 93%; bds: 50 coins, 94%.

136

5.4.20. Hoards closing 32–29 bc

code

name

act bds beu crr del gaj gur it6 mlo mog obi

Actium Belmonte del Sannio Mont Beuvray Cerriolo † Delos Gajine † Gura Padinii ‘Italy’ Maleo † Moggio Obislav (Dâmboviţa)

rrch 473 460 471 478 465 479 — — 480 470 —

country

closing date

‘good total’

32 32 32 32 32 32 32 32 32 32 32

40 53 32 37 648 88 232 444 65 69 50

Greece Italy France Italy Greece Fmr Yugoslavia Romania Italy Italy Italy Romania

Table 5.23: Hoards closing in 32 bc used in CA discussed in section 5.4.20. † These hoards also include quinarii of Augustus, RIC 1(2) 276, 29–27 bc, and are included in the next data set (see Table 5.25).

100 90 80 gaj 70 cumulative percentage

it6 obi 60

beu gur

50 40 30 20 10 0 157

147

137

127

117

107

97

87

77

67

57

47

37

year BC

Figure 5.83: Cumulative percentage graph of hoards closing in 32 bc. Solid lines: Italy; dashed lines: Romania; dotted lines: France and former Yugosalvia; dash-dot lines: Greece.

137

327 386 431 252

279 325 324 282 290 273

Rome Sicily moving with Mark Antony moving with Cn. Dom. Ahenobarbus moving with Octavian moving with Cassius and/or Brutus Gallia before issue 485

+3.5

+3.5

5. Comparing hoards — Correspondence Analysis

Italy Romania Greece former Yugoslavia France

obi

277

285 284 302 249 280 300 254 269 317 349260319 289 274 352 350 318 382 244 296 341 525 354 238 353 345281 346 286 255 306 361 348 231 299 308 344335 316 337 291 217 485 433 237 322 486 216 236 384 496 362 215 359 233 224 323 488, 519 511 270 250 -1.5 248 275 421 257 494 449 239 247 518 422 357 474 448 360 410 245 336 428 328 409 366 467 416 468 405 426 232 408 393 425 458 480 443 452 261 442

gur

gaj

mog 544 369

+1.5 542 543 462, 487, 526, 539

bds

mlo

crr

-1.5

act

del

+1.5

beu

it6

Figure 5.84: Variable map from CA of hoards closing in 32–29 bc. First and second axes of inertia, data points are RRC issue numbers.

-2.0

-2.0

500 and 32 earlier issues

Figure 5.85: Sample map from CA of hoards closing 32–29 bc. First and second axes of inertia, data points are hoards.

beu). Most of the more recent issues from Rome are plotted near the origin of the map. Issues 485 and 486 are very rare and the diagnostic statistics show that they are poorly represented on the map; issue 494 is much more common but is still poorly represented. Similarly, the three hoards plotted near the origin of the map (crr, mlo & mog) are poorly represented on this map, but are well represented by the third axis of inertia (not presented). The dominance of the legionary issue in the two Greek hoards, particularly that from Actium itself, is no surprise. The Belmonte del Sannio (bds) hoard has a structure so similar to the Greek hoards it is likely that it was a collection hidden without being added to, or circulating in, the Italian coinage pool. Two out of the three hoards plotted near the origin of the CA map, Cerriolo and Maleo (crr & mlo) also contain Augustan quinarii and thus close a few years after 32 bc when legionary denarii became more common in Italy. This leads one to suspect that the remaining hoard from Moggio may also have been deposited at a similar date. Four of the five most archaic hoards are from outside Italy/Greece 138

5.4.21. Hoards closing 29–28 bc

resulting in lower number of legionary denarii. We can see from this analysis, therefore, the highly variable distribution of these coins in the years immediately after the battle of Actium (cf. Chapter 7).

5.4.21

Hoards closing 29–28 bc

This period marks the end of the Roman Republic and the beginning of the Roman Empire under Octavian who took the title of Augustus in 27 bc. Augustus reformed various aspects of Roman coinage, particularly the aes denominations but maintained the denarius at a good weight and fineness (Burnett 1987). Thus, for the purposes of the analyses here the transition should have little effect on the hoarding pattern. In practice, however, the change from RRC to RIC 1(2) creates several problems. RIC 1(2) provides a much coarser dating scheme than RRC which creates blocks of coinage in the data if we were to analyse hoards by years of issue (Lockyear 1996b, pp. 230–46). Conversely, there is no equivalent to the ‘issues’ from RRC that have been employed up to now, and to analyse the hoards by individual RIC numbers would create overly sparse data sets. I have, therefore, created a set of issues by grouping together all coin types which were minted in the same place and given the same date bracket by Sutherland (Table 5.24). I acknowledge, however, that Sutherland’s catalogue has met with some criticism but in current circumstances this is appears the best solution. The second problem with these later data sets is that they are becoming increasingly sparse due to the long time span over which the denarius has circulated and the large numbers of coin types in circulation. I have, therefore, omitted from the CAs all issues with only one example in the data set as well as all coins before RRC 197. Comparisons were made between the results of including and excluding these issues and the relative positions of the hoards was barely affected although the maps were much better scaled and thus able to show patterning in the main mass of issues. This data set contained seventeen hoards closing in 29–28 bc (Table 5.25). They contained 6,756 denarii. Issues from 211–158 bc formed 0.9% of the data set and were omitted from the data set along with all unique examples leaving 6,673 denarii.48 Figure 5.86 are the cumulative percentage graphs of these data; Figures 5.87–5.88 present the maps from CA by issues. The hoards in this analysis come from all over western and central Europe. There is only one hoard from Romania in this analysis, Şeica Mică (sei). This hoard is of particular importance as it is one of the hoards used by Crawford in his estimates of the size of coin issues (Crawford 1974, Table L). As noted above, three hoards, Cerriolo, Gajine and Maleo (crr, gaj & mlo) end with the legionary denarius but also contained early imperial quinarii dating to 29–27 bc, hence their inclusion here. The cumulative percentage curves (Fig. 5.86b) show that there are not very large differences between hoards unlike the previous two data sets. The Niederlangen hoard (nie) is extremely archaic in structure, and has only one coin later than 64 bc. Crawford (RRCH 452) does not categorise this coin as ‘extraneous’ and given that the hoard comes from ‘free’ Germany, we cannot disregard it as 48 Obviously, trimming the data set in this fashion will change the percentage of a hoard formed by an issue. The change is rarely more than 0.05%, however. In the following all percentages quoted are the percentage of the trimmed data set, not the total of ‘good’ denarii.

139

5. Comparing hoards — Correspondence Analysis

‘issue’

RIC numbers

mint

dates

emr spa sp1 sp2 sp3 sp4 lu1 lu2 lu3 lu4 lu6 lu7 it1 it2 ro1 ro3 ro4 ro7 ro8 pel sam per un2 un3 un4 un5 un6

2a–10 33a–45 51–57 64–94 99–119 126–150b 165a–173a 174 178a–183 187a–193a 199 & 201a 207 & 210 250a–263 264–275a 287–322 338–343 351–367 399–410 412–322 472 & 473 475 515–526 540 541 543a–b 545 547b

Emerita Spain, uncertain mint 1 Spain, uncertain mint 2 (issue Spain, uncertain mint 2 (issue Spain, uncertain mint 2 (issue Spain, uncertain mint 2 (issue Lugdunum (issue i) Lugdunum (issue ii) Lugdunum (issue iii) Lugdunum (issue iv) Lugdunum (issue vi) Lugdunum (issue vii) Italy I Italy II Rome I (issue i) Rome I (issue iii) Rome I (issue iv) Rome I (issue vii) Rome I (issue viii) ?North Peloponnesian mint Samos Pergamum IV uncertain mint (issue ii) uncertain mint (issue iii) uncertain mint (issue iv) uncertain mint (issue v) uncertain mint (issue vi)

25–23 bc 19–18 bc 20–19 bc 19 bc 18 bc 18–17/16 bc 15–13 bc 12 bc 11–10 bc 11–10 bc 8–7 bc 2 bc–ad 4 32–29 bc 29–27 bc 19 bc 17 bc 16 bc 13 bc 19 bc 21 bc c. 21–20 bc 19–18 bc 17 bc after 27 bc before 27 bc 28–27 bc from 27 bc

i) ii) iii) iv)

Table 5.24: Imperial ‘issues’ derived from RIC 1(2) (Sutherland 1984). NB RIC numbers are those where examples of those coin types are present in the CHRR database as of February 2007. code

name

ala all bea cda cds clv crr cve es1 gaj lmp me1 mlo nie sei top vig

Cortijo del Álamo Allein Beauvoisin Castro de Alvarelhos Citânia de Sanfins Calvatone Cerriolo † Cologna Veneta Este Gajine † Lampersberg Meolo Maleo † Niederlangen Şeica Mică Topolovo Vigatto

rrch 464 — 459 — 463 — 478 — 466 479 468 — 480 452 456 457 475

country Spain Italy France Portugal Portugal Italy Italy Italy Italy Fmr Yugoslavia Austria Italy Italy Germany Romania Bulgaria Italy

closing date

‘good total’

29 29 29 29 29 29 32 28 29 32 29 29 32 29 29 29 29

130 180 195 3447 281 326 37 106 67 88 52 510 65 62 348 125 737

Table 5.25: Hoards closing 29–28 bc used in CA discussed in section 5.4.21 † Last denarius issue is RRC 544 but also contains quinarii of 29 bc and thus also occurs in the previous data set (see Table 5.23).

140

5.4.21. Hoards closing 29–28 bc

100 90 80 me1

cumulative percentage

70

crr mlo

60 50 40 30 20 10 0 157

147

137

127

117

107

97

87

77

67

57

47

37

47

37

year BC

(a) Solid lines: Italy; dashed lines: Spain and Portugal. 100 90

nie

80

cumulative percentage

70 60 sei 50

lmp

gaj

40 30 20 10 0 157

147

137

127

117

107

97

87

77

67

57

year BC

(b) Solid lines: Romania, Bulgaria and the former Yugoslavia; dotted lines: France, Germany and Austria; heavy dashed line cda (for comparison with Fig. 5.86a).

Figure 5.86: Cumulative percentage graphs of hoards closing 29–28 bc.

141

+2.0

5. Comparing hoards — Correspondence Analysis

Rome Sicily/Italy Osca (Spain) Cyrenaica? moving with Mark Antony moving with Ahenobarbus moving with Octavian moving with Cassius and/or Brutus Gallia Imperial issues before issue 486 228, 230, 256 267, 292, 326 371, 424, 470 297 477, 501, 532 234 469 468 201 320 313 241 443 462 216 243 304 311 406

-1.5

252 UN4

544

502 542

523 495

540 534

511 IT1

496 538 408 497

536

486 506

529 500 463

490 517

IT2 488

317

287

519

494 539

487

516

285

512

528

+3.0

344 337

284 324 342 429 546

275 345

-2.0

334 199

271

(411)

(278) (217)

(276)

Figure 5.87: Variable map from CA of hoards closing in 29–28 bc. First and second axes of inertia, data points are RRC issue numbers/Imperial issue groups. Issues before RRC 486 but with a relative contribution of over 400 are labelled. Issues off the plot are indicated, within parentheses if they have a relative contribution of less than 400.

extraneous without other cause. The hoard is poorly represented on the sample map (Fig. 5.88).49 The Şeica Mică and Gajine hoards (sei & gaj) are archaic in structure, whereas the Cerriolo and Maleo hoards (crr & mlo) are the most modern in profile. The CA of this data set was initially dominated by old and/or rare issues hence the removal of all unique examples as noted above. Even then, some of the more usual and generally older issues had extreme negative scores on the second axis and I have therefore presented maps which are ‘zoomed in’ to the majority of issues (Fig. 5.87). The positions of the omitted issues are indicated by arrows. It is difficult to assess geographical grouping in the sample map (Fig. 5.88) as we have only a few hoards from each country apart from Italy. The Italian hoards are, however, widely spread on the second axis and show little patterning. The previous analysis of this group of hoards was difficult to 49

Rerunning the analysis without Niederlangen made no appreciable difference to the results.

142

+2.0

5.4.21. Hoards closing 29–28 bc

Italy Spain Portugal Romania Bulgaria Germany Austria former Yugoslavia France

crr

mlo

lmp cve

clv ala cds

cda

bea all

vig

-1.5

+3.0

top nie es1 gaj

sei

-2.0

me1

Figure 5.88: Sample map from CA of hoards closing 29–28 bc. First and second axes of inertia, data points are hoards.

interpret (Lockyear 1996b, section 8.3.20) and an initial visual comparison of the results of an analysis by years v. by issues seemed to be quite different. The Procrustes plot shows, however, that although the direction of the axes has changed quite significantly, the relative positions of the hoards has not altered very much (Fig. 5.89). On the variable map (Fig. 5.88) the majority of the more recent issues are in the top right hand quadrant with the most of the more modern hoards of which four are from Italy. The negative end of the first axis, however is not simply the archaic hoards as previously. This analysis was difficult to interpret from the plot alone and so the diagnostic statistics were consulted. The quality for the Castro de Alvarelhos hoard (cda) is extremely good mainly due to its high relative contribution on the first axis. This hoard is currently the third largest in the chrr database. The size of the hoard is partly the explanation for the large block of issues RRC 228 to 532 at the negative end of the first axis, all of which have more than two examples but are only present in the that hoard. That group includes RRC 532 minted in Spain in 39 bc of which all 9 examples come from 143

5. Comparing hoards — Correspondence Analysis

● ●

●

500

●

●

0

●

●

● ● ● ●

●

−500

●

●

● ● ●

−500

0

500

1000

Figure 5.89: Procrustes Analysis of CAs of hoards closing 29–28 bc. First and second axes of inertia, data points are hoards.

the Castro de Alvarelhos hoard. The link to Spanish issues can also be seen in RRC 468 of which it had the highest percentage, RRC 469 of which it has the second highest percentage and RRC 470 of which it has the only two examples.50 RRC 443 continues to be associated with Iberian peninsula hoards with the three Iberian peninsula hoards having first, third and fourth highest percentages, the second highest percentage comes from the Topolovo hoard (top).51 Five of the seven examples from the un4 group of Imperial issues also come from Castro de Alvarelhos and one from Citânia de Sanfins (cds). One wonders if the ‘unknown mint’ may well be in Iberia. Only three issues have a very high relative contribution to the positive end of the first axis, RRC 463, 494 and 544. Thus the first axis is mainly representing the Castro de Alvarelhos hoard v. hoards with high quantities of those issues such as Cerriolo and Maleo (crr & mlo). Many of the eastern issues are also on the positive end of the first axis and so we are seeing indications of an east–west gradient in issues. The extreme size of the Castro de Alvarelhos does have a major influence on the CA and it seems that it is somewhat unusual in profile. On the second axis of inertia all the issues with a high relative contribution have a negative score.52 Of those 11 issues, all were struck before 55 bc and RRC 337–345 in during the Social War (91–88 bc). Meolo (me1) and Şeica Mică (sei) have the highest percentages of coins for the period 91–80 bc (31% & 40% respectively). The second axis is largely representing the distribution of coins of that period. It is highly unusual for a hoard from Italy to have an extremely archaic profile but the Meolo hoard is quite large with reasonable numbers of 50

RRC 468: 6.3%; RRC 469: 1.2%. cda: 6.3%; top: 5.6%; ala: 5.4%; cds: 3.6%. 52 RRC 199, 271, 275, 284, 324, 334, 337, 342, 344, 345, 429. 51

144

5.4.22. Hoards closing 19–15 bc

later coins making it unlikely that the hoard contains many extraneous coins although there are questions regarding the quality of the data.53 To summarise, the analysis has been difficult to interpret partly due to the large influence of the Castro de Alvarelhos hoard. That hoard does seem to be associated with Spanish issues, however, and eastern issues occur more commonly in some of the Italian hoards, and the Austrian hoard from Lampersberg (lmp). The second axis is partly highlighting the distribution of coins of the Social War period. The patterning in the data set is quite complex resulting in the first two axes only explaining 30.8% of the variation in the data and seven of the hoards having a quality of less than 100 for these maps. We should remember, however, that the maps are summarising variation a 17 × 265 table and so the analysis can still be judged to be successful.

5.4.22

Hoards closing 19–15 bc

The data set contained fourteen hoards closing in 19–15 bc (Table 5.26). They contained 2,434 denarii. Issues from 211–158 bc formed 0.2% of the data set and were omitted from the data set along with all unique examples leaving 2,315 denarii. Figure 5.90 is the cumulative percentage graph of these data; Figures 5.91–5.92 present the maps from CA by issues. This data set is less widely spread, geographically, than the previous one with seven hoards from Romania and Bulgaria, three from Italy and two from the Iberian peninsula. The cumulative percentage curves show that three hoards — Abertura, Bourgueil and Penamacor (abe, bou & pen) — have above average quantities of the most recent coinage, the three Italian hoards appear to have a fair amount of coins of the late 30s bc but less of the newer issues and most of the Romanian hoards are archaic in structure although the Medovo and Plopşor hoards (med & plp) cut across the the bulk of the lines in the early 90s bc showing that they have large number of these early Civil War issues. The first two axis of the CA only explain 23.9% of the variation in the data (Table 5.5), somewhat less than many of the other analyses. The spatial patterning on the sample map (Fig. 5.92) gives us confidence that the results will be of use. The hoard from Şpring has a poor score for quality and is thus poorly represented on this map. Comparing the object map with the variable map (Fig. 5.91) we can immediately see that a group of imperial issues are clustered in the bottom left quadrant along with the three hoards noted above as having above average quantities of the most recent coinage. More importantly, however, the imperial issues include many from Spanish mints54 and two of the three hoards are from Iberia. All three hoards have more than 10% of their contents from those five issues. The bottom right quadrant contains the three Italian hoards (pca, ssr & zar) and Conţeşti from Romania (cnt). The three Italian hoards have the highest percentages of legionary denarii (13.9–22.9%). The first axis is, therefore, mainly 53 Gorini (1974–1975) states that this hoard contained 515 coins. It is possible that some ten coins were lost but this is a insignificant proportion of the total. Of these, 213 are preserved at the Museum in Venice, the remainder were given to the landowner but are known from manuscript notes made by B. Forlatti. Gorini also notes that there appears to be only 2–4 coins of each magistrate apart from coins of 89–88 bc 54 spa, sp2, sp3, sp4 and emr. The first four have a high relative contribution (> 300) to the first axis.

145

5. Comparing hoards — Correspondence Analysis

code

name

rrch

country

1po abe bou brd cnt crn mai med pca pen plp spg ssr zar

Poiana † Abertura † Bourgueil Bordeşti Conţeşti † Cornii de Sus Maillé Medovo Palazzo Canavese Penamacor † Plopşor Şpring † Santo Stefano Roero Zara

— 496 493 — — — 488 490 486 502 — — 485 —

Romania Spain France Romania Romania Romania France Bulgaria Italy Portugal Romania Romania Italy Italy

closing date

‘good total’

15 15 18 16 15 18 19 19 19 15 19 15 19 18

141 38 689 43 141 110 421 150 155 81 52 49 97 271

Table 5.26: Hoards closing 19–15 bc used in CA discussed in section 5.4.22. † Also used in the next data set (see Table 5.27).

100 90

abe

80

bou

cnt

cumulative percentage

70 pen 60 mai 50 40 plp 30 20 10 0 157

147

137

127

117

107

97

87

77

67

57

47

37

27

year BC

Figure 5.90: Cumulative percentage graphs of hoards closing 19–15 bc.

146

17

+3.0

5.4.22. Hoards closing 19–15 bc

Osca, Spain mint moving with Mark Antony mint moving with Octavian Imperial issues issues before 528

543

529 377 200 233

444

266 495

322 417 454

313

306 224

216

538 UN4

437

SAM 246, 439, 485, 532 290 487 270 496 364 291 391 511 497 441 545 323 407 282 404 390 305 245 540 372 RO1 452 303, 351, 399, 419 388 353 296 382 356 285 336 259 279 354 314 445 293 292 401 409289 219 257 348 255 IT1 350 286 278 360 455 342 422 IT2 415 316 236 242 324 201 319 335 423 -2.0 486 208 352 328 542 308 346 210 345 311 341 430 349 300 304 528 462 416 386 405 425 204 469 299 378 474 473 238 302 231 284 448 340 301 357 367 273 392 410 244 525 318 327 317 239 SP2 PER 539 387 248 275 544 SP3 SPA 249 326 325 312 220 234 SP4 277 276 424 280 EMR 235 LU1 421 263 287 500

412 374

440

+3.0

205

256

-2.5

283

Figure 5.91: Variable map from CA of hoards closing 19–15 bc. First and second axes of inertia, data points are RRC issue numbers/Imperial issue groups.

representing the Spanish imperial issues v. legionary denarii RRC 544. The second axis is mainly representing earlier issues, especially those of the Civil Wars such as RRC 443, 444 and 454 all of which have a high relative contribution to the axis, v. the newer issues. The bulk of archaic hoards are from Romania and Bulgaria although the association of Civil War issues but not legionary denarii is a interesting detail. The maps present one curious puzzle. Why is the hoard from Conţeşti plotted with the Italian hoards when it is the most archaic hoard in the data set and has no legionary denarii? The original analysis by years of issue placed the hoard with other archaic hoards as would be expected (Lockyear 1996b, Fig. 8.62a). This hoard closes with a single example of four imperial issues.55 These four issues are very strongly associated with Abertura, Bourgueil and Penamacor hoards which accounts for Conţeşti’s position on the second axis. By looking at 55

One each of lu1, ro1, spa & sp2.

147

+3.0

5. Comparing hoards — Correspondence Analysis

Italy Spain Portugal Romania Bulgaria France

plp

brd crn

med 1po mai spg

-2.0

+3.0 zar bou

pca

abe cnt

pen

-2.5

ssr

Figure 5.92: Sample map from CA of hoards closing in 19–15 bc. First and second axes of inertia, data points are hoards.

issues with a high relative contribution to the first axis we can identify that some issues, almost certainly by pure chance, occur mainly in the Conţeşti hoard and the Italian ones, e.g., RRC 205 occurs only in Conţeşti, Palazzo Canavese and Santo Stefano Roero, and 11/ out of 19 examples of RRC 273 (58%) occur in those four hoards. The position of the Conţeşti does, therefore, reflect patterning in the data, but in this case we can argue that the patterning is largely due to chance variation. To summarise, of the newest issues the most plentiful appear to be those from Spain and, unsurprisingly, occur in hoards from the Iberian peninsula. RRC 544, the legionary issue, is now relatively common in Italy, and more so than the newer imperial issues, but is still quite unusual outside the Empire, e.g., Romania. The Romanian hoards, as well as being generally archaic in structure, are also associated with a number of Civil War issues.

148

5.4.23. Hoards closing 15–11 bc

5.4.23

Hoards closing 15–11 bc

The data set contained ten hoards closing in 15–11 bc (Table 5.27). They contained 1,437 denarii. Issues from 211–158 bc formed 0.6% of the data set and were omitted from the data set along with all unique examples leaving 1,365 denarii. Figure 5.93 is the cumulative percentage graph of these data; Figures 5.94–5.95 present the maps from CA by issues. The data set consists of seven Romanian hoards, one Italian hoard and two from the Iberian peninsula. The cumulative percentage graph (Fig. 5.93) shows the two Iberian peninsula hoards, which were also analysed in the previous section, still having the most modern in profile, and Conţeşti still the most archaic. The remaining Romanian hoards seem to split into two with Ciuperceni and Şpring (ciu & spg) having more coinage of the 40s bc. The first two axes of inertia ‘explain’ a higher percentage of the variation in the data than the CA in the previous previous section but the first and second axes are almost equally important explaining 17.2% and 14.7% of the variation respectively (Table 5.5). The hoard from Şpring continues to be poorly represented on the maps. The sample map (Fig. 5.95) clearly splits the archaic Romanian hoards from the three non-Romanian hoards on the first axis. These three have almost identical scores on that axis. Examining the variable map (Fig. 5.94) and the diagnostic statistics shows that RRC 544 is very important for the negative end of the first axis. This seems contrary to the results of the previous analysis where the first axis split the Iberian peninsula hoards away from this issue. In this data set the one Italian hoard — Gallignano — has very similar quantities of legionary denarii to the Iberian hoards (6.5–13.1%), and we should remember that on the second axis of inertia of the previous analysis the Iberian hoards and this issue had similar scores. Two of the Spanish imperial issues (spa & sp4) have a high relative contribution to this axis, both are relatively rare and have moderate numbers in the Gallignano hoard (4 out of 8 & 2 out of 3 respectively). Issues with a high relative contribution to the positive end of the first axis include many early issues such as RRC 328, 364, 382 and 408 (100–67 bc).56 There are only two examples each of sam and un3 all of which also come from Romanian hoards. The positive end of the second axis is associated with a selection of Spanish imperial issues. The negative end of the axis is again associated with a variety of early issues. Those with a high relative contributions include RRC 340, 352, 361 and 367 (90–82 bc).57 From the sample map and the cumulative percentage curve graphs we can see that Ciuperceni and Conţeşti (ciu & cnt) have high quantities of coins from 90–80 bc. To summarise, the first axis is mainly highlighting RRC 544 v. a wide variety of early issues and the second axis is mainly highlighting the Spanish imperial issues v. coins of 90–80 bc.

5.4.24

Hoards closing 8–2 bc

The data set contained eleven hoards closing in 8–2 bc (Table 5.28). They contained 1,780 denarii. Issues from 211–158 bc formed 0.2% of the data set 56

RRC 328: 10/10 from Romanian hoards; RRC 364: 18/23; RRC 382: 24/26; RRC 408:

7/8. 57

RRC 340: 11/30 with ciu & cnt; RRC 352: 6/13; RRC 361: 9/13; RRC 367: 4/13.

149

5. Comparing hoards — Correspondence Analysis

code

name

1po abe cet ciu cnt gal pen sg1 spg stb

Poiana † Abertura † Cetăţeni Ciuperceni Conţeşti † Gallignano Penamacor † Sfîntu Gheorghe Şpring † Strîmba

country

rrch — 496 — — — 505 502 — — 512

closing date

‘good total’

15 15 13 12 15 13 15 13 15 11

141 38 124 161 141 432 81 61 49 209

Romania Spain Romania Romania Romania Italy Portugal Romania Romania Romania

Table 5.27: Hoards closing 15–11 bc used in CA discussed in section 5.4.23. † Also used in in the previous data set (see Table 5.26).

100 90

abe

80

cnt

cumulative percentage

70 pen 60 50 40 30 20 10 0 157

147

137

127

117

107

97

87

77

67

57

47

37

27

year BC

Figure 5.93: Cumulative percentage graphs of hoards closing 15–11 bc.

150

17

5.4.24. Hoards closing 8–2 bc +2.5

SP3

216 LU1

511 EMR

226 287

SP2

413 321 233, 255, 528

SP4 473

469 284

200 414

390

408

308

324

245, 440, SAM 270 279

291 388 285 281 379 353 401 250 364 382 296

228, 323, UN3

244 328 448 544 363 337 289 350 267 494 486 389 384 319 300 UN4 232 345 274 415 215 344 346 240 249 500 387 313 496 348 405, 460, 474, 533 317 375207 325 452 RO7 357 236 374 -2.0 +2.0 416 SPA 311 286 426 IT2 282 542 378 454 354 406 341 273 199 275 276 231 431 204 278 444 464 298 407 252 237 304 385 340 318 RO1 433299 538 362 383 263, 326, 391, 432, 540, 546 352 412 425 302 343 409 367 280 335 238 392 356 IT1 316 421 277 265 525 239 256 214 361 205 380 394

462

342

465

423 327 234

235

mint moving with Mark Antony mint moving with Octavian Cyrenaica Imperial issues before 528

-3.0

424

220

Figure 5.94: Variable map from CA of hoards closing 15–11 bc. First and second axes of inertia, data points are RRC issue numbers/Imperial issue groups.

and were omitted from the data set along with all unique examples leaving 1,716 denarii. Figure 5.96 is the cumulative percentage graph of these data; Figures 5.97–5.100 present the maps from CA by issues. Geographically, the hoards fall into three groups: Romania and Bulgaria, Italy and Sicily, and Germany and the Netherlands. These groupings are not, however, very strongly reflected in the hoard profiles. The cumulative percentage graph (Fig. 5.96) shows the two Bulgarian hoards — Pravoslav and Stojanovo (pra & sto) — as having extremely different profiles. The Viile hoard (vii) has a very archaic profile as we have come to expect but the remaining Romanian hoards (brz & rdj) are more like the four Italian/Sicilian hoards (aqu, bag, es2 & pis). The Köln (I) hoard has a very modern profile but the Bylandse Waard hoard is more archaic (kl1 & byl). An initial CA of all issues post 156 bc was very dominated by Viile and Stojanovo due to groups of older issues which only occurred in one or other

151

+2.5

5. Comparing hoards — Correspondence Analysis

Italy Spain Portugal Romania

pen

abe

1po

stb gal

cet

-2.0

+2.0 spg sg1 ciu

-3.0

cnt

Figure 5.95: Sample map from CA of hoards closing 15–11 bc. First and second axes of inertia, data points are hoards.

hoard58 and the relatively small size of those two hoards. Removing all issues from the analysis with only a single example, as we have been doing for the last few analyses, reduces the number of variables from 241 to 178 and produces a better result. Even so, the maps presented here are ‘zoomed in’ with the position of three issues indicated by arrows (Fig. 5.97). The first axis of inertia (Fig. 5.98) splits four of the east European hoards into two groups with the extremely archaic Stojanovo and Viile hoards at the positive end and the the moderately archaic Breaza and Pravoslav hoards (brz & pra) at the negative end. The variable map shows a long spread-out arm of issues on the first axis associated with the two very archaic hoards. The negative end of the first axis only has two issues with a relative contribution of over 300, RRC 425 and 517, and these issues account for over 3.1% each of the Breaza 58

For example, RRC 202, 204, 208 etc. in the Viile hoard and RRC 207, 224, 290 etc. in the Stojanovo hoard.

152

5.4.24. Hoards closing 8–2 bc

code

name

aqu bag brz byl es2 kl1 pis pra rdj stj vii

Aquileia Bagheria Breaza Bylandse Waard Este Köln (I) Vico Pisano Pravoslav Răcătău de Jos I Stojanovo Viile

rrch 522 523 — 525 519 — 549 520 — — —

country

closing date

‘good total’

2 2 8 2 8 2 2 8 8 2 2

559 311 131 61 281 33 159 58 67 70 50

Italy Sicily Romania Netherlands Italy Germany Italy Bulgaria Romania Bulgaria Romania

Table 5.28: Hoards closing 8–2 bc used in CA discussed in section 5.4.24.

100 stj 90 80

vii

cumulative percentage

70 kl1

60 pis

50 40 30 20 10

pra

0 157

147

137

127

117

107

97

87

77

67

57

47

37

27

17

year BC

Figure 5.96: Cumulative percentage graphs of hoards closing 8–2 bc.

153

7

5. Comparing hoards — Correspondence Analysis

and Pravoslav hoards. No other hoards have more than 1.6% of these two issues. The three examples of RRC 517 in the Breaza hoard are extremely important, however, as these three are locally made cast copies (Poenaru Bordea & Ştirbu 1971). The Răcătău de Jos I hoard has a very low score for quality on this map and is poorly represented. The first axis is, therefore, mainly indicating very archaic v. moderately archaic hoards. Apart from the poorly represented Köln (I) hoard (kl1), which only has a relative contribution to the second axis of 33, the order on the second axis almost perfectly reflects the percentage of legionary denarii in the hoards as indicated by RRC 544’s relative contribution to that axis of 725.59 A number of other issues contribute significantly as well, however, such as lu1 and it2.60 The lack of legionary denarius in hoards from Romania and Bulgaria is somewhat unexpected given that they were issued in the east, although as we have seen earlier other eastern issues are associated with hoards from the region. The third axis of inertia ‘explains’ 12% of the variation in the data compared to the second axis’ 14.1%, and three hoards which are poorly represented in Figure 5.97 (byl, bag & kl1) have a a better score for quality for the maps of axes 2 v. 3. It is worthwhile, therefore, to examine this axis for this data set. Picking out just those issues with a very high relative contribution to this axis we find that only two issues are associated with the negative end of that axis, spa and RRC 442. Köln (I), Pravoslav and Bagheria have the highest percentages spa although in the case of Köln (I) this is a single coin because the hoard is only 31 coins after trimming. At the positive end of the axis we have a slew of 21 issues with a high relative contribution including RRC 528, 538, 540 and 545. These issues are all generally associated with Bylandse Waard (byl) which only has three imperial issues (one each of it2, lu6 & lu7) and no coins of RRC 544 which is why it did not fit into the pattern for either of the first two axes. It is also noteworthy that the combination of axes 2 and 3 allows geographical groups to form, essentially as a result of the distribution of pre-imperial issues. This data set is quite difficult to sort out clearly. The smaller numbers of hoards per region, the generally small size of the hoards, and the development of more complex patterning within the data makes clear cut patterns less common and subtle differences more significant. As a last observation, issue it1 has a date range of 32–29 bc and so in the cumulative percentage curve graphs and previous analyses by years of issue would be grouped along with RRC 544. The Pravoslav hoard, however, has the highest percentage of these issues (3 coins, 5.4%) but no coins of RRC 544 which illustrates once more that although analysis by issues is more complex to deal with due to the larger number of variables and the sparser data, it can show interesting patterning at a detailed level.

5.5

Summary, conclusions and problems

This section will attempt to stand back from the mass of detail presented in the previous section and to summarise the results. Firstly, a series of regional 59

In order of the second axis: pis: 24.6%, kl1: 0%, es2: 18.2%, aqu: 12.1%, bag: 4.3%, vii, byl, rdj, stj, pra & brz: 0% 60 e.g., lu1: pis: 4, 2.6%; kl1: 1, 3.2%; es2: 7, 2.5%; aqu: 1, 3.1%; bag: 2, 0.7%; vii: 0, 0%; byl: 0, 0%; rdj: 1, 1.6%; stj: 0, 0%; pra: 0, 0%; brz: 2, 1.6%.

154

155

425

444

391

528

IT2

261

LU2 237

546

273 401

RO3

428 517

529

452

461

543 511

300, 301, 389, 516

410

380

390

206, 283, 292

512 366 433 SPA 353 506 274 LU6 372 385 409 356 344 387 281 299 SP3 316 324 357 IT1464 319 SP4 342 285 220 463 329 RO8 383 416 SP2 407 434 382 284 405 394 362 442 458 LU3 289 363 277 343 336 291 496 236 337 318 296 388 RO1 443 348 350 393 379 354 341 314 340 335 SP1 384 352 248 231 413 451 364 540 LU1 EMR PEL 420 374 432 LU7 427 533 361 320, 486 544 UN2

RO4

445

453

500

321, 538

+3.0

328

392

260

308

317

306

311 304, 305

287

330

313

312

+4.5

238, 267

403

302, 386

303

Figure 5.97: Variable map from CA of hoards closing 8–2 bc. First and second axes of inertia, data points are RRC issue numbers/Imperial issue groups.

Imperial issues Republican issues

-2.0

446

526, RO7

423

-1.5

239, 257

5.5. Summary, conclusions and problems

156

-2.0

es2 pis

aqu

byl

rdj

kl1

bag

vii

stj

Figure 5.98: Sample map from CA of hoards closing in 8–2 bc. First and second axes of inertia, data points are hoards.

pra

brz

+3.0 -1.5

Italy Sicily Romania Bulgaria Germany The Netherlands

+4.5

5. Comparing hoards — Correspondence Analysis

+3.5

5.5.1. Regional patterns

Republican issues Imperial issues

287 538 321 274

261 281

545

391

367

291 431 417 236 284 329 542 345 372 277 449 231 341 316 356 206 237 469 440 452 283 407 220 465 361 357 429 346 529 342 464 486 320 285 458 385 543 533 314 363 289 267 344 500 IT2 374 413 337 319 299 -1.5 544 340 238 422 313 410 382 546 380 LU1 354 461 311 366 392 512 EMR, PEL 364 415 304 317 286 390 SP1 442 350 433 353 328 425 RO4 517 352 383 336 348 401 384 389 301 428 260 296 343 287 300516 RO3 405 432 273 453 335 RO1 324 308 442 RO8SP3 427 444 403 LU7 318 409 IT1 248 506 445 446 496 312 386 330 SPA 303 302 420 540

+3.5 239, 257 423 306

526 RO7

-2.0

UN2

Figure 5.99: Variable map from CA of hoards closing 8–2 bc. Second and third axes of inertia, data points are RRC issue numbers/Imperial issue groups.

summaries are presented, followed by a discussion of some numismatic aspects of the data. Finally, some observations on the use of CA in the analysis of hoard data will be made.

5.5.1

Regional patterns

Italy The majority of the hoards analysed come from Italy, including Sicily, Sardinia, Corsica, Elba and San Marino. For much of the period under examination the majority of the coins were minted there, mainly in Rome itself. For the early period, insufficient hoards were available for analysis to permit very detailed conclusions. However, the remarkable homogeneity of hoards of similar closing dates was noted. It is also possible that the earliest denarii remain

157

+3.5

5. Comparing hoards — Correspondence Analysis

Italy Sicily Romania Bulgaria Germany The Netherlands

byl

rdj vii es2 aqu brz

-1.5 pis

+3.5

bag stj

pra

-2.0

kl1

Figure 5.100: Sample map from CA of hoards closing in 8–2 bc. Second and third axes of inertia, data points are hoards.

better represented in Sicilian hoards than mainland Italian hoards despite having 60–90 years for their distribution to homogenise. The pattern for the period 118–97 bc has been explored in some detail in section 5.4.3–5.4.5. The original analysis of hoards from this period suggested that there were differences between Italy and the Iberian peninsula but because they were undertaken using years of issue as variables in the analysis the conclusions drawn at the time were tentative (Lockyear 1996b, sections 8.3.3–8.3.4). A fuller discussion if this pattern will be presented in the discussion of the Spanish and Portuguese evidence below (page 164). All the hoards from 92–87 bc analysed come from Italy or Sicily (section 5.4.6). The hoards generally split into two groups, the first of which is fairly homogenous and relatively archaic and the second of which are very modern in profile but widely varying. This is the result of the massive issues minted during and after the Social War (91–89 bc; Scullard 1982, pp. 63–8). The impact of these coins can be seen clearly in the next period (87–81 bc) where there are 158

5.5.1. Regional patterns

large differences in the coinage pool (section 5.4.7). The newest coins had not been in circulation for long enough for their distribution to become even and this, coupled with the large size of the issues in this period, results in large differences between hoards. A beneficial aspect of this is that these large differences allow us to examine the flow of coinage within the coinage pool. As might be expected, Sardinia and Isola Pantelleria have archaic local coinage pools, i.e., the newest coins had yet to reach there in quantity. More unexpected are the archaic hoards from northern Italy in Gallia Cisalpina. This area had been under Roman control for some time and although initially it had a different monetary history (Crawford 1985, chapter 5), one would have expected it to be integrated into the mainstream Italian economy by this period. Within peninsula Italy and Sicily the hoards are still variable but are relatively homogeneous. There is a large difference between them and the archaic hoards. Unfortunately, we have as yet only three northern Italian hoards from this period and caution is called for. It is, however, an interesting observation and one which would repay further research. The 70s bc are a period of increasing homogeneity between hoards (sections 5.4.8–5.4.11). This is a result of the distribution of the large issues of the Social War and the 80s bc slowly becoming even as the denarii circulate around Italy and Sicily, and the issues of the 70s being relatively small. By the end of the 70s the similarity between Italian hoards is so marked that some hoards, such as Villa Potenza and Ossero (vpt & oss) are almost identical. The CAs cease to be able to identify much meaningful variation between hoards. Perhaps the most notable aspect of this decade is the large numbers of hoards dated to 74 bc. Whereas only four hoards of 30+ denarii date to the Social War (91–89 bc), 23 hoards date to 74 bc using Crawford’s arrangement. Of the four issues dated by Crawford to 74 bc, RRC 394–397, all the hoards bar one have an example of RRC 394 which is relatively large issue. Using the Hersh-Walker dating scheme re-dates these hoards to 73–71 bc, 14 hoards closing with 394 now dated to 73 bc, five hoards to 72 bc and four to 71 bc. Apart from the odd hoard from Érd, RRC 394 only forms at most 4.7% of a coin hoard, and rarely more than 3%. If ‘violence’ is one cause of a peak in hoarding we can note that Spartacus’ revolt started in 73 bc and was crushed in 71 bc by Crassus and M. Lucullus (Scullard 1982, pp. 92–3). Although RRC 394 was a relatively small issue, it was much larger than those that followed. There are only eight coins of RRC 398 bc in all the hoards in the chrr database compared to 433 of RRC 394, there are only 23 coins of RRC 399 forming at most 1.5% but usually