Network Approaches to the Study of Corruption Johannes Wachs Supervisor: Janos´ Kertesz´ A Dissertation Submitted in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy in Network Science Central European University Budapest, Hungary CEU eTD Collection 2019 CEU eTD Collection Johannes Wachs: Network Approaches to the Study of Corruption, c 2019 All rights reserved. RESEARCHER DECLARATION I Johannes Wachs certify that I am the author of the work Network Approaches to the Study of Corruption. I certify that this is solely my own original work, other than where I have clearly indicated, in this declaration and in the thesis, the contributions of others. The thesis contains no materials accepted for any other degrees in any other institutions. The copyright of this work rests with its author. Quotation from it is permitted, provided that full acknowledgement is made. This work may not be reproduced without my prior written consent. Statement of inclusion of joint work I confirm that Chapter 3 is based on a paper which was written in collaboration with Taha Yasseri, Balazs´ Lengyel, and Janos´ Kertesz.´ I conceived of the idea to relate social capital with local corruption risk outcomes by comparing data from procurement contracts and the online social network. Dr. Kertesz´ and I conceived the details of the implementation. Dr. Lengyel and I collected data on municipalities. Dr. Yasseri and I developed the methods used. I carried out the analyses of the data. All authors contributed to the writing of the paper on which the chapter is based and gave final approval for publication. Dr. Kertesz´ endorses this statement with his signature below. I confirm that Chapter 4 reuses a plot from a working paper written in collab- oration with Mihaly´ Fazekas on the impact of corruption on public procurement market structure. The plot, which I drafted, serves as a prototypical example of the representation of procurement markets as networks. The remaining con- tents of the paper are independent of the contents of Chapter 4. Dr. Fazekas endorses this statement with his signature below. I confirm that Chapter 5 is based on a working paper which was written in CEU eTD Collection collaboration with Janos´ Kertesz.´ I conceived of the idea to use a co-bidding network framework to study collusion. Dr. Kertesz´ and I collaborated on de- veloping and improving the methods used in the paper. I collected the datasets used and implemented the methods. Dr. Kertesz´ and I and both contributed to 3 CEU eTD Collection ABSTRACT Though corruption is a broad notion encompassing many kinds of behavior, it always has a relational aspect. Consider how a driver bribes a policeman, how a minister steers a contract to build a hospital to his son-in-law’s construction company, how two managers from different firms agree to avoid competition in a region, or how a regulator goes easy on a potential future employer during an audit. The observation that interactions between people, firms, and institutions are where corruption happens is not a new one, but certainly merits further investigation. A better understanding of the relationship between the networks that these connections form and corruption can explain why corruption is so difficult to defeat. This thesis applies the methods of network science to the study of corruption and its relationship with markets and society. I argue that corruption emerges from specific patterns of interactions that can productively be described using networks. The dyads of actors engaging in a corrupt behavior, the driver and policeman, minister and son-in-law, etc., are embedded in networks of social relations that facilitate corruption. Within this framework, the thesis addresses several questions about corruption. Why does corruption persist in certain com- munities? How does corruption relate to the organization of markets? How does corruption emerge when it depends on cooperation in highly adverse cir- cumstances? I address these questions empirically using newly available micro- level data on corruption risks in public procurement. Starting with a study of Hungarian towns, I relate corruption risk in lo- cal government contracts to the structure of their social networks. I find that fragmented towns have higher corruption risk, while towns with residents that have diverse connections have less. This suggests that corruption is embedded in the social networks of places. Next I zoom out to the national level, com- paring the procurement markets, conceptualized as networks of issuers and CEU eTD Collection winners, of different EU countries. I find a strong relationship between cen- tralization and corruption risk. On the other hand, heterogeneity in market responses to changes in government across the EU suggests that corruption can be organized in many different ways. Finally, I investigate cartels, or groups of 5 firms that illegally agree to avoid competition. By drawing networks of firms that bid for the same contracts I highlight niches in markets where cartels are more likely to thrive. CEU eTD Collection ACKNOWLEDGEMENTS I would like to acknowledge several individuals who have, whether they real- ize it or not, significantly influenced this work. First I must thank my advisor Janos´ Kertesz´ for his continued support and confidence. His guidance has been invaluable and his patience is much appreciated. I thank my most immediate colleagues David Deritei, Orsi Vas´ arhelyi,´ and Tamer Khraisha for their commiseration and support throughout the PhD. They and the rest of the Department of Network and Data Science at CEU, including Olga Peredi, the other PhD students, and the various postdocs and memorable visitors, provided a wonderful environment to learn and work through turbu- lent social and political times. I am also grateful to my colleagues at the Government Transparency Insti- tute, Misi Fazekas and Agi´ Czibik, my host at the Oxford Internet Institute, Taha Yasseri, and friends and collaborators including Balint´ Daroczy,´ Agnes´ Horvat,´ Dorottya Szalay, Anna May, Casey Tompkins, Theresa Gessler, Gergo˝ Toth,´ and Balazs´ Lengyel. I thank Ancsa Hannak´ for suggesting that a PhD can be a worthwhile thing to do, and then encouraging me during its most difficult moments. I am lucky to have two families supporting me. I thank both the Czem´ an´ and Wachs clans for their unconditional love and support. My mother Kristin and father Karl know that they have given me endless encouragement and I hope they know how thankful I am for it. What they may not know is how important the example they set has been to me. I hope this work reflects that fact. Most importantly I want to thank Zsofi.´ She alone knows what went into this thesis, and her tireless encouragement always kept me on the tracks. She is my inspiration and partner in all things. I dedicate this thesis to her. CEU eTD Collection CEU eTD Collection CONTENTS Contents i List of Tables iii List of Figures vi 1 Introduction 1 2 Related Work 7 2.1 Experiments . 9 2.2 Models . 12 2.3 Measuring Corruption . 12 2.3.1 Survey and Perception-based Measures . 13 2.3.2 Administrative Data-based Measures . 14 2.3.3 Comparison of procurement-based indicators with perception-based indicators . 17 2.4 Corruption as Networked Phenomenon . 17 3 Social Networks and Corruption 21 3.1 Prelude . 21 3.2 Empirical Setting and Methods . 25 3.2.1 Public contracting . 25 3.2.2 Measuring social capital . 31 3.2.3 Models . 35 3.3 Results . 37 3.4 Discussion . 37 CEU eTD Collection 4 Corruption and Procurement Markets 41 4.1 Prelude . 41 4.2 Data . 43 4.3 Markets as Bipartite Networks . 45 i ii CONTENTS 4.3.1 The Core of Procurement Markets . 49 4.3.2 The Clustering of Corruption Risk in Markets . 58 4.3.3 Market Turnover and Change in Government . 65 4.4 Discussion . 69 5 Cartels 71 5.1 Prelude . 72 5.2 Results . 74 5.2.1 The 1980s Ohio School Milk Market . 75 5.2.2 Georgian Public Procurement Markets . 76 5.2.3 Simulation Model . 79 5.3 Discussion . 80 5.4 Methods and Data . 82 5.4.1 Co-bidding networks, group detection, and group features 82 5.4.2 Null models . 84 5.4.3 Agent Based Model . 84 5.4.4 Datasets . 86 6 Conclusion 87 7 Appendices 111 7.1 Social Networks and Corruption . 111 7.1.1 Description of iWiW data . 111 7.1.2 Relationship between fragmentation and diversity . 112 7.1.3 Model covariates and controls . 112 7.1.4 Model results, diagnostics, and feature importances . 113 7.2 Cartels . 118 7.2.1 Ohio School Milk Data . 118 7.2.2 Georgian Contracting Data . 119 CEU eTD Collection LIST OF TABLES 3.1 Elementary indicators of corruption risk on public contracts . 27 3.2 Municipality corruption risk regressed on social capital variables. 38 4.1 Procurement Market Network Summary Statistics . 48 4.2 Summary statistics of market cores. 56 4.3 Average National Procurement Market Edge-Clustered Modularity 60 4.4 Clustering of Single Bidding by Country . 64 4.5 Changes of Government, EU countries . 66 5.1 Cartel screen comparisons. 79 7.1 Descriptive statistics of key settlement-level variables and controls.114 7.2 Stepwise regressions predicting municipality corruption risk. 115 7.3 VIF scores for model predictors. 116 7.4 Lower inclusion threshold regression robustness test. 117 7.5 Summary statistics of the Ohio school milk market by year. 118 7.6 Georgian market summary statistics . 119 7.7 Georgian cartel screen size inclusion robustness tests. 120 CEU eTD Collection iii iv LIST OF TABLES CEU eTD Collection LIST OF FIGURES 2.1 Single bidding rates of EU countries. 18 2.2 Correlates of national single bidding .