Explore the issue of causal inference in agent-based computational models in a first-of-it’s-kind volume 

Agent-based Models and Causal Inference delivers an insightful investigation into the conditions under which different quantitative methods can legitimately hold to be able to establish causal claims. The book compares agent-based computational methods with randomized experiments, instrumental variables, and various types of causal graphs. It goes on to explain why there is no strong argument to believe that observational and experimental methods are qualitatively superior to simulation-based methods in their capacity to contribute to establishing causal claims. 

Organized in two parts, Agent-based Models and Causal Inference connects the literature from various fields, including causality, social mechanisms, statistical and experimental methods for causal inference, and agent-based computation models to help show that causality means different things within different methods for causal analysis, and that persuasive causal claims can only be built at the intersection of these various methods. 
Readers will also benefit from the inclusion of: 

• A thorough comparison between agent-based computation models to randomized experiments, instrumental variables, and several types of causal graphs. 
• A compelling argument that observational and experimental methods are not qualitatively superior to simulation-based methods in their ability to establish causal claims 
• Practical discussions of how statistical, experimental and computational methods can be combined to produce reliable causal inferences  

Perfect for academic social scientists and scholars in the fields of computational social science, philosophy, statistics, experimental design, and ecology, Agent-based Models and Causal Inference will also earn a place in the libraries of PhD students seeking a one-stop reference on the issue of causal inference in agent-based computational models. 

Gianluca Manzo is a research fellow in sociology at the French National Centre for Scientific Research (CNRS), and an affiliated researcher at the Institute for Analytical Sociology (IAS) at the University of Linköping. He has held positions of official visitor, instructor, or visiting professor at institutions including Nuffield College, Columbia University, the European University Institute (EUI), and the Universities of Oslo, Barcelona, Cologne, and Trento. 

List of Acronyms xi

List of Tables xii

Preface xiii

The Book in a Nutshell xvii

Introduction 1

1 The Book’s Question 3

2 The Book’s Structure 6

Part I: Conceptual and Methodological Clarifications 9

1 The Diversity of Views on Causality and Mechanisms 11

1.1 Causal Inference 11

1.2 Dependence and Production Accounts of Causality 13

1.3 Horizontal and Vertical Accounts of Mechanisms 17

1.3.1 Vertical versus Horizontal View 19

1.3.2 Horizontal versus Vertical View 21

1.4 Causality and Mechanism Accounts, and ABM’s Perception 22

2 Agent-based Models and the Vertical View on Mechanism 25

2.1 ABMs and Object-oriented Programming 26

2.2 ABMs and Heterogeneity 27

2.3 ABMs and Micro-foundations 28

2.4 ABMs and Interdependence 28

2.5 ABMs and Time 29

2.6 ABMs and Multi-level Settings 30

2.7 Variables within Statistical Methods and ABMs 31

3 The Diversity of Agent-based Models 33

3.1 Abstract versus Data-driven ABMs: An Old Opposition 34

3.2 Abstract versus Data-driven ABMs: Recent Trends 36

3.3 Theoretical, Input, and Output Realism 38

3.4 Different Paths to More Realistic ABMs 40

3.4.1 “Theoretically Blind” Data-driven ABMs 41

3.4.2 “Theoretically Informed” Data-driven ABMs 45

Part 2: Data and Arguments in Causal Inference 49

4 Agent-based Models and Causal Inference 51

4.1 ABMs as Inferential Devices 52

4.1.1 The Role of “Theoretical Realism” 52

4.1.2 The Role of “Output Realism” and Empirical Validation 54

4.1.3 The Role of “Input Realism” and Empirical Calibration 55

4.1.4 In Principle Conditions for Causally Relevant ABMs 57

4.1.5 Can Data-driven ABMs Produce Information on Their Own? 58

4.2 In Practice Limitations 59

4.2.1 ABMs’ Granularity and Data Availability 59

4.2.2 ABM’s Granularity and Data Embeddedness 61

4.3 From-Within-the-Method Reliability Tools 62

4.3.1 Sensitivity Analysis 64

4.3.2 Robustness Analysis 65

4.3.3 Dispersion Analysis 65

4.3.4 Model Analysis 66

5 Causal Inference in Experimental and Observational Methods 69

5.1 Causal Inference: Cautionary Tales 71

5.2 In Practice Untestable Assumptions 73

5.2.1 RCTs and Heterogeneity 73

5.2.2 IVs and the “Relevance” Condition 74

5.2.3 DAGs, Causal Discovery Algorithms and Graph Indistinguishability 76

5.3 In Principle Untestable Assumptions 79

5.3.1 RCTs and “Stable Unit Treatment Value Assumption” (SUTVA) 79

5.3.2 IVs and the “Exclusion” Condition 81

5.3.3 DAGs and Strategies for Causal Identification 83

5.3.3.1 DAGs and the “Backdoor” Criterion 83

5.3.3.2 DAGs and the “Front Door” Criterion 84

5.4 Are ABMs, Experimental and Observational Methods Fundamentally Similar? 85

5.4.1 Objection 1: ABM Lacks “Formal” Assumptions 86

5.4.2 Objection 2: ABM Lacks “Materiality” 89

5.4.3 Objection 3: ABMs Lack “Robustness” 91

5.5 A Common Logic: “Abduction” 94

6 Method Diversity and Causal Inference 95

6.1 Causal Pluralism, Causal Exclusivism, and Evidential Pluralism 97

6.2 A Pragmatist Account of Evidence 99

6.3 Evidential Pluralism and “Coherentism” 101

6.4 When is Diverse Evidence Most Relevant? 104

6.5 Examples of Method Synergies 106

6.5.1 Obesity: ABMs and Regression Models 106

6.5.2 Network Properties: ABMs and SIENA Models 109

6.5.3 HIV prevalence: ABMs and RCTs 111

6.5.4 HIV treatments: ABMs and DAG-based identification strategies 113

Coda 115

1 Possible Objections 116

1.1 Causation is Not Constitution 117

1.2 Lack of a Specific Research Strategy 118

1.3 A Limited Methodological Spectrum 119

2 Summary 121

References 127

Index 149