62,99 €
Mehr erfahren.
- Herausgeber: John Wiley & Sons
- Kategorie: Wissenschaft und neue Technologien
- Serie: Wiley Series in Computational and Quantitative Social Science
- Sprache: Englisch
This book is organized in two parts: the first part introduces the reader to all the concepts, tools and references that are required to start conducting research in behavioral computational social science. The methodological reasons for integrating the two approaches are also presented from the individual and separated viewpoints of the two approaches.The second part of the book, presents all the advanced methodological and technical aspects that are relevant for the proposed integration. Several contributions which effectively merge the computational and the behavioral approaches are presented and discussed throughout
Sie lesen das E-Book in den Legimi-Apps auf:
Seitenzahl: 308
Veröffentlichungsjahr: 2015
Ähnliche
Table of Contents
Cover
Title Page
Preface
1 Introduction: Toward behavioral computational social science
1.1 Research strategies in CSS
1.2 Why behavioral CSS
1.3 Organization of the book
Part I: CONCEPTS AND METHODS
2 Explanation in computational social science
2.1 Concepts
2.2 Methods
2.3 Tools
2.4 Critical issues: Uncertainty, model communication
3 Observation and explanation in behavioral sciences
3.1 Concepts
3.2 Observation methods
3.3 Tools
3.4 Critical issues: Induced responses, external validity, and replicability
4 Reasons for integration
4.1 The perspective of agent-based modelers
4.2 The perspective of behavioral social scientists
4.3 The perspective of social sciences in general
Part II: BEHAVIORAL COMPUTATIONAL SOCIAL SCIENCE IN PRACTICE
5 Behavioral agents
5.1 Measurement scales of data
5.2 Model calibration
5.3 Model classification
5.4 Critical issues: Validation, uncertainty modeling
6 Sophisticated agents
6.1 Common features of sophisticated agents
6.2 Cognitive processes
6.3 Cognitive structures
6.4 Critical issues: Calibration, validation, robustness, social interface
7 Social networks and other interaction structures
7.1 Essential elements of SNA
7.2 Models for the generation of social networks
7.3 Other kinds of interaction structures
7.4 Critical issues: Time and behavior
8 An example of application
8.1 The social dilemma
8.2 The original experiment
8.3 Behavioral agents
8.4 Learning agents
8.5 Interaction structures
8.6 Results: Answers to a few research questions
8.7 Conclusions
Appendix Technical guide to the example model
A.1 The interface
A.2 The code
References
Index
End User License Agreement
List of Tables
Chapter 05
Table 5.1 Tools for the calibration of single-level behavioral models with a single decision variable.
Table 5.2 Tools for the calibration of multilevel behavioral models with a single or multiple decision variables.
Chapter 06
Table 6.1 Main simple models of reinforcement learning and their characteristics.
Chapter 07
Table 7.1 An example of sociomatrix.
Table 7.2 Sociomatrix of Table 7.1 raised to power 2.
Table 7.3 Sociomatrix of Table 7.1 raised to power 3.
Table 7.4 Models for the generation of random networks.
Chapter 08
Table 8.1 Mean and standard deviation of the percentage of contribution to the public good and number of free riders in the original experiment (as percentage values).
Table 8.2 Results of the fixed effects regression model.
Table 8.3 Results of the random coefficients regression model.
Table 8.4 Results of the first differences regression model.
Table 8.5 Results of the ordered probit regression model.
Table 8.6 Example of inputs and outputs available at time 5.
Table 8.7 Main GP parameters and values.
List of Illustrations
Chapter 02
Figure 2.1 An abstract model of causal relationships in computational social science.
Chapter 05
Figure 5.1 The interface between agents (behavior) and the social environment.
Chapter 07
Figure 7.1 The example of undirected social network of Table 7.1 presented as a graph.
Figure 7.2 Random networks created with the four different algorithms and associated parameters.
Figure 7.3 Degree distribution of the four random networks presented in Figure 7.2.
Figure 7.4 Von Neumann and Moore neighborhoods in a two-dimensional grid space.
Chapter 08
Figure 8.1 Behavioral rule of player 40 represented as a decision tree.
Figure 8.2 Average contribution to the public good in the original experiment and in simulations with different types of agents.
Figure 8.3 Standard deviation of contributions to the public good in the original experiment and in simulations with different types of agents.
Figure 8.4 Average contribution to the public good in the original experiment and in simulations with different types of agents and random group assignment.
Figure 8.5 Standard deviation of contribution to the public good in the original experiment and in simulations with different types of agents and random group assignment.
Figure 8.6 Average contribution to the public good in simulations with fixed effects behavioral agents, random group assignment, and different levels of return on investments in the private good.
Figure 8.7 Average contribution to the public good in simulations with random coefficients behavioral agents, random group assignment, and different group sizes.
Figure 8.8 Average contribution to the public good in simulations with different types of agents, random group assignment, and 30 periods of interaction.
Figure 8.9 Average contribution to the public good in simulations with fixed effects behavioral agents and reinforcement learning agents, over 30 periods of interaction; comparison of two stable random interaction structures, fully connected groups of five agents and small world networks.
Appendix
Figure A.1 The interface of the simulation.
Figure A.2 Definition of input variables in the interface.
Figure A.3 Definition of graph plots in the interface.
Figure A.4 The window for editing the simulation code.
Guide
Cover
Table of Contents
Begin Reading
Pages
ii
iii
iv
ix
x
1
2
3
4
5
7
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
31
32
33
34
35
36
37
38
39
40
41
43
44
45
46
47
48
49
50
51
52
53
54
55
57
59
60
61
62
63
64
65
66
67
68
69
70
71
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
173
174
175
176
177
178
179
180
181
182
183
184
185
187
188
WILEY SERIES IN COMPUTATIONAL AND QUANTITATIVE SOCIAL SCIENCE
Embracing a spectrum from theoretical foundations to real-world applications, the Wiley Series in Computational and Quantitative Social Science (CQSS) publishes titles ranging from high-level student texts, explanation, and dissemination of technology and good practice, through to interesting and important research that is immediately relevant to social/scientific development or practice.
Other Titles in the Series
Vladimir Batagelj, Patrick Doreian, Anuška Ferligoj,
Nataša Kejžar – Understanding Large Temporal Networks and Spatial Networks: Exploration, Pattern Searching, Visualization and Network Evolution
Riccardo Boero – Behavioral Computational Social Science
Rense Corten – Computational Approaches to Studying the Co-evolution of Networks and Behavior in Social Dilemmas
Danny Dorling – The Visualisation of Spatial Social Structure
Gianluca Manzo (ed.) – Analytical Sociology: Actions and Networks
Brian Straughan, John Bissel, Camila C.S. Caiado, Michael Goldstein and Sarah Curt – Tipping Points: Modelling Social Problems and Health
Behavioral Computational Social Science
Riccardo Boero
Los Alamos National Laboratory, New Mexico, USA
This edition first published 2015© 2015 John Wiley & Sons, Ltd
Registered officeJohn Wiley & Sons, Ltd, The Atrium, Southern Gate, Chichester, West Sussex, PO19 8SQ, United Kingdom
For details of our global editorial offices, for customer services and for information about how to apply for permission to reuse the copyright material in this book please see our website at www.wiley.com.
The right of the author to be identified as the author of this work has been asserted in accordance with the Copyright, Designs and Patents Act 1988.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, except as permitted by the UK Copyright, Designs and Patents Act 1988, without the prior permission of the publisher.
Wiley also publishes its books in a variety of electronic formats. Some content that appears in print may not be available in electronic books.
Designations used by companies to distinguish their products are often claimed as trademarks. All brand names and product names used in this book are trade names, service marks, trademarks or registered trademarks of their respective owners. The publisher is not associated with any product or vendor mentioned in this book.
Limit of Liability/Disclaimer of Warranty: While the publisher and author have used their best efforts in preparing this book, they make no representations or warranties with respect to the accuracy or completeness of the contents of this book and specifically disclaim any implied warranties of merchantability or fitness for a particular purpose. It is sold on the understanding that the publisher is not engaged in rendering professional services and neither the publisher nor the author shall be liable for damages arising herefrom. If professional advice or other expert assistance is required, the services of a competent professional should be sought.
Library of Congress Cataloging-in-Publication data applied for
A catalogue record for this book is available from the British Library.
ISBN: 9781118657300
Cover Image: maikid/iStockphoto
Preface
I developed the ideas and perspectives presented in this book during several years of work and interactions in many different places and with many people. I should probably thank all of them personally and individually and not only for what I present here.
Among the many institutions I should thank, there are the Universities of Torino and Brescia in Italy, the University of Surrey in the United Kingdom, and the Los Alamos National Laboratory in the United States. I have had the chance to “live” in those institutions, and I consider the seminars, the talks, and the discussions I had in those places with both staff and students as probably the most inspiring ones for the development of my research interests and approach.
I also have to admit that the fact of having worked for several years in the field of public policy has helped me greatly. When scientists face real-world problems, they can learn much, both because they have the chance to put their knowledge to use in a real-life situation (and to fear the results) and because they receive unique feedback from policymakers and stakeholders. I want to thank in particular those I met in European local governments who shared with me their own viewpoints, unconsciously supporting my learning.
Among the many individuals to whom I owe a thanks, there are, in a disordered order, Pietro Terna, Flaminio Squazzoni, Nigel Gilbert, Peter Hedström, Paul Ormerod, Laura Bianchini, Brian K. Edwards, Marco Castellani, Giangiacomo Bravo, Gianluigi Ferraris, Matteo Morini, Michele Sonnessa, Donatella Pasqualini, Marco Novarese, Salvatore Rizello, Massimo Egidi, Marco Monti, Rosaria Conte, and finally (and collectively) the students of the PhD program in economics and complexity at the University of Torino. I consider several of them not only fellow researcher but also friends, with some of them I have spoken for entire years, while with others I have had the chance to interact only for short periods and few words. However, each one of them has taught me something special about science in general and social sciences in particular, and for that, I am very grateful.
The points of view presented here are obviously my own only. Because I conceive social research as a never-ending process of innovation and learning of new methods, I hope that this book will contribute to new explorations and ideas.
1Introduction: Toward behavioral computational social science
The main assumption of this book is that individual behavior and social phenomena are somehow connected and that the investigation of that connection is central for all social sciences.
The work presented here can be classified as a methodological one since it deals with methods. With extreme synthesis, it presents the methods available for putting together the studying of individual behavior, as developed in behavioral sciences, with the many tools that today compose the approach called “computational social science” (from now on CSS).
The ideas and methods presented here have originated in different domains, and it is very difficult today to find an exhaustive and comprehensive description of them. The book thus aims not only at theoretically discussing a unified methodological approach but also at providing the readership with all the necessary information to experiment with the approach. Obviously, given its physical constraints, the preparation of this book has meant much selection and not all concepts and tools are explained from scratch and in details. However, average readers with interests in the scientific explanation of social phenomena can surely comprehend what is discussed here and can use the many references provided to master all the topics and tools.
Before presenting how the book is organized and how selective reading can be conducted on it, this introduction focuses on the first question readers should ask: what is the use of this approach?
1.1 Research strategies in CSS
CSS has recently emerged because of many important technological and conceptual advancements.
Undoubtedly, it is the “Big Data” approach and the large availability of data associated with it that allow today studying large-scale social phenomena that were impossible even just a few years ago. At the same time, it is the availability of (cheap) computing power that allows storing, managing, and analyzing those datasets.
From the conceptual viewpoint, relatively new scientific tools such as those of social networks analysis, complexity, and other approaches finally find in such an abundance of social and behavioral data the chance to be applied and tested.
Along with the promises of the potentially fruitful integration of all those innovations, CSS seems today to give social sciences the possibility to overcome the well-known limits of more traditional approaches. Heterogeneity of individuals, nonlinearity of systems and behavior, and the lack of capability to effectively put in relationship social structures and social phenomena with individual behavior are just a few of the many examples of limits that potentially can be overcome.
Much of the research in CSS is today still aimed at exploring the potentials of the approach, but different research strategies have already emerged, and consequently, different methodologies have emerged too.
In particular, CSS can be considered today as a self-standing approach to social sciences because it provides tools and methods to pursue any kind of scientific research strategy.
Research strategies in science are only a few. Following an order that is not intended to imply any ranking of importance, it is firstly possible to explore the data in order to describe or classify it. It is an activity that is always needed in scientific investigations and that can provide first-hand and novel information about completely unknown phenomena and systems.
Second, it is possible to establish relationships in the data. Using statistical and other models, it is in fact possible to observe that some of the variables appear to be connected, changing in similar or opposite ways. Variables that are somehow related are the first candidates to consider, select, and investigate deeper.
Most of the contemporary research works in CSS, for instance, the ones belonging to the Big Data and science of networks approaches, adopt one or both of these two research strategies.
CSS however allows also pursuing the third kind of scientific research strategy, which is the investigation of causality. Similarly to any other scientific domain, in social sciences, the investigation of causality requires the availability not only of data but also of tools for modeling. Modeling is the formalization procedure that ultimately allows developing, testing, and validating knowledge. In CSS, modeling is pursued by the approach of social simulation where adopted modeling tools are explicitly addressed at dealing with the peculiar complexities of social systems.
1.2 Why behavioral CSS
Behavioral CSS is aimed at investigating causality of social phenomena and it thus relies on modeling tools developed in social simulation.
Further, it refers to the need for behavioral information. The need for the integration of the two approaches is motivated by the unique form that causal relations have in social systems.
In fact, social phenomena can be conceived as different from phenomena in natural systems because of their complexity. The complexity of social phenomena, moreover, is surely characterized by the most complex object we know, which is the human brain, but not only by that. Social phenomena, in fact, are complex because their causes always involve both individual behavior and some of the many features of the social structure (e.g., institutions, social norms, ways of social interaction, etc.). These features of social complexity are the ones that the behavioral CSS approach explicitly acknowledges and that it tries to effectively deal with.
Most of the critiques to established research methods in social sciences are addressed at the same time. Causal explanations in traditional approaches are incomplete and often unreliable because the complexity of social systems is reduced, either from the perspective of individuals or from the one of the social structure. Established approaches are often obliged to apply reductionism because of the technical limits of the analytical tools that are adopted.
Today, CSS and behavioral sciences create the opportunity to overcome traditional limits and to finally deal with the individual, the social structure, and the relation between the two.
Behavioral CSS can thus be seen from three different perspectives. From the one of behavioral sciences, it can provide researchers with the tools to extend their investigations toward social contexts of interaction. Such an extension can improve knowledge about behavior because of the many feedbacks going from the social structure to individual behavior and vice versa.
From the perspective of researchers in social simulation, the integration of behavioral knowledge in their models can finally allow rigorous validation and external validity.
From the perspective of social sciences in general, behavioral CSS enriches the researcher’s toolset with an approach that is explicitly addressed at the investigation of causality of social phenomena intended as a social causal mechanisms.
Because of what just said, the behavioral CSS approach presented in this book is aimed at readers in social sciences in general and particularly at those in behavioral sciences and social simulation.
The book discusses the approach and presents several tools in order to allow readers with different backgrounds experimenting with and perhaps extending it.
1.3 Organization of the book
Being aimed at presenting and discussing a methodological approach that puts together different tools and traditions, the book is organized in two parts. In the first one, there is the presentation of the main concepts and methods developed in the two approaches that are integrated (i.e., CSS and behavioral sciences). The first part also includes a short discussion of the advantages of the approach.
The second part takes a more applied and technical perspective and it presents methods for tool integration. In particular, because in order to investigate causality with behavioral CSS researchers have to rely on modeling, that part initially discusses how to integrate results of behavioral analyses in models. Secondly, there is a chapter that discusses how to model structures of social interaction.
In conclusion, the second part ends with the presentation of an example of application of the approach, mainly aimed at didactic purposes. In particular, it is presented the preparation and implementation of a model of a social phenomenon, starting from the collection of behavioral data, passing through its analysis, and arriving at the specification and formalization of behavior and interaction in the model. Some applications of the model are sketched too in order to give the reader the intuition of potential uses and results in terms of causal explanations.
Readers experienced in social simulation and agent-based modeling can probably read the book without spending much attention to Chapter 2, although the concepts related to causality that are presented there are crucial for the evaluation of the approach.
Readers coming from behavioral sciences can avoid Chapter 3 where common tools of that approach are shortly introduced.
Readers particularly doubtful about the analytical advantages provided by the approach should start reading Chapter 4 and then read the section presenting results in Chapter 8. If reading those chapters changes their minds, they can go backward to the rest of the book to better comprehend the reasons why the approach can be analytically effective and powerful.
Readers who intend to adopt the modeling approach presented here but who are not experienced in social simulation should spend particular attention to the second part of the book. The first three chapters of that part illustrate the algorithms that allow modeling behavior in agents and interaction between agents. The fourth chapter of that part provides examples of application of many of those algorithms. Finally, in the appendix, the technical implementation of what presented in Chapter 8
Part ICONCEPTS AND METHODS
2Explanation in computational social science
What is computational social science (CSS) and how researchers working in this field think and work? This chapter aims at shortly introducing the main concepts, methods, and tools underlying the emerging field of CSS.
CSS is an intrinsically interdisciplinary approach that uses several concepts originally developed in different domains. Furthermore, its development takes also advantage of the availability of innovative modeling tools, of the increasing computing power of computers, and of everyday larger and detailed datasets about social phenomena.
In order to understand CSS, it is thus needed to understand the convergence of concepts and methods that constitutes it. Further, while the CSS approach is today made by several different tools applied to different research strategies, for the sake of this work, we focus only on one specific research strategy, that is, the investigation of causality in order to explain social phenomena. Therefore, we select concepts and tools that have influenced the development of CSS with that in mind, and we focus our discussion on how scientific explanations (i.e., investigations of causality) are conceived in CSS.
The first section of this chapter is devoted to the introduction of the main concepts underlying CSS, namely, the ones derived from the fields of complexity, social mechanisms, and social networks. They all contribute to the definition of CSS because they provide CSS with concepts, methodologies, methods, and tools. Furthermore, the section outlines the contributions of two other important elements influencing the development and definition of CSS, which are the increasing availability of data regarding socioeconomic domains and the increasing computational power available to researchers.
The following section then focuses on the methods used in CSS and particularly on the predominant role played by agent-based modeling, network analysis, and social simulation in general.
The fourth section of this chapter is a very short introduction to the general and common characteristics of the software tools available for conducting causality research in CSS, aiming at allowing inexperienced readers to evaluate research needs and tools potentials before tackling steep learning curves.
The chapter concludes with a short discussion of some important practical critical issues related to the CSS approach to explain social phenomena.
2.1 Concepts
CSS derives from the convergence of concepts and tools, and concepts and tools themselves in turn are often the output of other innovations such as the availability of new large datasets and computing power.
The understanding of concepts underlying CSS is thus necessary in order to understand what CSS is and to more easily appreciate contributions in this field. However, it is hard to defend that each single constituent of CSS is a novelty. On the contrary, disciplines composing social sciences have debated and developed the same concepts underlying CSS for decades if not centuries, and they have done that often in autonomous and independent way, in different times, under different labels, and with different impacts. The novelty and the definition of CSS thus do not descend from a single concept, but from the adoption of a broader, consistent, and inclusive perspective.
2.1.1 Causality
The first important contribution to the development of CSS derives from complexity science, that is to say the science that studies complex systems (Eve et al. 1997, Foster 2004).
Complex systems are systems defined as, in general, characterized by the presence of nonlinearity and by being far from equilibrium. Nevertheless, they are not chaotic systems, and they have several properties, like the possibility to generate outcomes difficult to forecast or understand by just looking at the elements of which they are composed.
In complex systems terminology, there is a micro level which is made of system components and a macro level that is made of aggregated and systemic outcomes. A good and classic example is the temperature of materials, a macro feature absent in the micro level of atoms.
The research field coping with complex systems was developed within physical sciences, in particular in the fields of nonlinear and nonequilibrium thermodynamics (Prigogine and Stengers 1979). More recently, it has been extended to social sciences.
The modeling frameworks developed in complexity, for instance, in the field of statistical mechanics, mostly rely on simulation techniques. Nonlinearity in fact often means the absence of closed form analytic solutions, and thus, simulation is the unique way to explore the solution space and to understand model outcomes.
Complexity thus carries on the attention toward nonlinearity, on out-of-equilibrium dynamics rather than on static equilibria, in the usage of simulation as the analytical tool to investigate models of systems that can be out of equilibrium, and finally on the presence of macro, systemic outcomes that “emerge” from micro heterogeneous components. In other words, the approach provides concepts and tools to investigate causality in complex systems.
Regarding the focus of complexity on emergent systemic properties, one of the most known and studied of them is self-organization. Complex adaptive systems (Holland 1975, 2006) are a subcategory of complex systems capable to organize themselves autonomously to cope with external shocks and to adapt to a changing environment. Any kind of living organism is a good example of a complex adaptive system. Most social systems are good examples too.
Another strand of research that has influenced the development of CSS is the one focusing on social causal mechanisms as explanations for social phenomena. This field of research provides CSS with a specific form of explanations (i.e., mechanisms) for social phenomena. Social mechanisms are, in other words, how causal links are organized and have to be searched for in social systems.
Definitions of social mechanisms are many. For instance, Merton, who introduced middle-range theorizing as the middle way between general social theories and descriptions, proposed mechanisms as constituents of middle-range theories. Mechanisms in his words are “social processes having designated consequences for designated parts of the social structure” (Merton 1949, pp. 43–44). The relation between social processes and the social structure helps pointing out also the presence of different levels in social structures, so that mechanisms work between “entities at different levels, for instance between individuals and groups” (Stinchcombe 1991, p. 367).
Mechanisms, however, can be defined also via their explanatory power. For example, Harré (1970) noted that a mechanism needs to have a key role in explaining a social phenomenon, while Elster (1998) reports the evolution of his interpretation of mechanisms, saying, for instance, that in Elster (1983) he conceived mechanisms as antonyms of black boxes, in connection with a reductionist strategy in social science, while in more recent works, he was referring to mechanisms as antonyms of scientific laws. Mechanisms, in summary, have to be white boxes because they are explanations of the causality of a social phenomenon, but they take a specific form that is far from the one of scientific laws.
Combining the ideas of mechanisms as social processes and as explanations of social phenomena, Boudon (1998, p. 172) presents the definitive definition of a social mechanism as the “well articulated set of causes responsible for a given social phenomenon.” The need for mechanisms derives from the characteristics of causality in social phenomena. Other kinds of explanations of social phenomena (e.g., laws) do not allow effectively capturing causality of these systems due to their reductionism.
Further, social mechanisms do not imply a mechanicistic and static vision of social systems, but on the contrary, they suggest one that shows features like self-organization, self-adaptation, and purposive behavior as in complex systems. In fact, as early pointed out for neural cognitive mechanisms in human beings, there are also mechanisms capable to (purposely or not) adapt the operations of other mechanisms to produce different results in the same external conditions (Hayek 1952). Similarly, social mechanisms define social systems where the internal structure can be modified and the range of operation can be extended depending upon experience.
Explanations in the form of social mechanisms are not in the form of a law (either deterministic or probabilistic). Social mechanisms also differ from explanations in the form of statistical relationships because they are made by causal relationships of which statistical correlations are only an observation of some outcomes.
Social mechanisms, that have found a systematic representation in sociology in the approach of analytical sociology (Hedström and Swedberg 1998a), are not new in social sciences, but they support CSS in identifying two main aspects of the scientific explanations looked for in this approach.
The first one is the focus on causality: in CSS, it is essential to investigate the complex set of causes that generate social phenomena. The degree of complexity of causes cannot be reduced and must be effectively coped with. Consequently, CSS mainly exploits modeling tools that allow rigorously investigating the complex set of causes giving rise to social phenomena.
The second consequence of the social mechanism approach for CSS relates the attention on the distinction and relationship between micro and macro.
Social mechanisms consider two sets of abstract elements, actors, and their “staging,” and they are useful because they allow understanding their systemic effects (Hernes 1998). The distinction between “macrostructural and microindividual levels” (Van Den Berg 1998) allows investigating both how social structure influences individual behavior and how individual behavior determines social structure.
The distinction between micro and macro is thus not just a trivial separation between individuals and the aggregation. For instance, social mechanisms allow for the possibility of considering institutions and other kinds of actors as micro entities. Similarly, explanations neither have to be reduced to methodological individualism (i.e., to individual actions) nor to social structures. Social mechanisms neither ignore the relevancy of actors nor of structures, but they consider both as necessary ingredients for understanding social phenomena.
The micro–macro separation is thus dependent on the phenomenon of interest and on the elements of analysis, and it is led by the fact that some mechanisms needed to explain the phenomenon work between two conceptual different levels of social systems, one of which is greater than the other in the sense that it contains the other. For instance, if the phenomenon of interest is a particular dynamics happening in a society, the macro level could be that social group, with all its characteristics, rules, and so forth, while the elements aiding the description of the macro level could be the elements of the micro level, that is to say individuals and their relationships.
In summary, the distinction considered in social mechanisms between the micro and the macro is characterized by the following features:
It is a relative, nonabsolute separation depending on the social phenomenon on focus and on the research process carried on by the scholar studying it.
It considers a micro level constituted by elements (from individuals to nations) who act and interact.
It considers a macro level that represents the system made by the micro elements, which is often conceived as a structure (which can be composed of roles, social norms, etc.) emerged from social actions by micro elements: the only requirement is that the features of interest of such a “macro” system must be observable (i.e., some features of the macro must be measurable).
The micro and the macro are separate for analytical purposes, but for the same reason, they are put in relationship for explaining social phenomena: social mechanisms in fact often point out the effects of the macro on the micro and how the micro generates the macro.
An abstract model can be used to synthesize and categorize how explanations of social phenomena usually work with social mechanisms and in CSS. In particular, it is possible to consider a modified version of the macro–micro–macro model introduced by Coleman (1986), often dubbed as the “Coleman’s boat.” The aim of the model is to represent three different kinds of causal links determining social phenomena. The different kinds of causes are separated here for the sake of simplicity and for improving the description and the modeling of social phenomena (see Section 2.3 when talking about tools used in CSS).
As in Figure 2.1, which schematically presents the abstract model inspired by Coleman, the framework considers micro and macro levels and their reciprocal influences. As the figure shows, social mechanisms can be differentiated depending on the direction of the connection between macro and micro levels. In the figure, numbers represent causal relationships that contribute to explain a social phenomenon, and the direction of the arrow indicates the direction of causality.
Figure 2.1An abstract model of causal relationships in computational social science.
Causal relationships where the macro level influences the micro elements are represented in Figure 2.1 with the arrow labeled “1.” They include dynamics such as “situational mechanisms” from analytical sociology (Hedström and Swedberg 1998b), “second-order emergence” from social simulation (Gilbert 2002), and “immergence” from the fields of cognitive modeling and social simulation (Castelfranchi 1998).
For instance, many belief-formation and preference-formation models have at their core the presence of mechanisms which link the social structure or other macrosocial characteristics to the beliefs, desires, opportunities, and behavior of actors.
In the field of economics, a common example is the influence that macroeconomic variables such as the official interest rate can have on households and firms beliefs and choices.
Another example of this kind of causal mechanism is constituted by the complex set of constraints and opportunities faced by individuals in any kind of social dynamics. In fact, the environment that surrounds individuals, either the social structure or the natural environment, is a macro-level entity that defines opportunities and constraints of actors living there.
Similarly, another example could be the one of some learning processes where the feedbacks do not come from other micro actors but from the macro level. Moreover, a set of social norms can influence beliefs and actions of individuals, in particular when individuals have low levels of reflexivity or when social identity is strong.
The second arrow in Figure 2.1, the one labeled “2,” describes causes entirely located at the micro level and causes that affect the macro level by means of interaction and aggregation at the micro level.
They include dynamics happening at the micro level that do not involve interdependence with the macro level and also those that do not strictly require social interaction, as, for instance, utilitarian decision-making. But they also include any other kind of action and choice at the micro level if not influenced by the macro one.
Thus, many of those causes are based upon interaction among micro elements, including processes such as imitation and communication. Social interaction, moreover, often defines how micro elements come to define the macro level or to modify it.
Along the arrow labeled “2” takes place all the processes of social interaction that in CSS often relies on the concepts developed in social network analysis (SNA).
Starting from some seminal works of social psychologists and sociologists (e.g., Milgram 1977), SNA is a branch of sociometrics that has developed tools explicitly aimed at observing, measuring, and statistically investigating the structure of individuals’ interaction. In recent times, SNA has experienced a renewed attention due to the availability of vast datasets about social interactions on social media and by scholars in the field of complexity.
