© Copyright JASSS

JASSS logo ------

Epistemological Aspects of Computer Simulation in the Social Sciences (Lecture Notes in Computer Science)

Squazzoni, Flaminio (ed.)
Springer-Verlag: Berlin, 2009
ISBN 9783642011085 (pb)

Order this book

Reviewed by Roy Wilson
Jury Simulation Research, Pittsburgh, USA

Cover of book This volume contains twelve contributions collectively spanning social science, computer science, business, environmental science, information systems, and philosophy. Frank, Squazzoni, and Troitzsch provide an excellent introduction (which I cannot match) and Gilbert and Ahrweiler suggest (in "The Epistemologies of Social Simulation Research") that: (1) there is no "one best way" of "doing" social science simulation; (2) it is important to understand why and how computer simulation depends on context and purpose. Tobias Lorenz (in "Abductive Fallacies with Agent-based Modeling and System Dynamics") notes (quoting Donella Meadows and Jennifer Robinson), that "[d]ifferent modeling paradigms cause their practitioner to define different problems, follow different procedures, and use different criteria to evaluate the results". I heartily recommend this volume (modulo my own context, purpose, and paradigm).

I am a practitioner of computational (sociological) social psychology. My perspective on computational social science is based on journeyman-level training and experience with mathematics, philosophy, computer science, history, and (to me, most importantly) sociology. I carry out agent-based simulation research concerning jury deliberations using a model of social influence in small, task-oriented, groups. Of course, the notion that behavior in jury deliberation can be simulated is hardly new (Hastie, Penrod, and Pennington 1983).

Because in situ observation of jury deliberations is (in the US) essentially impossible (much like the observation of thermonuclear weapon detonation), a critical issue is whether the simulation results have epistemic value. I believe that readers of the volume can benefit by "revisioning" their previous work from the perspectives offered by relevant papers. In the hope of illustrating this point, I sketch an argument for the epistemic value of jury simulations based on the group-decision model described in (Wilson 2007): Better examples may exist, but this is the work with which I am most familiar. The sketch is constructed using one chapter that falls into each of the three thematic categories identified in the Introduction.

Generative Explanation

The cognitive scientist Rosaria Conte (in "From Simulation to Theory (and Backward)") refines Epstein's notion of generative explanation: "Generative explanation requires a theory of the linked chain of events from those causes to effects, otherwise there is no generative explanation but mere reproduction of the effect". The group-decision model (in Wilson 2007) links behavioral "causes" to network "effects" back to behavioral "causes" (but see (Doreian 2001) on the "sticky wicket" of causality in social networks). Conte also argues for "defining social properties and entities as ontological[ly] dependent on lower levels but endowed with autonomous causal powers", a welcome formulation to a social realist (see Archer 1995). To its credit, this volume also provides a contrary view (articulated by Camille Roth in "Reconstruction Failures: Questioning Level Design") of the utility of ontological levels.

Empirical Foundations and Validation of Simulation Models

Although some demographic data can be collected and the initial belief of potential jurors regarding the plaintiff/defendant can be "estimated", the only available outcome variable is the verdict, or lack of one. Hence, the task of predicting an outcome invites a structuralist, "rule-based", approach to model-building. How are the rules to be formulated? As noted by economists Ormerod and Rosewell (in "Validation and Verification of Agent-based Models in the Social Sciences") "behavioural rules in many ABMs typically contain certain stochastic elements" - as is the case for the group-decision model. For these authors, "[o]ne key test [of the outcomes] is that the behavioural rules should be capable of justification using evidence from outside the model". The probabilistic transition rules that govern the group-decision model are rooted in the expectation states research program (over fifty years old) that is itself based on both mathematical theory formulation and experimental assessment of that theory.

Emergence

There are many ways to conceptualize emergence. Martin Neumann conjectures (in "Emergence as an Explanatory Principle in Artificial Societies: Reflection on the Bottom-Up Approach to Social Theory") that "... emergence ... should allow us to answer questions such as how cooperative relations among individuals emerge and become stable, or how social institutions, norms, and values evolve". Huneman (2008), in a paper not in this volume, offers the following computational account of emergence: "a state of simulation process is weakly emergent if there is no shorthand to get to it except by running the simulation (this defines the incompressibility criterion of emergence)". In the group-decision model, each social network state generates a (in this case, stable) social structure: the task-related status order that governs participation (and through it, social influence). Each social network state other than the first cannot (I suspect) be determined (except perhaps probabilistically) from the prior network state.

Summary and Conclusion

So, returning to the question of epistemic value, why should anyone take seriously results produced by a jury simulation? First, the simulation enables a generative explanation (in Conte's sense) of the simulated outcomes and, further, is consistent with social realism. Second, the simulation is grounded in a venerable theoretical/experimental research program. Third, the underlying model describes how group social structure "emerges" and becomes stable. Finally, each state (other than the first) of the simulation process satisfies the incompressibility criterion of emergence. In my (biased) view, these are reasons to take the simulation results "seriously". Of course, no reader (or writer!) will be convinced by this mere sketch which may, however, suggest another way readers can benefit from this volume: by "revisioning" their work (and that of others) from the perspectives offered by the contributors to the volume.


* References

ARCHER MS (1995) Realist Social Theory: The Morphogenetic Approach. Cambridge: Cambridge University Press

DOREIAN P (2001) 'Causality in Social Network Analysis'. Sociological Methods and Research, vol. 30, pp. 81-114

HASTIE R, PENROD A and PENNINGTON N (1983) Inside the Jury. Cambridge: Harvard University Press

HUNEMAN P (2008) 'Emergence Made Ontological? Computational versus Combinatorial Approaches'. Philosophy of Science, vol. 75, pp. 595-607

WILSON RW (2007) 'Simulating the Effect of Social Influence on Decision-Making in Small, Task-Oriented, Groups'. Journal of Artificial Societies and Social Simulation, vol. 10, no. 4: http://jasss.soc.surrey.ac.uk/10/4/1.html

-------

ButtonReturn to Contents of this issue