Analytical Sociology and Agent-Based Modeling: Is Generative Sufficiency Sufficient?

Abstract

Building mechanisms-based, black box–free explanations is the main goal of analytical sociology. In this article, I offer some reasons to question whether some of the conceptual and methodological developments of the analytical community really serve this goal. Specifically, I argue that grounding our computer modeling practices in the current definition of mechanisms posits a serious risk of defining an ideal-typical research path that neglects the role that the understanding of the generative process must have for a black box–free explanation to be met. I propose some conceptual and methodological alternatives, and I identify some collective challenges that the analytical community should tackle in order not to deviate from its main goal.

Analytical sociology and its ‘syntax’ for explanation

“…Although the unraveling of the generative process does not appear as a step in the seven-step sequence of Manzo’s ideal-typical research circle (2014b: 7–9), his proposal can also be interpreted as prone to recognizing the necessity of this understanding. Several reasons for this can be found in León-Medina (2017). Manzo’s most explicit reference to this task can be found in a paper of 2014 where he refers to the ‘thorny issue’ of understanding the inner workings of the model: By inner workings I mean the numerical details of the way in which agents evolve, their interactions and the feedback effects that take place once the simulation is launched.…”

Understanding Mechanistic Explanation as A Strategy of Analytical Sociology

“…The analytical approach is founded on the premise that proper explanations detail the 'cogs and wheels' through which social outcomes are brought about, and it is driven by a commitment to realism. Empirically false assumptions about human motivation, cognitive processes, access to information, or social relations cannot bear the explanatory burden in a mechanistic explanation no matter how well they predict the outcome to be explained (Leon-Medina, 2017). In other words, an explanation that is mechanistic in nature should be able to provide facts that are consistent with such explanation.…”

Multiobjective Genetic Programming Can Improve the Explanatory Capabilities of Mechanism-Based Models of Social Systems

“…e generative approach commits the modeler to at least a limited form of retroduction-the simulation, as redescription, is scrutinised for its ability to reproduce the concrete phenomenon, in so far as the latter is observable in empirical data. e simulation parameters often have to be manipulated in order to achieve good similarity; in the computational modeling community, this process is known as calibration [5] and is commonly identified as belonging to best practice programs for analytical sociological research [6]. If a simulation can be calibrated successfully, then the redescription it encodes is said to pass the generative sufficiency test-it remains a candidate explanation for the phenomenon [1].…”

“…In a contrasting role socialisation mechanism, individuals gradually adopt and internalize drinking practices that are compatible with the roles they hold [21][22][23]. A difference in drinking disposition is calculated when individuals gain or lose roles (equations (6) and 7). e new disposition to drink (equation 8) is a function of this difference in disposition and a modifier (equation (9)), calculated using the number of days the new role has been held and the speed of socialisation.…”

“…To account for role socialisation, the disposition to drink is gradually reduced the longer an individual holds a role and is gradually increased if an individual loses a role. e full disposition effect to apply is calculated using equations (6) and (7). e proportion of that effect to apply after a particular number of days of socialisation is calculated using the logistic function in equation (9).…”