Bridging the Micro-Macro Divide in Empirical Research. Moving Beyond Empirically Calibrated Agent-Based Models to Study Aggregation Processes.

While empirical methodology to examine contextual drivers of individual behavior is well developed (e.g., multilevel regression framework), few approaches exist to empirically study real-world aggregation processes. Building on the ‘multiple membership’ extension of the multilevel regression framework (Goldstein 2011), I develop a conceptually reversed multilevel model that allows examining aggregation processes with regression analysis. By including a weighted sum in the regression equation and endogenizing the weights of this sum, the generalized multiple membership multilevel model (MMMM) allows studying complex nonlinearities of micro-level effects on macro-level outcomes. Using a threshold model of collective behavior as an example, I demonstrate by simulation that the generalized MMMM is able to capture this aggregation process while both micro- and macro-level regressions lead to erroneous conclusions. I also discuss advantages of the generalized MMMM vis-à-vis empirically calibrated agent-based modeling as an alternative approach to the empirical study of micro-macro linkages. Finally, I illustrate how to fit the generalized MMMM for a variety of outcomes (linear, logit, conditional logit, Cox, Weibull) in a user-friendly way using the developed R package ‘rmm’.