Separating the Wheat from the Chaff: Learning about key drivers of temporal social interaction behavior
Because social interaction behavior between actors in a network is driven by many potentially important endogenous variables (which summarize the past behavior between actors), exogenous variables (such as dyadic or actors’ attributes), and possible interactions between these variables, identifying which of these variables have a crucial role to explain social interaction behavior is a challenging problem in exploratory research. To tackle this problem, regularization techniques are developed, such as the Bayesian lasso, or the horseshoe prior, which allow researchers to separate the wheat from the chaff in large complex models resulting in simpler explanations of social interaction behavior. Alternatively in confirmatory research, network theory may dictate which variables play a more dominant role than other variables, which can be translated to statistical hypotheses with order constraints on the network effects of interest. Testing such hypotheses is done using Bayes factors which are flexible for such tests and known to be statistical consistent. These methods have resulted in a better understanding of information sharing behavior between colleagues in large organizations, and to better understand (nonlinear) interaction processes of new workers.
Karimova et al. (2022a). Separating the Wheat from the Chaff: Bayesian Regularization in Dynamic Social Networks.
Karimova et al. (2022b). Honey, I shrunk the irrelevant effects! Simple and Fast Shrinkage Using Approximate Bayesian Regularization. In preparation.
Vieira Generoso et al. (2022). Bayesian mixed-effect models for independent dynamic social network data.
Mulder et al. (2021). BFpack: Flexible Bayes Factor Testing of Scientific Theories in R. Journal of Statistical Software.