A tutorial on generative modeling's potential for advancing research using implicit measures

A central project of social cognition researchers is using attitudes measures to predict behavior. We use the example of predicting behavior with implicit measures of attitudes to highlight benefits of generative models in place of (or alongside) summary scores of behavior on such measures. We propose that the field incorporate generative models because they require researchers to clarify assumptions and consequences of behavioral explanations, thereby leading to better reliability, validity, generalizability, and clarity. All of which should eventuate in improved theory-building. To that end, we show 1) why generative models will be useful for their research questions and 2) how to create and use a generative model. We use the example of the tug-of-war model of attitude associations to show how behavior on the IAT can be modeled, to posit a richer explanation of behavior on the IAT, and to show how model parameters can be used to predict behavior.