What race and gender stand for: using Markov blankets to identify constitutive and mediating relationships

A growing body of research points to the limitations of conceptualizing and measuring race and gender using a single, time-invariant categorical variable. Researchers have argued that the complex processes underlying race and gender cannot meaningfully be reduced into these categories, and that these measures tend to generate essentialist misconceptions. Yet even if more nuanced measures of race and gender have been developed, most datasets in social sciences still contain single categorical variables to measure these constructs. In this paper, I argue that one way of empirically investigating the meaning and role that these variables play in a specific system is by identifying their Markov blanket, which is composed of the variables carrying all information about the variable of interest. I illustrate this exploratory approach by searching for the Markov blanket of race and gender in a nationally representative dataset containing a wide range of factors related to child development.