Race as a Causal Variable

• In Event: Methods to Study Diversity

Abstract

Various scholars have grappled with conceptual difficulties in applying the potential outcomes framework to the study of race. In a prominent view, some have argued that exposure designs can identify the causal effect of race (Sen and Wasow 2016). In this design, research subjects are exposed (experimentally or observationally) to members of an out-group. In this way, researchers seek to evaluate group contact or threat hypotheses of racial prejudice.

Nonetheless, conceptual difficulties remain. If one accepts philosophical arguments that race is a social construct and that phenotypical markers (such as skin color, accent, clothing) used to classify individuals into racial categories do not correspond to objective attributes of individuals, then we are left with two elements of the concept of race: the marker and the social content ascribed to the marker. These social factors include the prescribed social or economic role of group members, norms, ideologies, and the political context (plus interactions among all of these). While the phenotypical marker is not socially meaningful in itself, it is the unifying element in racial discrimination. In contrast, the social content associated with a marker can be highly heterogeneous across social contexts (Haslanger 2000).

In many cases, empirical researchers vary an aspect of the social context in order to vary the exposure of research subjects to the presence of people with a racial marker. In doing so, researchers seek to identify the effect of exposure to the marker. However, problems arise for causal identification. The estimate generated from such an empirical model will be a combination of the effect of exposure to the marker and changing contextual social factors. This problem may arise due to information (related to the argument of Bueno de Mesquita and Tyson 2020) but it need not do so. In a simple model, we show formally how this feature of race results in problems for causal inference. We also contrast empirical studies in which this problem arises with studies in which social factors are plausibly held constant and, if so, the estimate generated from exposure to a racial marker is the effect of racial prejudice. We conclude by emphasizing the importance of engaging in well-designed causal research about race and the difficulties in doing so.

Authors

• David Robert Foster, University of California, Berkeley

• Joseph Warren, University of California, Berkeley

• Alan Nigel Yan, University of California, Berkeley