Disentangling Race and Party: What <i>Callais</i>&nbsp;Asks and What Statistics can Answer

<p>In&nbsp;<i>Louisiana v. Callais</i>, the Supreme Court gave clear instructions to plaintiffs bringing redistricting claims under Section 2 of the Voting Rights Act: any analysis of the relationship between race and vote choice, as generally mandated by the second and third preconditions of&nbsp;<i>Thornburg v. Gingles</i>, must account for, or condition on, party affiliation. On its face, this is a simple, easy-to-implement instruction, mostly requiring that party be included as some sort of control in statistical analyses of vote choice. The command is nonetheless hard to reconcile with modern statistical inference, however, and the Court's instructions have generated significant confusion as to how, or whether, they can best be implemented.</p> <p><br></p> <p>In this Article, we begin reconciling the tension between&nbsp;<i>Callais</i>’s instructions and the task of interpretable statistical analysis of voting data. We do so by prescribing a conceptual framework and set of pragmatic implementation strategies that satisfy the&nbsp;<i>Callais</i>&nbsp;majority's instructions while also meeting the assumptions required for sound statistical inference. Focusing on the updates to the second and third&nbsp;<i>Gingles&nbsp;</i>preconditions, we first consider the Court's instructions in light of existing scholarly paradigms of causal and descriptive inference. Though the&nbsp;<i>Callais</i>&nbsp;majority seems to rely on a causal theory of vote choice, the inquiry the majority mandates is extremely difficult, if not impossible, to pursue within a causal framework. An analyst who interprets&nbsp;<i>Callais</i>&nbsp;as mandating causal interpretation will need to make several unrealistic assumptions to establish any statistically valid causal inferences. Under such unrealistic assumptions, any analysis risks yielding gravely misleading or biased conclusions. Despite the&nbsp;<i>Callais</i>&nbsp;majority's causal framing, we therefore do not recommend this approach.</p> <p><br></p> <p>A more statistically sound approach is for analysts to focus on the simpler, descriptive relationship reflected in the conditional correlation between race and vote choice, provided its non-causal interpretation is made explicit. Such analyses are not without their own assumptions, however. Specifically, analysts of voting rights must operate within acceptable bounds of two key parameters: (1) the correspondence between race and party, and (2) the association between race and ideology (candidate choice) conditional on party affiliation. High values of the former combined with low values of the latter can produce misidentified statistical models and specious inferences, even in an otherwise straightforward descriptive analysis. To translate this strategy into pragmatic advice, we provide an empirical analysis of these two parameters, showing how their values constrain analysts’ ability to draw meaningful inferences from a&nbsp;<i>Callais</i>-compliant voting-rights analysis.</p> <p><br></p> <p>We conclude by providing concrete recommendations for analysts attempting to implement the majority’s instructions in&nbsp;<i>Callais</i>. In doing so, we note the mismatch between legal and statistical frameworks of inference, and we encourage future analysts to consider carefully which interpretations are feasible and justifiable under&nbsp;<i>Callais</i>.</p>