The Salience of Race to State Policy: Measurement and Implications for Research and Practice

Background & Purpose

Devolution of policymaking authority from the federal to the state level has many appealing features, including more targeted policy design, more locally democratic responsiveness, and opportunities for policy experimentation. In policy areas such as social welfare and criminal justice, devolution also raises concerns about the influence of the politics of race and racism on policy design and implementation. Research often finds a relationship between the salience of race to state policymaking in these areas and the selection of more stringent policy tools.   

Social theory offers related but differing views on the processes underlying these patterns. While almost all acknowledge a role for both the demographic composition of the target population and attitudes of the racial majority, some theories place greater emphasis on one or the other. Empirically, there is no clear way to measure the salience of race to policy design and implementation. Studies have used various demographic (e.g., racial composition of the state population) and attitudinal (e.g., proportion of whites expressing negative views of blacks) measures. Does the choice of measure matter, or are these various methods interchangeable?  

Methods

I identify or generate several measures of the salience of race to state policy and politics, including demographic measures and measures of the prevalence of negative racial views among whites. These variables include items drawn from different survey datasets and operationalizing different racial attitudes (e.g., “old fashioned racism,” attitudes toward integration). Some attitudinal measures are generated from cross-tabulation by state while others are created from multi-level regression with post-stratification, a model-based procedure.

I examine the simple correlations between the measures to gauge their association with one another. I then use each variable as a predictor in models of various aspects of state welfare policy from 2001 to 2010, including cash assistance coverage under TANF, maximum cash assistance benefit for a family of three, and presence of a TANF family cap. Continuous dependent variables are modeled using linear regression and discrete ones using Firth logistic regression. The models control for other relevant factors such as state government ideology and economic conditions.

Results

With the exception of one of the attitudinal measures, all measures are strongly correlated. When used as predictors in empirical models, the variables also produce consistent results with regard to coefficient sign. Patterns of statistical significance are inconsistent, however. Different variables produce statistically significant results for different outcomes or in different sets of years.  

Conclusions and Implications

The convergence of measures suggests an analyst can be fairly confident in uncovering similar patterns regardless of approach used to operationalize racial salience. The findings also raise concerns, however. First, it is challenging to disentangle underlying processes given the correlation of the measures, creating difficulties both in testing theory and in practically determining change strategies for policy-oriented social workers. Second, the inconsistencies in statistical significance despite consistency in sign highlight limitations in the concept of “significance” and lingering challenges in inference.