The equation was estimated with logistic regression on ANES data, separately for each election, from 1952 to 2012.
Figure 2 gives the expected difference in turnout between Blacks and Whites for each presidential election year. For each election year taken separately, the logit regression results are used to predict the turnout probability for an average American who happens to be White (White, male, and average in age, education and income). This calculation is then repeated, this time to predict the turnout rate for an average American who happens to be Black (Black, male, and average in age, education and income). The “average” turnout for Whites is then subtracted from the “average” turnout for Blacks, and it is this difference that appears in Figure 2.