A reader of Seltzer's comment may get the incorrect impression that I interpreted Turnout-Manufacturing Correlation as a proxy for the extent of southern institutions. For example, Seltzer writes, “Fleck improves upon my specification by quantifying the specific political factors that characterized the South. His regressions on voting include Turnout (measured at the district level) and Turnout-Manufacturing Correlation.” The view that Turnout-Manufacturing Correlation “characterized the South” is Seltzer's, not mine. As my article explains, Turnout-Manufacturing Correlation provides a useful proxy for the composition of the electorate, conditional on the number of people who voted. In fact, my article states explicitly that the point of using Turnout-Manufacturing Correlation is “not to reflect … the number of people who voted, but whether those people favoring the FLSA most strongly were also among those most likely to vote” (p. 36). Thus, consistent with a logical interpretation of Turnout-Manufacturing Correlation, the two variables I described in my article as “directly related to the extent of disfranchisement” were Turnout and the number of African Americans (p. 48). Seltzer does not mention why he believes he can use Turnout-Manufacturing Correlation to proxy for southern political institutions.

Two preliminary points of clarification are in order. First, my article and Seltzer's comment consider two votes, and, although Seltzer never mentions the fact, my findings differed greatly between the votes. For the vote to recommit, I concluded that “Neither of the turnout-related variables has a consistent sign across specifications, and neither provides clear evidence of how congressional support depended on turnout” (p. 42). Thus, even if his findings were based on sound methodology, Seltzer's claim that these two variables mattered little on the vote to recommit would merely confirm my conclusions. Second, Seltzer's “formula” for the estimated probability of a yes vote should be the cumulative standard normal distribution evaluated at β′x, but Seltzer's F(−β′x) simplifies to one minus the value of the cumulative standard normal distribution evaluated at β′x. Thus, what Seltzer actually reports is a formula for the probability of a no vote.

That is why my article does not interpret Turnout-Manufacturing Correlation as a proxy for southern institutions. Note that my article's reasoning for interpreting Turnout-Manufacturing Correlation as a proxy for the composition of the electorate does not depend on the effect (or lack of effect) of disfranchisement on Turnout-Manufacturing Correlation: A higher value of Turnout-Manufacturing Correlation proxies for greater concentration of turnout in manufacturing counties, regardless of whether disfranchisement increased Turnout-Manufacturing Correlation, decreased Turnout-Manufacturing Correlation, or left Turnout-Manufacturing Correlation unchanged.

The means of Turnout-Manufacturing Correlation are 0.17 for New England, 0.09 for the South, and −0.15 for the entire sample. New England's mean Turnout, at 0.70, demonstrates the absence of southern-style disfranchisement; the mean is 0.29 for the South (the region with the lowest Turnout) and 0.54 for the entire sample.

One can estimate the magnitude of understatement if one is willing to operate under assumptions that Seltzer makes implicitly in his exercise (that is, that Turnout and Turnout-Manufacturing Correlation fully account for southern political institutions, and that Turnout and Turnout-Manufacturing Correlation each vary because of the degree of disfranchisement and noise). Let δ represent the extent of disfranchisement, such that one cannot directly observe δ, but can observe proxies: Turnout (denoted x₁) and Turnout-Manufacturing Correlation (denoted x₂). Then, with β′x as estimated by my probits, the marginal effect of δ will be:

To index the degree to which the effect of Turnout mis-measures the total effect of δ on β′x, define k such that kβ₁(∂x₁/∂δ)=β₁(∂x₁/∂δ)+β₂(∂x₂/∂δ). Thus, k indicates the degree to which the effect of Turnout understates (when k>1) or overstates (when k<1) the effect of δ on β′x. (Seltzer claims that ∂β′x/∂δ≅0; this requires k≅0.) Even though δ cannot be observed, k can be estimated. Following the logic of instrumental variables estimation, one can regress Turnout-Manufacturing Correlation on Turnout, then use the coefficient on Turnout as an estimate of (∂x₂/∂δ)/(∂x₁/∂δ). For the South, that coefficient is 0.749, which, combined with β₁ and β₂, yields estimates of k ranging from 1.5 to 1.9. This means that an increase in disfranchisement sufficient to reduce Turnout by, say, 0.01, would have an effect on β′x perhaps 50–90 percent larger than indicated by the coefficient on Turnout. In sum, when addressing Seltzer's key empirical question with the data that actually reflect disfranchisement, the results suggest that the coefficients on Turnout should be scaled up by 50–90 percent, a striking reversal of Seltzer's claim that they should be scaled down by 100 percent.

Two additional points further elucidate the nature of the joint effects of Turnout and Turnout-Manufacturing Correlation. First, within the South itself, as within the country as a whole, Turnout and Turnout-Manufacturing Correlation both have large, positive estimated effects on the vote to pass the FLSA. This confirms that my article's conclusions (unlike Seltzer's) hold for the region that actually had southern political institutions. Second, the estimated effects of Turnout are not sensitive to whether I control for Turnout-Manufacturing Correlation, and, similarly, the estimated effects of Turnout-Manufacturing Correlation are not sensitive to whether I control for Turnout. This provides further confirmation that my results for Turnout and Turnout-Manufacturing Correlation do not involve the kind of offsetting relationship that Seltzer imagines.

For evidence, see Poole and Rosenthal, “Spatial Mapping” and Congress. As Poole and Rosenthal show, the end of disfranchisement caused a change in the party balance and ideological positions of southern delegations to Congress and, therefore, caused a change in southern roll-call voting patterns. Thus, in view of the actual effects of removing southern political institutions, it is clear that Seltzer made mistakes by disregarding party and by dismissing ideology as “not the variable of interest” in his exercise. Although Seltzer cites Poole and Rosenthal's Congress as his reference with respect to interpreting the ideology variables, Seltzer's treatment of ideology patently contradicts what Poole and Rosenthal actually write. Indeed, Poole and Rosenthal specifically discuss how ideology's predictive power for minimum-wage voting reflects the effects of disfranchisement on ideology (see, for example, Congress, pp. 140–42).

Note, for example, that given the disproportionately involuntary nature of African American nonvoting relative to white nonvoting (Key, Southern Politics), one should expect southern political institutions to produce a negative sign on the African American variable, even when controlling for Turnout. And, as my article reports, I did indeed find the sign to be negative.

Seltzer raises another issue for future research: measurement error in the wage variable. Note, however, that the form of measurement error mentioned by Seltzer is addressed in the robustness tests discussed in my article. Specifically, I verified the robustness of my results by using data on manufacturing wages per manufacturing worker (p. 34). In calculating the number of manufacturing workers, the Census Bureau weighted their figures by the fraction of the year that workers were employed; therefore, the manufacturing wage variable addresses the form of measurement error mentioned by Seltzer. The manufacturing wage variable also controls for within-state heterogeneity, whereas none of Seltzer's (“Political Economy”) economic variables do. Seltzer's data set codes, for example, districts encompassing logging areas of Northern California with wage, industry, and race data identical to those it assigns to districts in Los Angeles. Because all of Seltzer's economic variables are coded in this manner, the obvious problem with measurement error is in Seltzer's previous work. One final remark is important for avoiding confusion: The wage variable that Seltzer's comment criticizes is the wage variable that Seltzer provided when I requested the data from his “Political Economy” (“state median wage” in Seltzer, “Sen37-3.wk1”). My article (p. 50) cites Seltzer as the source of the variable.