Race, place, and power.

Author:Stephanopoulos, Nicholas O.
Position:IV. Descriptive Representation C. Drivers through Conclusion, with footnotes and appendix, p. 1371-1408
  1. Drivers

    Of course, the above analysis does not consider any of the variables other than minority population share that may have driven the changes over time in descriptive representation. In particular, it does not consider the variables that Gingles prioritized above all others: segregation and polarization. To determine their impact on representation, and whether it shifted as a result of Gingles, I construct a series of models. Black or Hispanic seat share is the dependent variable in all cases, while segregation and polarization (decomposed into minority political cohesion and white crossover voting (243)) are the key independent variables. Consistent with the relevant literature, I also include several more factors that may be linked to the proportion of minority legislators: both black and Hispanic population shares, (244) the average population of a state house district, (245) whether a state is covered by section 5 of the Voting Rights Act, (246) and whether a state uses any multimember districts. (247) As in my earlier models, I include fixed effects for years and states as well. (248) And I run four models in total: for black and Hispanic voters, before and after Gingles. (249)

    Figure 5's first two charts display the pre- and post-Gingles relationships between segregation and descriptive representation, for blacks and Hispanics, holding all other variables at their means for the entire 1972-2012 period. (250) For blacks, segregation did not have a statistically significant connection with the share of black legislators before Gingles, (251) As black populations grew more geographically concentrated, that is, they did not receive materially more or less representation. But after Gingles, the relationship between segregation and black seat share becomes strongly positive and significant. (252) In this era, denser black populations elect more of their preferred candidates, with their seat share benefit relative to the previous period rising in tandem with their level of segregation. This benefit is about 0.5 percentage points when segregation is at 40% (5.3% to 5.6%), and roughly 3 percentage points when segregation is at 70% (6.4% to 9.2%).

    For Hispanics, in contrast, there is not a meaningful connection between segregation and the proportion of Hispanic legislators in either timeframe.253 Both before and after Gingles, more spatially isolated Hispanic populations do not obtain significantly more or less representation. If anything, the pre-Gingles link between these variables was slightly positive, while the post-Gingles tie is slightly negative. In wake of the decision, that is, more clustered Hispanic populations elect somewhat fewer Hispanic candidates than they did beforehand (though the gap between periods is small, 0.1 to 1.7 percentage points, for all segregation levels).

    These results validate the hypothesis that geographic compactness did not confer a representational advantage prior to Gingles, (254) In this era, greater spatial concentration did not lead to the election of more black or Hispanic legislators, likely due to the large-scale cracking and packing of minority populations. (255) The results also support the claim that Gingles fundamentally reshaped the relationship between black segregation and representation.256 Greater black density now produces sizeable gains in black seat share, especially compared to the previous period. But the results do not reveal any similar transformation in the link between Hispanic segregation and representation. This link remains as weak in the current timeframe as before the Court's intervention. How come?

    One possibility is that Hispanics are not sufficiently clustered to benefit from the Gingles framework. I found earlier that Hispanic segregation has been substantially lower than black segregation over the last five censuses. (257) Greater residential integration is generally desirable, but it may have the drawback of making Gingles's first prong harder to satisfy and remedial districts harder to design. (258) Another explanation is that, relative to blacks, Hispanics might reap fewer electoral dividends from any given level of segregation. Because a higher proportion of Hispanics are ineligible to vote, (259) and a lower share of eligible Hispanics actually go to the polls, (260) a similar spatial distribution could yield worse political outcomes. And still another option is that Hispanic voters may continue to be the victims of widespread vote dilution. The lion's share of litigation under section 2 has involved black plaintiffs, (261) so it is possible that many more districts could be drawn in which Hispanic voters would be able to elect their preferred candidates. Additional research is necessary to assess these divergent reasons for the absent connection between Hispanic segregation and representation.

    Turning to Gingles's second and third prongs, Figure 5's next two charts show the pre- and post-Gingles relationships between polarization and descriptive representation, again for blacks and Hispanics and holding all other variables at their overall means. (262) For blacks, polarization was positively linked to the share of black legislators before Gingles, though this finding masks a positive coefficient for black political cohesion (the second prong) but a negative coefficient for white crossover voting (the third one). (263) After Gingles, there is no longer a connection between polarization and descriptive representation, but relative to the previous period, the latter's level is higher for any level of the former. (264) This seat share benefit is about 3.5 percentage points when polarization is at 30% (4.1% to 7.5%), and close to 0.5 percentage points when polarization is at 60% (6.5% to 7.2%).

    For Hispanics, neither Gingles prong is related to the proportion of Hispanic legislators, either before or after the Court's intervention. In both eras, greater Hispanic political cohesion and more extensive white crossover voting do not result in the election of appreciably more or fewer Hispanic officials. (265) However, descriptive representation is higher in the post-Gingles timeframe than in the previous period at all levels of polarization. This seat share boost amounts to about 1 percentage point over the entire polarization range.

    These results confirm both the prediction that Gingles made it easier for polarized groups to elect their preferred candidates and the caveat that polarization may have a more complex relationship with representation than does segregation. (266) On the positive side, both blacks and Hispanics are now represented by larger shares of their preferred candidates, at all polarization levels, than they were before Gingles. This is a notable achievement even if the seat share gains--ranging from 0.5 percentage points for more polarized black voters, to 1 percentage point for all Hispanic voters, to 3.5 percentage points for less polarized black voters--are not necessarily enormous.

    More ambiguously, neither Gingles prong rises to statistical significance in three of the four models. And in the one model in which the prongs register (for black voters prior to Gingles), they point in opposite directions. As noted earlier, these mixed findings may stem from the divergent legal and functional implications of polarization, which muddy any statistical analysis of its impact. (267) They may also stem from the fact that polarization is measured using results from presidential elections, while representation is assessed at the state house level. If it were possible to track polarization using state house election results comparable across states and years, it is conceivable that clearer conclusions would emerge. As Elmendorf and his coauthors have observed, "the issue space of national politics may be quite different from the issue space of [state] politics, leading to divergent patterns of racial polarization." (268)

    Next, recall the hypothesis that desegregation might be making it harder for black voters to satisfy Gingles' s first prong and for remedial districts to be drawn around them. (269) (This claim is inapplicable to Hispanic voters, whose integration has not increased in recent years.) To test the hypothesis, I used the post-Gingles model for black voters to generate predicted seat shares under two scenarios: first, if black-white segregation had stayed at its 1992 level for the next two decades (roughly 55%); and second, given the decline in black-white segregation that actually occurred over this period (from 55% to 45%). I again held all other variables at their means--except for the year, whose varying fixed effect determined the proportion of black legislators in conjunction with the varying extent of segregation.

    Figure 5's fifth chart displays these predictions. If blacks had not integrated from 1992 to 2012, their expected seat share in the average state would have increased from 7% to 8.5%, thanks to a rise over time in the year fixed effect. But because of blacks' integration, their expected seat share actually grew to only 7.5%, or about 1 percentage point less than in the counterfactual scenario. The increase in the year fixed effect was mostly offset by the decline in segregation over this timeframe. Accordingly, desegregation has not yet reduced black descriptive representation, but it has prevented it from growing as quickly as it otherwise would have. Going forward, if the desegregative trend continues, it may start to eat away at the proportion of black legislators, especially if the year fixed effect stops rising.

    Lastly, to ensure the robustness of my results, I vary my estimation strategy in several respects. As in my earlier analysis of the drivers of polarization, I replace spatial with aspatial segregation, voting patterns with ideological preferences, and the state fixed effects with state random effects. (270) My principal findings are mostly unaffected by this variation...

To continue reading