Competitiveness Analysis for Adopted and Alternative Congressional District Plans in Arizona Joseph Stewart, Jr., Ph.D. Professor, Political Science University of New Mexico January 31, 2003 I have been asked to analyze the competitiveness of the Arizona Congressional districts used in the 2002 general elections and of an alternative the Downtown Competitive (DTC) plan. I earlier (April 18, 2002) submitted an analysis of this issue, but now an election has provided new data for analysis. I again use the Mayhew 1 standard, i.e. a competitive district is one in which the winner receives less than 55% of the vote. In conducting the analyses that follow, I have relied largely upon data provided by Tony Sissons, Research Advisory Services, Phoenix. Specifically, Sissons provided to me a precinct-level file which contains demographics, vote totals, Congressional district assignment used in the 2002 general election, and proposed assignment to Congressional districts under the DTC plan. The original file did not include precinct-level election returns for either Mohave or Yavapai County. These data have subsequently been added. Data for Mohave County was again provided by Mr. Sissons. The only data which I did not receive from Mr. Sissons are the data for Yavapai County, which were collected from the County web site, with links to each precinct (http://www.co.yavapai.az.us/elections/2002/1105/results/byprecinctindex.asp). All of these data are collected in the attached Excel file: AZCONG02_DTC_MASTER.xls. While the earlier competitiveness analysis was a matter of prediction, the competitive analysis of the current plan is now a matter of description. The AIRC s expert, Dr. McDonald, had concluded that the adopted plan modestly accomplishes the goal of drawing competitive districts based on his projection of one competitive district in the adopted plan. While I disagree with the characterization of one competitive district as modestly accomplish[ing] the goal of drawing competitive districts, only one competitive district was achieved by the AIRC. Table 1 shows the results of the 2002 general election. Only in the 1 st Congressional District was there any level of competition, with the Republican winner achieving only 51.88% of the two-party vote. In every other district, the winner was able to achieve at least 61% of the two-party vote. 1 David R. Mayhew (1974). The Case of the Vanishing Marginals, Polity. Mayhew s standard is based on his observation that incumbents who lost reelection has often won their previous election by less than 55% of the vote, signaling to potential challengers, their funders, and the media that the incumbent might be vulnerable. 1
TABLE 1: RESULTS OF 2002 ARIZONA CONGRESSIONAL ELECTIONS DISTRICT WINNER % DEMOCRATIC VOTE COMPETITIVENESS 1 Renzi 48.12% Competitive/ Republican 2 Franks 37.89% Republican 3 Shadegg 31.03% Republican 4 Pastor 70.78% Democratic 5 Hayworth 37.21% Republican 6 Flake 32.37% Republican 7 Grijalva 61.42% Democratic 8 Kolbe 34.66% Republican Margins can be inflated by incumbency status. Five incumbents Shadegg, Pastor, Hayworth, Flake, and Kolbe participated in the 2002 general elections in Arizona. To control for the incumbency effect, I accept Dr. McDonald s finding presented earlier that the incumbency advantage in Arizona congressional elections is 9.4 percentage points. To account for this advantage, I add 9.4 percentage points to the share of the twoparty vote obtained by the non-incumbent s party in a district. If the two shares were now added, the sum would be 109.4% of the vote, so each candidates vote share is divided by 109.4% to standardize the two-party vote share, i.e. to convert it back to a pair of percentages that add to 100%. 2 This computation then results in an adjusted share of the two-party vote that deflates the advantage held by the incumbent. Table 2 presents the results of the same analysis as the table above, adjusting and highlighting the figures for the five districts with incumbents in the races. Given the presentation of just the adjusted % Democratic vote, these figures are lower than in the previous table where there is a Democratic incumbent; higher where there is a Republican incumbent. In no case, however, does the adjustment significantly change 2 This method is used because it does not assume that the incumbency advantage is the same across all values of the incumbents vote shares. For example, we would not expect that an incumbent who garnered 48% of the two-party vote benefited as much from incumbency as one who garnered 70% of the two-party vote. Thus, a simple addition or subtraction method of accounting for incumbency advantage seems to be inappropriate. 2
the status of the competitiveness of the district. The margins of victory by the five incumbents are so wide, that there is no evidence that the opposition party would have been competitive absent their presence. TABLE 2: RESULTS OF 2002 ARIZONA CONGRESSIONAL ELECTIONS ADJUSTED FOR INCUMBENCY EFFECTS DISTRICT WINNER ADJUSTED % DEMOCRATIC VOTE COMPETITIVENESS 1 Renzi 48.12% Competitive/ Republican 2 Franks 37.89% Republican 3 Shadegg 36.96% Republican 4 Pastor 64.70% Democratic 5 Hayworth 42.61% Republican 6 Flake 38.18% Republican 7 Grijalva 61.42% Democratic 8 Kolbe 40.27% Republican How would the adopted plan fare compared to alternatives? As indicated earlier, I have been asked specifically to compare the adopted plan to the Downtown Competitive (DTC) plan. This analysis involves reassigning precincts from the extant district to districts proposed in the DTC and designated in the data provided to me by the letters A through H. 3 The assumption in this analysis is that voters would have voted for the candidate of the same party even if they were presented with a different pair of candidates. No assumptions are made about who those candidates might be. 3 These designations are included in the AZCONG02_DTC_MASTER.xls file in the fourth and fifth columns (DTC and CD). As points of reference, District A encompasses most of the voters in current District 2, plus a significant block of those in current District 3; District B, significant parts of current Districts 3, 4, and 5; District C, mainly current District 1, with some of current District 2; District D, mainly current District 4, with some addition from current Districts 2, 3, and 7; Districts E and F redivide current Districts 5 and 6; District G, mainly current District 7, with slight additions from current Districts 1 and 8; and, District H, largely current District 8, with slight additions from current District 7. The cross-tabulation of voters in these two sets of districts is found in the attached file: DTC02_T2.xls. 3
The same kind of data as presented above are presented in Table 3 for the districts as configured in the DTC. The DTC configuration produces an additional competitive district, with two districts being within the +/- 5 percentage point range of 50% of the two-party vote. All other districts appear to be safely in the hands of one or the other of the two major parties. TABLE 3: VOTING PATTERNS FROM 2002 CONGRESSIONAL ELECTIONS IN DTC PROPOSED DISTRICTS DISTRICT % DEMOCRATIC VOTE COMPETITIVENESS A 36.37% Republican B 45.61% Competitive/ Republican C 47.47% Competitive/ Republican D 63.46% Democratic E 29.99% Republican F 34.46% Republican G 85.38% Democratic H 35.62% Republican These results, however, are based on voting confounded by incumbency effects. Again, adjustments must be made, and because Arizona now has eight sitting members of the House, the analysis is reconducted as if the majority party in each of these districts benefited from an incumbency advantage. 4 Thus, as above, the minority party in each district is credited with a 9.4 percentage point bonus, the two-party share is recomputed and standardized as before, and Table 4 presents the results of the adjusted competitiveness analysis. The overall conclusion does not change, although the level of competition within the competitive districts would appear to be heightened. The adjustment makes both Districts B and C appear to be very competitive, were these open seats. 4 Even voters who do not live within the boundaries of the district of a Member of Congress have a greater opportunity to hear about the incumbent, particularly if media markets are not highly segmented. Thus, perhaps the 2002 Arizona might be seen as being more competitive if we control for the effect of incumbency. 4
If the extant plan has modestly accomplished the goal of achieving competitiveness, the DTC appears to have a plan that could double that accomplishment. TABLE 4: VOTING PATTERNS IN PROPOSED DTC DISTRICTS ASSUMING NO INCUMBENTS DISTRICT ADJUSTED % DEMOCRATIC VOTE COMPETITIVENESS A 41.84% Republican B 50.29% Competitive/ Democratic C 51.99% Competitive/ Democratic D 58.01% Democratic E 36.01% Republican F 40.09% Republican G 78.04% Democratic H 41.16% Republican Compliance with the Voting Rights Act (VRA) Of course, no plan could be adopted if it led the state out of compliance with the VRA. The data used here show that Arizona s Hispanic Voting Age Population (HVAP) comprises 21.32% of the state s VAP, and two of the eight Representatives are Hispanic. Table 5 displays the VAP and relative proportions of HVAP and White VAP for the districts proposed under the DTC. 5
TABLE 5: VAP AND ETHNICITY OF VAP IN PROPOSED DTC DISTRICTS VAP HVAP % WVAP% A 494,455 9.05% 84.30% B 489,729 23.71% 66.28% C 459,558 10.91% 68.84% D 423,655 48.17% 40.12% E 488,731 9.18% 85.57% F 459,187 14.23% 78.24% G 452,909 44.78% 45.24% H 495,461 15.02% 77.79% TOTALS 3,763,685 21.32% 68.96% Given the current level of Hispanic representation, to what extent can we provide evidence that Hispanics will have an opportunity to elect representatives of their choice? Given the earlier conclusion of Dr. Lisa Handley, a consultant for the AIRC that [v]oting in Arizona is frequently, but not always, racially/ethnically polarized, the districts of the most interest are D and G, where Hispanics comprise a significant proportion of the VAP, and no group comprises a majority. The simplest test of racially polarized voting is homogeneous precinct analysis in which the voting patterns within highly segregated precincts, i.e. > 90% of one racial/ethnic group, are compared to those similarly segregated, but of the another racial/ethnic group. District D contains no such precincts, but District G contains thirteen > 90% HVAP precincts and four > 90% WVAP precincts. While this is a small proportion of the 238 6
precincts in that district, we can see if a pattern is evident. Fortunately, all seventeen precincts were in District 7 in the 2002 election, so all voters faced a choice between Democrat Raul Grijalva and Republican Ross Hieb. In the thirteen Hispanic precincts, Grijalva garnered 63.6% of the 8,477 two-party votes cast. In the four Anglo precincts, Grijalva netted only 37.1% of the 3,837 two-party votes cast. This evidence would suggest ethnically polarized voting. 5 This suggests the need for additional analysis. The next step is to conduct a two-equation bivariate regression analysis. Unlike the homogeneous precinct analysis, the regression analysis uses information on all of the precincts. Thus, estimates of voter behavior can be derived even where there are no homogenous precincts. A two-equation method controls for differential turnout across ethnic groups, which may be important in Arizona. Dr. Handley has noted in earlier documents on this case that minorities in Arizona have a lower turnout rate than do whites. First, the percentage of the VAP that voted for a candidate is regressed on the Hispanic percentage of the VAP. Second, the percentage of the VAP that voted for any other candidate in the race is regressed on the Hispanic percentage of the VAP. These equations yield estimates of the percentage of the Hispanic VAP voting for the candidate of interest; of the percentage of the Hispanic VAP voting for candidates other than the candidate of interest; the percentage of the non-hispanic VAP voting for the candidate of interest; and the percentage of the non-hispanic VAP voting for other candidates. The correction for possible turnout differentials is achieved by dividing the percentage of an ethnic group that voted for the candidate of interest by the sum of that percentage plus the percentage of that group voting for other candidates. This analysis is conducted for both Districts G and D in the DTC plan. District G is considered first because of the homogeneous precinct analysis. Two sets of equations are estimated. As noted above, a Hispanic candidate ran in 2002 in most of the precincts which would be included in District G, but, in addition, some precincts which were in District 1 in 2002 would be added under the DTC plan. In those precincts, voters also had a chance to vote for or against a Hispanic candidate (George Cordova), so a separate analysis is done with them included. The analysis reveals that 78.1 % of Hispanics supported Grijalva, and 56.7% of non- Hispanic voters did so. Thus, while there is a clear differential, it appears that Grijalva would have won regardless of the ethnic composition of the voters in his district. When the precincts from District 1 are added, 79.5% of Hispanic voters supported the Hispanic candidate they were offered, but only 52.7% of the non-hispanic voters did so. The addition of these precincts moves the electorate toward more polarization, but, still a majority of non-hispanic votes supported the Hispanic candidate. 5 These percentages are not very different from Grijalva s overall level of support. He won with 61.42% of the two-party vote. 7
Turning to District D of the DTC plan, the same analysis is performed. Most of the precincts in this proposed district include voters who, in 2002, faced a choice between the Hispanic Democratic candidate Ed Pastor and Republican Jonathan Barnert. In addition, the plan would add precincts in which the voters in 2002 chose between Hispanic Democrat Randy Camacho and Republican Trent Franks and some precincts from the district which featured the Grijalva v. Heib race just discussed. Again, the regression analysis shows that if one looks at just the precincts where voters chose between Pastor and any other candidate, a majority, 53.5% of non-hispanic voters supported Pastor. When the additional precincts are added to the analysis, the number of non-hispanics supporting the Hispanic candidates drops below a majority to 44%. In both sets of precincts the statistical estimate of percentage of Hispanic voters supporting Pastor or a Hispanic candidate is beyond empirical possibilities. In sum, while the homogeneous precinct analysis raises a red flag, the subsequent statistical analysis suggests that the DTC enhances competitiveness, but not at the expense of Hispanics ability to elect candidates of choice. 8