Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

Gerrymandering or geography? How Democrats won the popular vote but lost the Congress in 2012:

Gerrymandering or geography? How Democrats won the popular vote but lost the Congress in 2012: This article assesses whether the antimajoritarian outcome in the 2012 US congressional elections was due more to deliberate partisan gerrymandering or asymmetric geographic distribution of partisans. The article first estimates an expected seats–votes slope by fitting past election results to a probit curve, and then measures how well parties performed in 2012 compared to this expectation in each state under various redistricting institutions. I find that while both parties exceeded expectations when controlling the redistricting process, a persistent pro-Republican bias is also present even when maps are drawn by courts or bipartisan agreement. This persistent bias is a greater factor in the nationwide disparity between seats and votes than intentional gerrymandering. Keywords Congress, legislative elections, gerrymandering, 2012 American elections Leading into the 2012 general election in the United States, geographic districts. It is unsurprising that Republicans much of the media’s prognostication focused on the possi- won more than their fair share of seats in states where they bility that President Barack Obama might win reelection drew the maps. However, Democrats also underperformed with a majority of the Electoral College yet a minority of under bipartisan maps, and gained only small advantages the popular vote. In retrospect, Obama won a comfortable from their own maps, suggesting their main issue is not ger- popular vote victory, but the same election saw a parallel rymandering, but districting itself. “antimajoritarian” outcome in the House Representatives: The observation that Republicans appear to have a natural Republicans won just 49.4% of the aggregated two-party advantage in the geographic dispersion of their voters is not vote and yet won 54% of the seats. just a recent one. Erikson studied this phenomenon in north- On the surface, Republican partisan gerrymandering ern districts in the 1960s, concluding that “the tendency appears to explain this disparity. The argument that toward a Republican gerrymander in the distribution of con- Democrats underperformed in their seat share due to stituency vote” was “the ‘natural’ state of affairs” and “more Republican control of redistricting in many large states is an accident of geography than the intentional creation of relatively simple. Firstly, it is certainly true that Republicans Republican legislatures” (Erikson, 1972: 1241–1243). controlled this process in more states, representing more In the 1970s, this bias seemed to reverse to the benefit of seats. In addition, in each of these states, Democrats won Democrats, largely due to overwhelming Democratic control fewer seats than any reasonable allocation of the popular of districting in the South (see e.g. Brunell, 1999; McGhee, vote would suggest was “fair.” For example, Republicans 2012). In recent years, however, Erikson’s thesis has received won a large majority of the seats in Pennsylvania, North Carolina, and Michigan, despite losing the mean popular Washington University in St. Louis, USA vote by district in each state. Corresponding author: However, the problem for Democrats might actually be Nicholas Goedert, Postdoctoral Research Associate, Washington more fundamental: the current geographic distribution of University in St. Louis, Campus Box 1063, One Brookings Drive, partisans now leaves Democrats at a disadvantage so long St. Louis, MO 63130, USA. as congressional representation is based on contiguous Email: [email protected] Creative Commons CC BY-NC: This article is distributed under the terms of the Creative Commons Attribution-NonCommercial 3.0 License (http://www.creativecommons.org/licenses/by-nc/3.0/) which permits non-commercial use, reproduction and distribution of the work without further permission provided the original work is attributed as specified on the SAGE and Open Access pages (http://www.uk.sagepub.com/aboutus/openaccess.htm). 2 Research and Politics Figure 1. Seats–votes curve in US congressional elections, 1972–2012. renewed attention. Hirsch, for example, examines the 2000 for individual candidates into seats. It has been almost uni- redistricting cycle and asserts that “Democratic concentra- versally observed that electoral systems employing single- tions in urban areas make it easier for Republicans to gerry- member districts yield seat majorities that exaggerate vote mander successfully…[and] relatively harder for Democrats majorities (Lijphart, 1999; McDonald, 2009; Rae, 1967). to gerrymander successfully” (Hirsch, 2003: 196). Chen and To the extent that this exaggeration is not biased to favor Rodden (2013) use random districting simulations of Florida one party, it is often seen as a feature of such systems rather and other states to argue that the Democratic Party is disad- than a bug, creating governing mandates out of what would vantaged even under neutral districting methods, tracing this otherwise be the confusion of unstable plurality coalitions. bias back to urban population shifts during the industrial The exaggeration tends to take the shape of a probit or logit revolution. In addition, through a case study of several ideo- function, although the slope (i.e. the sensitivity) of the logically neutral proposals to redistrict Virginia, Altman and curve has been found to vary widely among electoral sys- McDonald conclude “there may be some modest truth to the tems (e.g. King and Browning, 1987; Taagepera and claim that urban Democrats are inefficiently concentrated Shugart, 1989; Tufte, 1973). within their urban communities from a redistricting stand- Tufte (1973) proposed that a system of districting must point” (Altman and McDonald, 2013: 830). pass two tests to be “minimally democratic.” Firstly, it must Several recent trends, however, might cast doubt on the be responsive such that an increase in votes for one party will lasting relevance of Erikson’s assertion. These include translate into an increase in seats, and secondly, it must be more sophisticated and varied redistricting institutions and unbiased in treating both parties alike. We therefore start tools and changing demographic patterns, particularly the from the premise that a fair assignment of seats to parties will dramatic rise in Hispanic population. This note takes a first be not be biased in favor of one party, but also will not require cut at adjudicating this question as applied to the 2012 elec- proportional representation. Rather, we will assume that a tion results. party should expect to win a proportion of seats in line with historical patterns found in modern congressional elections. The “fair expectation” for seats given a vote share is Estimating the seats–votes curve thus estimated by imputing a responsiveness slope that is To assess the bias in maps of individual states, we must first average for all congressional elections since the nationwide establish how a “fair” map might translate the popular vote implementation of equal-population districts. Figure 1 Goedert 3 shows the relationship between national vote share and estimate above, translates into winning 56% of seats. In seats won in congressional elections since 1972, as well as actuality, however, Democrats won only 5 of Michigan’s 14 a fit line using both probit (solid line) and ordinary least seats (36%), 20% less than the expected number of seats in squares (OLS; dashed line). Within the observed range, that state. these two methods yield almost identical results, indicating Each state is coded for redistricting control by that a 1% increase in vote share will produce about a 2% Republicans, Democrats, or some other institution (e.g. increase in seat share. Thus, winning 55% of the vote will commission, court, bipartisan agreement) to assess whether generally yield about 60% of the seats. The estimated 2012 Republicans exceeded their expected seat share more when result (not included in the fit line) falls far below this line, they controlled the redistricting process. Table 1 shows bias demonstrating the Democrats’ underperformance com- results for five categories of states, with negative numbers pared with historical averages. in the last column indicating the degree of pro-Republican The probit curve has a slope coefficient of 0.026, repre- bias. The first three subheads show states with at least six senting responsiveness, and a constant of −0.040 (where congressional districts with maps drawn by Republicans, the independent variable is the Republican percentage point Democrats, and bipartisan agreement/courts, respectively, advantage in the aggregated popular vote, and the depend- while the last two subheads show states with the largest ent variable is share of seats won). This coefficient of 0.026 Hispanic populations and those in the Deep South, catego- is used throughout the analysis to represent the “expected” ries that will be analyzed separately. responsiveness of the seats–votes curve, equivalent to the ρ term in King and Browning’s (1987) model. Seats won versus seats expected by In lieu of using national election data to measure the redistricting control responsiveness of congressional seats to votes, we can alter- nately estimate this slope using state-by-state election data If the overall pro-Republican bias in the national election from the same 1972–2010 period, using mean two-party outcome was due predominantly to Republicans control- vote share by district as the independent variable, and state- ling the districting process in more states, we should expect wide seat share as the dependent variable, similar to the to observe opposing biases of similar magnitudes in indi- 2012 results presented below. This method (detailed in vidual states when Republicans and Democrats controlled Table A1 of Supplementary Material) yields a slope coeffi- the process. In addition, we would expect little or no bias in cient of 0.0234. In addition, unopposed races in the South, states where maps were drawn by courts or bipartisan particularly in the first two decades, distort this result: the agreement. At first glance, neither of these hypotheses coefficient estimate is 0.0271 if the South is excluded. seems true. Using this method, we can also include fixed effects for dec- In every state districted by Republicans, Democrats won ade, none of which are significant. Although the bias in con- fewer seats than their historical expectation, and in six gressional maps appears to vary over time, there is little cases they underperformed by 20% or more. It appears as variation in responsiveness, either within this period or though Republicans gained dramatic benefits across the when comparing the last 40 years to earlier decades in the board from holding the reins of districting. 20th century. Imputing the lowest slope value under this In contrast, Democrats only slightly exceeded their method (0.0234) still yields substantively very similar expected seat share in the three states—Illinois, results (shown in Table A2 of Supplementary Material). Massachusetts, and Maryland—where they controlled the process, gaining just a fractional seat above expectation in each. For instance, Illinois Democrats won a smaller major- Methodology for vote share and ity in their delegation than Republicans won in Pennsylvania seat share or Ohio, despite winning a much larger vote share. Although Drawing on the 2012 election results, I have calculated winning all of Massachusetts’ nine districts may seem a each party’s mean vote share across each state’s congres- wildly inequitable distribution, by winning 76% of the sional districts, using mean rather than the aggregate share mean vote Massachusetts Democrats could expect to win so that each district is weighted equally regardless of turn- 91% of the seats under a “fair” map. If John Tierney had out and unopposed races can be included. Where a candi- won 1% less in his MA-6 race, Democrats would have date ran completely unopposed, I have assigned that slightly underperformed their expected share. While candidate’s party 100% of the vote; where a candidate ran Democrats underperformed by an average of 19% under against only minor parties, I have assigned the opposing Republican gerrymanders, they only exceeded expectation party the vote share of the minor candidates. I then compare by 5% under these Democratic gerrymanders. the mean vote share with the expected seat share under a In addition, we observe bias even where we should “fair” map with zero bias and a historically average seats– expect none in the redistricting process. Democrats also votes curve. For example, Michigan Democrats won a fell short of expectation in several states with bipartisan or mean vote share of 53%, which, when we apply the slope court-drawn maps. For example, despite a constitutional 4 Research and Politics Table 1. Seats won versus mean vote share by gerrymandering party: 2012 congressional elections. Republican gerrymanders Dem. Dem. Dem. seats Won–Exp. State CDs Vote share Seats won Expected Difference Indiana 9 46% 22% 42% −20% Michigan 14 53% 36% 56% –20% Missouri 8 43% 25% 36% –11% North Carolina 13 51% 31% 52% –21% Ohio 16 48% 25% 46% –21% Pennsylvania 18 51% 28% 51% –23% Tennessee 9 38% 22% 28% –6% Virginia 11 49% 27% 48% –21% Wisconsin 8 51% 38% 52% –15% Total 106 48% 28% 47% –19% Democratic gerrymanders Dem. Dem. Dem. seats Won–Exp. State CDs Vote share Seats won Expected Difference Illinois 18 56% 67% 63% 4% Massachusetts 9 76% 100% 91% 9% Maryland 8 64% 88% 76% 11% Total 35 63% 80% 75% 5% Bipartisan or court gerrymanders Dem. Dem. Dem. seats Won–Exp. State CDs Vote share Seats won Expected Difference Colorado 7 50% 43% 50% –7% Florida 27 48% 37% 46% –9% Kentucky 6 39% 17% 29% –12% Minnesota 8 57% 63% 63% –1% New Jersey 12 57% 50% 65% –15% New York 27 67% 78% 81% –3% Washington 10 53% 60% 57% 3% Total 97 55% 54% 61% –7% High Hispanic population states Dem. Dem. Dem. seats Won–Exp. State CDs Vote share Seats won Expected Difference Arizona 9 48% 56% 45% 10% California 53 60% 72% 70% 2% New Mexico 3 54% 67% 59% 8% Nevada 4 51% 50% 53% –3% Texas 36 43% 33% 37% –3% Total 105 53% 56% 56% 0% Deep South states Dem. Dem. Dem. seats Won–Exp. State CDs Vote share Seats won Expected Difference Alabama 7 35% 14% 21% –7% Georgia 14 39% 36% 28% 8% Louisiana 6 32% 17% 20% –4% Mississippi 4 39% 25% 29% –4% South Carolina 7 41% 14% 31% –17% Total 38 37% 24% 26% –2% Goedert 5 amendment prohibiting Republican legislators from using biased against Democrats with one exception. Because we partisanship to draw maps in Florida, the GOP neverthe- therefore might expect these states to be much differently less managed to win 17 seats with 51.4% of the vote, sur- impacted by both urbanization and the gerrymandering passing expectation by 2.5 seats. Even under bipartisan party compared to the rest of the nation, they are excluded gerrymandering in New York, in which Democrats won 21 from the regression analysis below. of 27 seats, their vote share suggested they should have won 22. Across the seven states with bipartisan or court Regression results gerrymanders, Republicans exceeded expectation by an average of 7%. To more directly approach Chen and Rodden’s (2013) argu- So how many seats did this underlying disadvantage ment that Democrats are disadvantaged due to their heavy cost the Democrats? If we imagine that these bipartisan or concentration in cities, I analyzed these results using an court maps were unbiased, and that Democrats and OLS regression, including 2010 US Census data on race Republicans received equal benefit from their own maps and urbanization. Table 2 depicts regression results with (for example, a 12% advantage as an average), this would each state weighted by number of districts, excluding five have yielded 16 or 17 additional seats, likely getting the Deep South states and states with only one or two districts. Democrats within a couple seats of the majority. By con- The dependent variable is the difference between trast, the disparity between the number of seats gerryman- Democratic seats won and the number of seats expected dered by Republicans compared to Democrats likely costs given their vote share. A high positive value is a map dis- Democrats about nine seats. This initial analysis reveals torted in favor of Democrats, while a high negative value is that geography is a slightly greater factor than intentional a map distorted in favor of Republicans. Dummy variables gerrymandering in explaining why Democrats won fewer are assigned for partisan redistricting procedures; the seats than expected from their vote share. excluded category is bipartisan or court-drawn maps. In If there is any area of the country where the geographic addition, controls are included in some models for the per- distribution of partisans has not led to an underrepresenta- cent of the population that lives in urban areas or that is tion of Democrats, we might expect to observe it where African-American or Hispanic. The “Hispanic Dummy” in Democratic voting strength does not hew as closely to the Model 1 is a “1” for the five most heavily Hispanic states. black/white or urban/rural divide. In particular, we find this Model 1 reaffirms the three central conclusions from pattern interrupted in areas with very large Hispanic popu- Table 1. Firstly, the effect of partisan control of the district- lations, as Hispanics tend to be both less saturated in their ing process is significant and in the expected direction. support for Democrats and more geographically dispersed Secondly, as we can see from the negative and significant than African-Americans living in large urban areas. In the constant, which captures the bias in states with a bipartisan five states with the highest proportion of Hispanics or court-drawn map and without a large Hispanic popula- (Arizona, California, New Mexico, Nevada, and Texas), tion, maps are distorted in favor of Republicans even when Democrats won a seat share very close to expectation in we control for partisan gerrymanders. Finally, this distor- each state, despite not controlling the process in any of tion is not present in the case of the most heavily Hispanic them. It is possible that non-partisan commissions in states. California and Arizona may have contributed to greater Model 2 tests the effect of minority population propor- fairness, but the ease of drawing geographically large, tions, includes controls for state size and overall partisan- majority Hispanic districts in these states, (e.g. AZ-4, ship of the state, and also yields a closer test of the Chen CA-16, CA-51, and TX-23) might have also mitigated the and Rodden (2013) hypothesis by including the urbaniza- advantage Republicans have in other regions given the dis- tion variable. Chen and Rodden hypothesize that the distor- tribution of their voters. tion is due to population shifts toward urban areas. If this The final subhead of Table 1 depicts results from five were true, we would expect more distortion against states in the Deep South. In these states, voting is highly Democrats in heavily urbanized states. Consistent with racial polarized and, unlike most of the rest of the nation, Table 1 and Model 1, a larger Hispanic population reduces much of the African-American population is rural. In addi- bias against Democrats, but the size of the African- tion, amendments to the Voting Rights Act (VRA) have American population has no significant effect on distortion, been interpreted to require the drawing of African- and we see no effect for urbanization. American-majority or African-American-influence dis- Model 3, including only states with more than six dis- tricts across rural parts of these states, with district maps tricts, paints a different picture, showing a significant nega- requiring Department of Justice preclearance under the tive coefficient for urbanization. Among larger states, VRA. Past research has suggested that this may constrain which likely include both urban and rural areas, heavily maps to resemble Republican gerrymanders even when urbanized states (e.g. New Jersey and Pennsylvania) are drawn by another party (Goedert, 2012; Hill, 1995; Lublin, more often heavily distorted against Democrats than more 1999), and we do see that results in these states are slightly rural states (e.g. Minnesota and Wisconsin) after 6 Research and Politics Table 2. Regression results. Democrat % seats won Hisp. dum minus % seats expected Model 1 Model 2 >6 CDs Democratic gerrymander 9.13* 10.1** 16.6*** (4.63) (4.79) (4.75) Republican gerrymander −11.2*** −4.08 −13.6** (2.89) (3.81) (4.86) Percent African-American – −0.41 −0.29 (0.26) (0.24) Percent Hispanic – 0.58** 0.77*** (0.22) (0.24) Urbanization – 0.046 −0.72** (0.22) (0.34) Democratic vote – 0.32 0.11 (0.21) (0.24) Number of seats – −0.29* −0.16 (0.16) (0.18) Hispanic dummy 9.95*** – – (3.11) Constant −5.52** −25.5 45.0 (2.26) (15.8) (29.2) Observations 33 33 21 R-squared 0.557 0.641 0.829 Notes: Standard errors in parentheses. Data points weighted by state size. ***p < 0.01, **p < 0.05, *p < 0.10. Table 3. Seats won versus mean vote share by gerrymandering party: 2012 presidential vote (summary). Summary table Dem. Dem. Dem. seats Won–Exp. CDs Vote share Seats won Expected Difference Republican gerrymanders 106 50% 28% 50% −22% Democratic gerrymanders 35 61% 80% 72% 8% Bipartisan or court gerrymanders 97 57% 63% 64% −1% High Hispanic population 105 55% 57% 59% −2% Deep South states 38 43% 21% 36% −15% controlling for the gerrymandering party. Furthermore, the districts, further suggesting pro-Republican bias. We can coefficients for partisan maps increase when we limit the substitute Obama’s margin for the congressional election sample to larger states, possibly indicating the greater flex- result to measure bias under the various redistricting regimes. ibility parties have in drawing districts in such states. The results of replicating Table 1 using presidential elec- tion results are summarized in Table 3 and are detailed in Table A3 of Supplementary Material. In the case of partisan Robustness check: Presidential maps and heavily Hispanic states, the average bias is very election results similar to the bias under the actual congressional election Although the current congressional map has thus far only results. Notably, the difference in bias between Republican seen one cycle of election results, there has been another and Democratic gerrymanders remains the same at 14%. election held across all 435 of these districts that we can use However, the pro-Republican bias under bipartisan and court to test the robustness of this paper’s finding: the 2012 presi- gerrymanders largely disappears. There are likely two expla- dential election. Despite winning with 52.0% of the two- nations for this difference. Firstly, President Obama won party popular vote, Obama won only 209 congressional three districts in Minnesota and five districts in New York Goedert 7 Declaration of conflicting interest with 52% or less of the vote, which might be described as luck. However, this result also suggests that the asymmetry The author declares that there is no conflict of interest. in the geographic distribution of partisans is not constant across states and regions. In some “bluish” states, the more Funding conservative areas such as upstate New York and rural This research received no specific grant from any funding agency Minnesota may be only marginally Republican. These dis- in the public, commercial, or not-for-profit sectors. tricts may be won by Republicans in a nationally tied elec- toral environment but captured by Democrats in a climate Notes somewhat more favorable to them, such as Obama’s 4% 1. While Hirsch argues that the combination of redistricting popular vote victory. In contrast, in the Deep South where the control and geographic imbalance biased the 2002 election more conservative regions are deeper “red,” probably exag- results against the Democrats by 25 seats, he does not dis- gerated by VRA considerations, the bias against Democrats tinguish between these two factors in that estimate, and he is actually exacerbated as their vote majority increases. argues that almost all bias can be located within four states with Republican-controlled districting. 2. The linear method estimates an average slope of 2.02 for Conclusion the past 40 years, compared with Tufte’s (1973) average of 2.09 for the preceding 70 years. Tufte justifies using a linear Both the state-by-state results and aggregated regression estimate, as opposed to probit or logit, because the major- analysis suggest that while deliberate partisan gerrymander- party vote shares rarely fall outside of the 35–65% range. ing produces additional seats for the districting party, parti- However, as vote shares in several states in the 2012 election san gerrymandering is not a sufficient explanation for the fall outside of this range, a curve that will deal more appro- overall antimajoritarian outcome. Instead, pro-Republican priately with extreme values is needed for our purpose. bias is observed under all districting regimes. In addition, 3. The constant in this regression represents approximately a 3–4% bias in favor of Democrats over this period. When bro- the regression results offer possible support for the Chen ken down by decade (shown in Table A1 of Supplementary and Rodden (2013) thesis that urbanization has created bias Material), the bias estimate aligns with past research in show- while also forecasting its possible demise if patterns of ing Democratic bias in the 1970s and 1980s, shifting toward rapid Hispanic population growth continue. Republican bias in the 2000s (e.g. King and Gelman, 1991; It is important to note the limits to these conclusions. McGhee, 2012), possibly due to the same gerrymandering Firstly, while asymmetric population distributions are a plausi- and geography trends observed here for 2012. The sign of ble explanation for persistent bias, and one supported by previ- this bias is reversed under the state elections data method ous research, they are not the only possible cause. For example, (also in Table A1 of Supplementary Material), with the dif- one might claim that incumbency could give Republicans ference likely attributable to the method of imputation for advantages in more marginal districts (see McGhee, 2012). unopposed races. If estimated using a logit function on the This article does not attempt to isolate that cause. national data, the slope coefficient is .0415, with all results substantive unchanged. This analysis does not imply that Democrats are 4. In cases where a candidate runs unopposed and no votes are doomed to the minority even for the next decade. It does collected, no votes are added to the national total, but a 100% indicate they are unlikely to retake the House in an essen- vote share is imputed into the state result. This will lead to tially tied national election. Yet national elections are not a difference in responsiveness between the methods where usually this close: Democrats reversed a Republican ger- unopposed incumbents are predominantly one party. About rymander in Pennsylvania, Virginia, Ohio, and Michigan 70% of such races in the data set occur in the South, with in 2006 or 2008 (all states with aggressive Republican about two-thirds of those being Democrats in the 1970s and maps). The 2012 maps leave the Democratic Party several 1980s. openings; for example, Republicans now sit in five 5. Obviously, Democrats could not have hoped to perform bet- Pennsylvania districts won by Obama in 2008. To win ter in Massachusetts than they did. At the state level, how- these seats, Democrats will need the electorate to look like ever, this example illustrates the national phenomenon of Democrats failing to maximize their vote by oversaturating 2006 or 2008, but this is far from unprecedented: their support in certain areas. In addition, Democrats con- Democrats won the popular vote by at least 5 points in 12 trolled the process in Arkansas, but won none of its four of the last 20 cycles. But given the unequal concentra- seats; earning an average of 35% of the vote across this state tions of vote share in most states, not just those with would have predicted winning one seat under a “fair” map. Republican gerrymanders, a Democratic majority will be 6. This average disparity is extremely close to the 6% dispar- a bit more difficult than it should be. ity observed nationwide, as the Democrats’ 1% popular vote advantage is estimated to correspond to 52% of seats Supplementary Material expected, compared to 46% of seats actually won. 7. Reducing the Republican bias by 7% in the 238 seats under The entire Supplementary Material is available at: http://bit. Republican, Democratic, or Bipartisan control in Table 1 nets ly/1jOtnma 8 Research and Politics the Democrats (238 × .07) = 16.7 seats. If we instead assume Goedert N (2012) Southern redistricting under the VRA: A the level of bias shown in Table 1, but allocate 70 seats to both model of partisan tides. In: the 2012 state politics & policy Democratic and Republican control (rather than 35 and 106, conference, Houston, TX. Available at: http:// 2012sppcon respectively), this reduces the number of Republican seats in ference.blogs.rice.edu/files/2012/02/VRA-Partisan-Tides- Republican-controlled maps by (36 × .19) = 6.8 seats, and SPPC-version1.pdf. increases the number of Democrats in Democratic-controlled Hill KA (1995) Does the creation of majority black districts aid maps by (35 × .05) = 1.75 seats (for a total of 8.6 seats). Republicans? An analysis of the 1992 elections in eight 8. The exception here is Georgia, which is biased toward southern states. Journal of Politics 57(2): 384–401. the Democrats despite being districted in 2011 by Hirsch S (2003) The United States House of Unrepresentatives: Republicans. This is likely attributable to the novel What went wrong in the latest round of congressional redis- strategy of “minority influence” districts employed in a tricting. Election Law Journal 2(2): 179–216. Democratic gerrymander in the 2000s, a strategy upheld King G and Browning RX (1987) Democratic representation and in Georgia v. Ashcroft (2003), combined with the need to partisan bias in congressional elections. American Political avoid retrogression from this map to achieve VRA clear- Science Review 81(4): 1251–1273. ance in the next decade. King G and Gelman A (1991) Systemic consequences of incum- 9. Because of the inclusion of other controls with continuous bency advantage in U.S. House Elections. American Journal values in Models 2 and 3, the value of the constant is no of Political Science 35(1): 110–138. longer inherently meaningful. Lijphart A (1999) Patterns of Democracy: Government Forms 10. The coefficients for Republican gerrymanders between mod- and Performance in Thirty-Six Countries. New Haven, CT: els are different at p < .05, but not significantly different for Yale University Press. Democratic gerrymanders. Lublin D (1999) The Paradox of Representation. Princeton, NJ: 11. This explanation seems less plausible given that Mitt Princeton University Press. Romney won 52% of congressional districts despite losing McDonald M (2009) Seats to votes ratios in the United States. the national popular vote by 4%. Unpublished Paper, George Mason University. Available at: http://www.democracy.uci.edu/files/democracy/docs/con ferences/mcdonaldgmu.doc (accessed 31 March 2014). References McGhee E (2012) The history of the House votes-seats dis- Altman M and McDonald MP (2013) A half-century of Virginia crepancy, in two graphs. The Monkey Cage (Political redistricting battles: Shifting from rural malapportionment to Science blog). Available at: http://themonkeycage.org/ voting rights to public participation. University of Richmond blog/2012/11/19/the-history-of-the-house-votes-seats-dis Law Review 47: 771–831. crepancy-in-two-graphs/ (accessed 31 March 2014). Brunell TL (1999) Partisan bias in U.S. congressional elections, Rae D (1967) The Political Consequences of Electoral Laws. New 1952–1996. American Politics Research 27(3): 316–337. Haven, CT: Yale University Press. Chen J and Rodden J (2013) Unintentional gerrymandering: Political Taagepera R and Shugart MS (1989) Seats & Voters: The Effects geography and electoral bias in legislatures. Quarterly Journal & Determinants of Electoral Systems. New Haven, CT: Yale of Political Science 8(3): 239–269. University. Erikson R (1972) Malapportionment, gerrymandering, and party Tufte ER (1973) The relationship between seats and votes in two- fortunes in congressional elections. American Political party systems. The American Political Science Review 67(2): Science Review 66(4): 1234–1245. 540–554. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Research & Politics SAGE

Gerrymandering or geography? How Democrats won the popular vote but lost the Congress in 2012:

Goedert, Nicholas
Research & Politics , Volume 1 (1): 1 – Apr 1, 2014
Download PDF
8 pages

Loading next page...
 
/lp/sage/gerrymandering-or-geography-how-democrats-won-the-popular-vote-but-MeB2WITFSO

References (20)

Publisher
SAGE
Copyright
Copyright © 2022 by SAGE Publications Ltd, and Stichting Research and Politics, unless otherwise noted. Manuscript content on this site is licensed under Creative Commons Licenses.
ISSN
2053-1680
eISSN
2053-1680
DOI
10.1177/2053168014528683
Publisher site
See Article on Publisher Site

Abstract

This article assesses whether the antimajoritarian outcome in the 2012 US congressional elections was due more to deliberate partisan gerrymandering or asymmetric geographic distribution of partisans. The article first estimates an expected seats–votes slope by fitting past election results to a probit curve, and then measures how well parties performed in 2012 compared to this expectation in each state under various redistricting institutions. I find that while both parties exceeded expectations when controlling the redistricting process, a persistent pro-Republican bias is also present even when maps are drawn by courts or bipartisan agreement. This persistent bias is a greater factor in the nationwide disparity between seats and votes than intentional gerrymandering. Keywords Congress, legislative elections, gerrymandering, 2012 American elections Leading into the 2012 general election in the United States, geographic districts. It is unsurprising that Republicans much of the media’s prognostication focused on the possi- won more than their fair share of seats in states where they bility that President Barack Obama might win reelection drew the maps. However, Democrats also underperformed with a majority of the Electoral College yet a minority of under bipartisan maps, and gained only small advantages the popular vote. In retrospect, Obama won a comfortable from their own maps, suggesting their main issue is not ger- popular vote victory, but the same election saw a parallel rymandering, but districting itself. “antimajoritarian” outcome in the House Representatives: The observation that Republicans appear to have a natural Republicans won just 49.4% of the aggregated two-party advantage in the geographic dispersion of their voters is not vote and yet won 54% of the seats. just a recent one. Erikson studied this phenomenon in north- On the surface, Republican partisan gerrymandering ern districts in the 1960s, concluding that “the tendency appears to explain this disparity. The argument that toward a Republican gerrymander in the distribution of con- Democrats underperformed in their seat share due to stituency vote” was “the ‘natural’ state of affairs” and “more Republican control of redistricting in many large states is an accident of geography than the intentional creation of relatively simple. Firstly, it is certainly true that Republicans Republican legislatures” (Erikson, 1972: 1241–1243). controlled this process in more states, representing more In the 1970s, this bias seemed to reverse to the benefit of seats. In addition, in each of these states, Democrats won Democrats, largely due to overwhelming Democratic control fewer seats than any reasonable allocation of the popular of districting in the South (see e.g. Brunell, 1999; McGhee, vote would suggest was “fair.” For example, Republicans 2012). In recent years, however, Erikson’s thesis has received won a large majority of the seats in Pennsylvania, North Carolina, and Michigan, despite losing the mean popular Washington University in St. Louis, USA vote by district in each state. Corresponding author: However, the problem for Democrats might actually be Nicholas Goedert, Postdoctoral Research Associate, Washington more fundamental: the current geographic distribution of University in St. Louis, Campus Box 1063, One Brookings Drive, partisans now leaves Democrats at a disadvantage so long St. Louis, MO 63130, USA. as congressional representation is based on contiguous Email: [email protected] Creative Commons CC BY-NC: This article is distributed under the terms of the Creative Commons Attribution-NonCommercial 3.0 License (http://www.creativecommons.org/licenses/by-nc/3.0/) which permits non-commercial use, reproduction and distribution of the work without further permission provided the original work is attributed as specified on the SAGE and Open Access pages (http://www.uk.sagepub.com/aboutus/openaccess.htm). 2 Research and Politics Figure 1. Seats–votes curve in US congressional elections, 1972–2012. renewed attention. Hirsch, for example, examines the 2000 for individual candidates into seats. It has been almost uni- redistricting cycle and asserts that “Democratic concentra- versally observed that electoral systems employing single- tions in urban areas make it easier for Republicans to gerry- member districts yield seat majorities that exaggerate vote mander successfully…[and] relatively harder for Democrats majorities (Lijphart, 1999; McDonald, 2009; Rae, 1967). to gerrymander successfully” (Hirsch, 2003: 196). Chen and To the extent that this exaggeration is not biased to favor Rodden (2013) use random districting simulations of Florida one party, it is often seen as a feature of such systems rather and other states to argue that the Democratic Party is disad- than a bug, creating governing mandates out of what would vantaged even under neutral districting methods, tracing this otherwise be the confusion of unstable plurality coalitions. bias back to urban population shifts during the industrial The exaggeration tends to take the shape of a probit or logit revolution. In addition, through a case study of several ideo- function, although the slope (i.e. the sensitivity) of the logically neutral proposals to redistrict Virginia, Altman and curve has been found to vary widely among electoral sys- McDonald conclude “there may be some modest truth to the tems (e.g. King and Browning, 1987; Taagepera and claim that urban Democrats are inefficiently concentrated Shugart, 1989; Tufte, 1973). within their urban communities from a redistricting stand- Tufte (1973) proposed that a system of districting must point” (Altman and McDonald, 2013: 830). pass two tests to be “minimally democratic.” Firstly, it must Several recent trends, however, might cast doubt on the be responsive such that an increase in votes for one party will lasting relevance of Erikson’s assertion. These include translate into an increase in seats, and secondly, it must be more sophisticated and varied redistricting institutions and unbiased in treating both parties alike. We therefore start tools and changing demographic patterns, particularly the from the premise that a fair assignment of seats to parties will dramatic rise in Hispanic population. This note takes a first be not be biased in favor of one party, but also will not require cut at adjudicating this question as applied to the 2012 elec- proportional representation. Rather, we will assume that a tion results. party should expect to win a proportion of seats in line with historical patterns found in modern congressional elections. The “fair expectation” for seats given a vote share is Estimating the seats–votes curve thus estimated by imputing a responsiveness slope that is To assess the bias in maps of individual states, we must first average for all congressional elections since the nationwide establish how a “fair” map might translate the popular vote implementation of equal-population districts. Figure 1 Goedert 3 shows the relationship between national vote share and estimate above, translates into winning 56% of seats. In seats won in congressional elections since 1972, as well as actuality, however, Democrats won only 5 of Michigan’s 14 a fit line using both probit (solid line) and ordinary least seats (36%), 20% less than the expected number of seats in squares (OLS; dashed line). Within the observed range, that state. these two methods yield almost identical results, indicating Each state is coded for redistricting control by that a 1% increase in vote share will produce about a 2% Republicans, Democrats, or some other institution (e.g. increase in seat share. Thus, winning 55% of the vote will commission, court, bipartisan agreement) to assess whether generally yield about 60% of the seats. The estimated 2012 Republicans exceeded their expected seat share more when result (not included in the fit line) falls far below this line, they controlled the redistricting process. Table 1 shows bias demonstrating the Democrats’ underperformance com- results for five categories of states, with negative numbers pared with historical averages. in the last column indicating the degree of pro-Republican The probit curve has a slope coefficient of 0.026, repre- bias. The first three subheads show states with at least six senting responsiveness, and a constant of −0.040 (where congressional districts with maps drawn by Republicans, the independent variable is the Republican percentage point Democrats, and bipartisan agreement/courts, respectively, advantage in the aggregated popular vote, and the depend- while the last two subheads show states with the largest ent variable is share of seats won). This coefficient of 0.026 Hispanic populations and those in the Deep South, catego- is used throughout the analysis to represent the “expected” ries that will be analyzed separately. responsiveness of the seats–votes curve, equivalent to the ρ term in King and Browning’s (1987) model. Seats won versus seats expected by In lieu of using national election data to measure the redistricting control responsiveness of congressional seats to votes, we can alter- nately estimate this slope using state-by-state election data If the overall pro-Republican bias in the national election from the same 1972–2010 period, using mean two-party outcome was due predominantly to Republicans control- vote share by district as the independent variable, and state- ling the districting process in more states, we should expect wide seat share as the dependent variable, similar to the to observe opposing biases of similar magnitudes in indi- 2012 results presented below. This method (detailed in vidual states when Republicans and Democrats controlled Table A1 of Supplementary Material) yields a slope coeffi- the process. In addition, we would expect little or no bias in cient of 0.0234. In addition, unopposed races in the South, states where maps were drawn by courts or bipartisan particularly in the first two decades, distort this result: the agreement. At first glance, neither of these hypotheses coefficient estimate is 0.0271 if the South is excluded. seems true. Using this method, we can also include fixed effects for dec- In every state districted by Republicans, Democrats won ade, none of which are significant. Although the bias in con- fewer seats than their historical expectation, and in six gressional maps appears to vary over time, there is little cases they underperformed by 20% or more. It appears as variation in responsiveness, either within this period or though Republicans gained dramatic benefits across the when comparing the last 40 years to earlier decades in the board from holding the reins of districting. 20th century. Imputing the lowest slope value under this In contrast, Democrats only slightly exceeded their method (0.0234) still yields substantively very similar expected seat share in the three states—Illinois, results (shown in Table A2 of Supplementary Material). Massachusetts, and Maryland—where they controlled the process, gaining just a fractional seat above expectation in each. For instance, Illinois Democrats won a smaller major- Methodology for vote share and ity in their delegation than Republicans won in Pennsylvania seat share or Ohio, despite winning a much larger vote share. Although Drawing on the 2012 election results, I have calculated winning all of Massachusetts’ nine districts may seem a each party’s mean vote share across each state’s congres- wildly inequitable distribution, by winning 76% of the sional districts, using mean rather than the aggregate share mean vote Massachusetts Democrats could expect to win so that each district is weighted equally regardless of turn- 91% of the seats under a “fair” map. If John Tierney had out and unopposed races can be included. Where a candi- won 1% less in his MA-6 race, Democrats would have date ran completely unopposed, I have assigned that slightly underperformed their expected share. While candidate’s party 100% of the vote; where a candidate ran Democrats underperformed by an average of 19% under against only minor parties, I have assigned the opposing Republican gerrymanders, they only exceeded expectation party the vote share of the minor candidates. I then compare by 5% under these Democratic gerrymanders. the mean vote share with the expected seat share under a In addition, we observe bias even where we should “fair” map with zero bias and a historically average seats– expect none in the redistricting process. Democrats also votes curve. For example, Michigan Democrats won a fell short of expectation in several states with bipartisan or mean vote share of 53%, which, when we apply the slope court-drawn maps. For example, despite a constitutional 4 Research and Politics Table 1. Seats won versus mean vote share by gerrymandering party: 2012 congressional elections. Republican gerrymanders Dem. Dem. Dem. seats Won–Exp. State CDs Vote share Seats won Expected Difference Indiana 9 46% 22% 42% −20% Michigan 14 53% 36% 56% –20% Missouri 8 43% 25% 36% –11% North Carolina 13 51% 31% 52% –21% Ohio 16 48% 25% 46% –21% Pennsylvania 18 51% 28% 51% –23% Tennessee 9 38% 22% 28% –6% Virginia 11 49% 27% 48% –21% Wisconsin 8 51% 38% 52% –15% Total 106 48% 28% 47% –19% Democratic gerrymanders Dem. Dem. Dem. seats Won–Exp. State CDs Vote share Seats won Expected Difference Illinois 18 56% 67% 63% 4% Massachusetts 9 76% 100% 91% 9% Maryland 8 64% 88% 76% 11% Total 35 63% 80% 75% 5% Bipartisan or court gerrymanders Dem. Dem. Dem. seats Won–Exp. State CDs Vote share Seats won Expected Difference Colorado 7 50% 43% 50% –7% Florida 27 48% 37% 46% –9% Kentucky 6 39% 17% 29% –12% Minnesota 8 57% 63% 63% –1% New Jersey 12 57% 50% 65% –15% New York 27 67% 78% 81% –3% Washington 10 53% 60% 57% 3% Total 97 55% 54% 61% –7% High Hispanic population states Dem. Dem. Dem. seats Won–Exp. State CDs Vote share Seats won Expected Difference Arizona 9 48% 56% 45% 10% California 53 60% 72% 70% 2% New Mexico 3 54% 67% 59% 8% Nevada 4 51% 50% 53% –3% Texas 36 43% 33% 37% –3% Total 105 53% 56% 56% 0% Deep South states Dem. Dem. Dem. seats Won–Exp. State CDs Vote share Seats won Expected Difference Alabama 7 35% 14% 21% –7% Georgia 14 39% 36% 28% 8% Louisiana 6 32% 17% 20% –4% Mississippi 4 39% 25% 29% –4% South Carolina 7 41% 14% 31% –17% Total 38 37% 24% 26% –2% Goedert 5 amendment prohibiting Republican legislators from using biased against Democrats with one exception. Because we partisanship to draw maps in Florida, the GOP neverthe- therefore might expect these states to be much differently less managed to win 17 seats with 51.4% of the vote, sur- impacted by both urbanization and the gerrymandering passing expectation by 2.5 seats. Even under bipartisan party compared to the rest of the nation, they are excluded gerrymandering in New York, in which Democrats won 21 from the regression analysis below. of 27 seats, their vote share suggested they should have won 22. Across the seven states with bipartisan or court Regression results gerrymanders, Republicans exceeded expectation by an average of 7%. To more directly approach Chen and Rodden’s (2013) argu- So how many seats did this underlying disadvantage ment that Democrats are disadvantaged due to their heavy cost the Democrats? If we imagine that these bipartisan or concentration in cities, I analyzed these results using an court maps were unbiased, and that Democrats and OLS regression, including 2010 US Census data on race Republicans received equal benefit from their own maps and urbanization. Table 2 depicts regression results with (for example, a 12% advantage as an average), this would each state weighted by number of districts, excluding five have yielded 16 or 17 additional seats, likely getting the Deep South states and states with only one or two districts. Democrats within a couple seats of the majority. By con- The dependent variable is the difference between trast, the disparity between the number of seats gerryman- Democratic seats won and the number of seats expected dered by Republicans compared to Democrats likely costs given their vote share. A high positive value is a map dis- Democrats about nine seats. This initial analysis reveals torted in favor of Democrats, while a high negative value is that geography is a slightly greater factor than intentional a map distorted in favor of Republicans. Dummy variables gerrymandering in explaining why Democrats won fewer are assigned for partisan redistricting procedures; the seats than expected from their vote share. excluded category is bipartisan or court-drawn maps. In If there is any area of the country where the geographic addition, controls are included in some models for the per- distribution of partisans has not led to an underrepresenta- cent of the population that lives in urban areas or that is tion of Democrats, we might expect to observe it where African-American or Hispanic. The “Hispanic Dummy” in Democratic voting strength does not hew as closely to the Model 1 is a “1” for the five most heavily Hispanic states. black/white or urban/rural divide. In particular, we find this Model 1 reaffirms the three central conclusions from pattern interrupted in areas with very large Hispanic popu- Table 1. Firstly, the effect of partisan control of the district- lations, as Hispanics tend to be both less saturated in their ing process is significant and in the expected direction. support for Democrats and more geographically dispersed Secondly, as we can see from the negative and significant than African-Americans living in large urban areas. In the constant, which captures the bias in states with a bipartisan five states with the highest proportion of Hispanics or court-drawn map and without a large Hispanic popula- (Arizona, California, New Mexico, Nevada, and Texas), tion, maps are distorted in favor of Republicans even when Democrats won a seat share very close to expectation in we control for partisan gerrymanders. Finally, this distor- each state, despite not controlling the process in any of tion is not present in the case of the most heavily Hispanic them. It is possible that non-partisan commissions in states. California and Arizona may have contributed to greater Model 2 tests the effect of minority population propor- fairness, but the ease of drawing geographically large, tions, includes controls for state size and overall partisan- majority Hispanic districts in these states, (e.g. AZ-4, ship of the state, and also yields a closer test of the Chen CA-16, CA-51, and TX-23) might have also mitigated the and Rodden (2013) hypothesis by including the urbaniza- advantage Republicans have in other regions given the dis- tion variable. Chen and Rodden hypothesize that the distor- tribution of their voters. tion is due to population shifts toward urban areas. If this The final subhead of Table 1 depicts results from five were true, we would expect more distortion against states in the Deep South. In these states, voting is highly Democrats in heavily urbanized states. Consistent with racial polarized and, unlike most of the rest of the nation, Table 1 and Model 1, a larger Hispanic population reduces much of the African-American population is rural. In addi- bias against Democrats, but the size of the African- tion, amendments to the Voting Rights Act (VRA) have American population has no significant effect on distortion, been interpreted to require the drawing of African- and we see no effect for urbanization. American-majority or African-American-influence dis- Model 3, including only states with more than six dis- tricts across rural parts of these states, with district maps tricts, paints a different picture, showing a significant nega- requiring Department of Justice preclearance under the tive coefficient for urbanization. Among larger states, VRA. Past research has suggested that this may constrain which likely include both urban and rural areas, heavily maps to resemble Republican gerrymanders even when urbanized states (e.g. New Jersey and Pennsylvania) are drawn by another party (Goedert, 2012; Hill, 1995; Lublin, more often heavily distorted against Democrats than more 1999), and we do see that results in these states are slightly rural states (e.g. Minnesota and Wisconsin) after 6 Research and Politics Table 2. Regression results. Democrat % seats won Hisp. dum minus % seats expected Model 1 Model 2 >6 CDs Democratic gerrymander 9.13* 10.1** 16.6*** (4.63) (4.79) (4.75) Republican gerrymander −11.2*** −4.08 −13.6** (2.89) (3.81) (4.86) Percent African-American – −0.41 −0.29 (0.26) (0.24) Percent Hispanic – 0.58** 0.77*** (0.22) (0.24) Urbanization – 0.046 −0.72** (0.22) (0.34) Democratic vote – 0.32 0.11 (0.21) (0.24) Number of seats – −0.29* −0.16 (0.16) (0.18) Hispanic dummy 9.95*** – – (3.11) Constant −5.52** −25.5 45.0 (2.26) (15.8) (29.2) Observations 33 33 21 R-squared 0.557 0.641 0.829 Notes: Standard errors in parentheses. Data points weighted by state size. ***p < 0.01, **p < 0.05, *p < 0.10. Table 3. Seats won versus mean vote share by gerrymandering party: 2012 presidential vote (summary). Summary table Dem. Dem. Dem. seats Won–Exp. CDs Vote share Seats won Expected Difference Republican gerrymanders 106 50% 28% 50% −22% Democratic gerrymanders 35 61% 80% 72% 8% Bipartisan or court gerrymanders 97 57% 63% 64% −1% High Hispanic population 105 55% 57% 59% −2% Deep South states 38 43% 21% 36% −15% controlling for the gerrymandering party. Furthermore, the districts, further suggesting pro-Republican bias. We can coefficients for partisan maps increase when we limit the substitute Obama’s margin for the congressional election sample to larger states, possibly indicating the greater flex- result to measure bias under the various redistricting regimes. ibility parties have in drawing districts in such states. The results of replicating Table 1 using presidential elec- tion results are summarized in Table 3 and are detailed in Table A3 of Supplementary Material. In the case of partisan Robustness check: Presidential maps and heavily Hispanic states, the average bias is very election results similar to the bias under the actual congressional election Although the current congressional map has thus far only results. Notably, the difference in bias between Republican seen one cycle of election results, there has been another and Democratic gerrymanders remains the same at 14%. election held across all 435 of these districts that we can use However, the pro-Republican bias under bipartisan and court to test the robustness of this paper’s finding: the 2012 presi- gerrymanders largely disappears. There are likely two expla- dential election. Despite winning with 52.0% of the two- nations for this difference. Firstly, President Obama won party popular vote, Obama won only 209 congressional three districts in Minnesota and five districts in New York Goedert 7 Declaration of conflicting interest with 52% or less of the vote, which might be described as luck. However, this result also suggests that the asymmetry The author declares that there is no conflict of interest. in the geographic distribution of partisans is not constant across states and regions. In some “bluish” states, the more Funding conservative areas such as upstate New York and rural This research received no specific grant from any funding agency Minnesota may be only marginally Republican. These dis- in the public, commercial, or not-for-profit sectors. tricts may be won by Republicans in a nationally tied elec- toral environment but captured by Democrats in a climate Notes somewhat more favorable to them, such as Obama’s 4% 1. While Hirsch argues that the combination of redistricting popular vote victory. In contrast, in the Deep South where the control and geographic imbalance biased the 2002 election more conservative regions are deeper “red,” probably exag- results against the Democrats by 25 seats, he does not dis- gerated by VRA considerations, the bias against Democrats tinguish between these two factors in that estimate, and he is actually exacerbated as their vote majority increases. argues that almost all bias can be located within four states with Republican-controlled districting. 2. The linear method estimates an average slope of 2.02 for Conclusion the past 40 years, compared with Tufte’s (1973) average of 2.09 for the preceding 70 years. Tufte justifies using a linear Both the state-by-state results and aggregated regression estimate, as opposed to probit or logit, because the major- analysis suggest that while deliberate partisan gerrymander- party vote shares rarely fall outside of the 35–65% range. ing produces additional seats for the districting party, parti- However, as vote shares in several states in the 2012 election san gerrymandering is not a sufficient explanation for the fall outside of this range, a curve that will deal more appro- overall antimajoritarian outcome. Instead, pro-Republican priately with extreme values is needed for our purpose. bias is observed under all districting regimes. In addition, 3. The constant in this regression represents approximately a 3–4% bias in favor of Democrats over this period. When bro- the regression results offer possible support for the Chen ken down by decade (shown in Table A1 of Supplementary and Rodden (2013) thesis that urbanization has created bias Material), the bias estimate aligns with past research in show- while also forecasting its possible demise if patterns of ing Democratic bias in the 1970s and 1980s, shifting toward rapid Hispanic population growth continue. Republican bias in the 2000s (e.g. King and Gelman, 1991; It is important to note the limits to these conclusions. McGhee, 2012), possibly due to the same gerrymandering Firstly, while asymmetric population distributions are a plausi- and geography trends observed here for 2012. The sign of ble explanation for persistent bias, and one supported by previ- this bias is reversed under the state elections data method ous research, they are not the only possible cause. For example, (also in Table A1 of Supplementary Material), with the dif- one might claim that incumbency could give Republicans ference likely attributable to the method of imputation for advantages in more marginal districts (see McGhee, 2012). unopposed races. If estimated using a logit function on the This article does not attempt to isolate that cause. national data, the slope coefficient is .0415, with all results substantive unchanged. This analysis does not imply that Democrats are 4. In cases where a candidate runs unopposed and no votes are doomed to the minority even for the next decade. It does collected, no votes are added to the national total, but a 100% indicate they are unlikely to retake the House in an essen- vote share is imputed into the state result. This will lead to tially tied national election. Yet national elections are not a difference in responsiveness between the methods where usually this close: Democrats reversed a Republican ger- unopposed incumbents are predominantly one party. About rymander in Pennsylvania, Virginia, Ohio, and Michigan 70% of such races in the data set occur in the South, with in 2006 or 2008 (all states with aggressive Republican about two-thirds of those being Democrats in the 1970s and maps). The 2012 maps leave the Democratic Party several 1980s. openings; for example, Republicans now sit in five 5. Obviously, Democrats could not have hoped to perform bet- Pennsylvania districts won by Obama in 2008. To win ter in Massachusetts than they did. At the state level, how- these seats, Democrats will need the electorate to look like ever, this example illustrates the national phenomenon of Democrats failing to maximize their vote by oversaturating 2006 or 2008, but this is far from unprecedented: their support in certain areas. In addition, Democrats con- Democrats won the popular vote by at least 5 points in 12 trolled the process in Arkansas, but won none of its four of the last 20 cycles. But given the unequal concentra- seats; earning an average of 35% of the vote across this state tions of vote share in most states, not just those with would have predicted winning one seat under a “fair” map. Republican gerrymanders, a Democratic majority will be 6. This average disparity is extremely close to the 6% dispar- a bit more difficult than it should be. ity observed nationwide, as the Democrats’ 1% popular vote advantage is estimated to correspond to 52% of seats Supplementary Material expected, compared to 46% of seats actually won. 7. Reducing the Republican bias by 7% in the 238 seats under The entire Supplementary Material is available at: http://bit. Republican, Democratic, or Bipartisan control in Table 1 nets ly/1jOtnma 8 Research and Politics the Democrats (238 × .07) = 16.7 seats. If we instead assume Goedert N (2012) Southern redistricting under the VRA: A the level of bias shown in Table 1, but allocate 70 seats to both model of partisan tides. In: the 2012 state politics & policy Democratic and Republican control (rather than 35 and 106, conference, Houston, TX. Available at: http:// 2012sppcon respectively), this reduces the number of Republican seats in ference.blogs.rice.edu/files/2012/02/VRA-Partisan-Tides- Republican-controlled maps by (36 × .19) = 6.8 seats, and SPPC-version1.pdf. increases the number of Democrats in Democratic-controlled Hill KA (1995) Does the creation of majority black districts aid maps by (35 × .05) = 1.75 seats (for a total of 8.6 seats). Republicans? An analysis of the 1992 elections in eight 8. The exception here is Georgia, which is biased toward southern states. Journal of Politics 57(2): 384–401. the Democrats despite being districted in 2011 by Hirsch S (2003) The United States House of Unrepresentatives: Republicans. This is likely attributable to the novel What went wrong in the latest round of congressional redis- strategy of “minority influence” districts employed in a tricting. Election Law Journal 2(2): 179–216. Democratic gerrymander in the 2000s, a strategy upheld King G and Browning RX (1987) Democratic representation and in Georgia v. Ashcroft (2003), combined with the need to partisan bias in congressional elections. American Political avoid retrogression from this map to achieve VRA clear- Science Review 81(4): 1251–1273. ance in the next decade. King G and Gelman A (1991) Systemic consequences of incum- 9. Because of the inclusion of other controls with continuous bency advantage in U.S. House Elections. American Journal values in Models 2 and 3, the value of the constant is no of Political Science 35(1): 110–138. longer inherently meaningful. Lijphart A (1999) Patterns of Democracy: Government Forms 10. The coefficients for Republican gerrymanders between mod- and Performance in Thirty-Six Countries. New Haven, CT: els are different at p < .05, but not significantly different for Yale University Press. Democratic gerrymanders. Lublin D (1999) The Paradox of Representation. Princeton, NJ: 11. This explanation seems less plausible given that Mitt Princeton University Press. Romney won 52% of congressional districts despite losing McDonald M (2009) Seats to votes ratios in the United States. the national popular vote by 4%. Unpublished Paper, George Mason University. Available at: http://www.democracy.uci.edu/files/democracy/docs/con ferences/mcdonaldgmu.doc (accessed 31 March 2014). References McGhee E (2012) The history of the House votes-seats dis- Altman M and McDonald MP (2013) A half-century of Virginia crepancy, in two graphs. The Monkey Cage (Political redistricting battles: Shifting from rural malapportionment to Science blog). Available at: http://themonkeycage.org/ voting rights to public participation. University of Richmond blog/2012/11/19/the-history-of-the-house-votes-seats-dis Law Review 47: 771–831. crepancy-in-two-graphs/ (accessed 31 March 2014). Brunell TL (1999) Partisan bias in U.S. congressional elections, Rae D (1967) The Political Consequences of Electoral Laws. New 1952–1996. American Politics Research 27(3): 316–337. Haven, CT: Yale University Press. Chen J and Rodden J (2013) Unintentional gerrymandering: Political Taagepera R and Shugart MS (1989) Seats & Voters: The Effects geography and electoral bias in legislatures. Quarterly Journal & Determinants of Electoral Systems. New Haven, CT: Yale of Political Science 8(3): 239–269. University. Erikson R (1972) Malapportionment, gerrymandering, and party Tufte ER (1973) The relationship between seats and votes in two- fortunes in congressional elections. American Political party systems. The American Political Science Review 67(2): Science Review 66(4): 1234–1245. 540–554.

Journal

Research & Politics – SAGE

Published: Apr 1, 2014

Keywords: Congress; legislative elections; gerrymandering; 2012 American elections

There are no references for this article.