The Pseudo-Paradox of Partisan Mapmaking and Congressional ... · competitiveness is explained by...
Transcript of The Pseudo-Paradox of Partisan Mapmaking and Congressional ... · competitiveness is explained by...
The Pseudo-Paradox of Partisan Mapmaking and Congressional Competition
Nicholas Goedert Washington University in St. Louis
November 2013
ABSTRACT: This paper uses empirical evidence from four decades of congressional elections to examine how redistricting institutions influence congressional seat competition under changing partisan tides. In particular, the paper finds that partisan gerrymanders induce greater competitiveness as national tides increases, largely due to unanticipated consequences of waves adverse to the map-drawing party. These results explain the “pseudo-paradox” that less competition in the national congressional popular vote historically predicts greater competition in individual local races. In contrast, bipartisan maps are shown to induce lower competition, and nonpartisan maps higher competition, under all electoral conditions and competitiveness measures.
Word Count: 6678 (all-inclusive); 5197 (body text only)
1
The link between electoral competition and accountability has been a focus of both public
and academic debate since the founding of democracy. In the United States in recent years,
much of this debate has targeted congressional elections, with good-government groups and the
media worrying about the overwhelming number of seemingly entrenched incumbents, and
scholars of American politics trying to explain why the number of close elections seemed to be
low and declining. This paper addresses one common explanation for declining competition,
gerrymandering, but with a less common focus: maps drawn for partisan advantage, specifically
in interaction with national partisan tides. We find that partisan gerrymandering explains both
variation in congressional competition, and the perverse connection between national partisan
balance and the prevalence of close elections at the district level.
Much of this concern about noncompetitive elections and entrenched incumbency has
highlighted the recent period in the late 1990’s and early 2000’s when competition appears
historically low. But this period was immediately followed by three consecutive elections from
2006 to 2010 where a significant number of incumbents were defeated, and the make-up of the
U.S. Congress changed dramatically. And it does not seem like the research has caught up with
this era, either by arguing that these elections were a fluke, or that the overall pattern of reduced
competition has now been reversed.
This paper argues that partisan gerrymandering has had a profound effect on
congressional competition, both during times of close partisan balance and times when one party
is ascendant, but that this effect can only be observed in interaction with these short-term
national electoral trends. Specifically, we find evidence that trends in the competitiveness of
congressional races have been largely driven by “backlash” responses of partisan gerrymanders
to national wave elections. This pattern among partisan maps also generates a somewhat
2
paradoxical inverse relationship between national and district competition, where greater parity
between parties at the national level correlates with reduced competition at the local level in
congressional campaigns. While the electoral effects of bipartisan gerrymanders and nonpartisan
commission maps tend to be resilient to tides (with nonpartisan maps generating consistently
more close races than bipartisan ones), partisan maps suppress competition when the national
electoral environment is closely balanced, but incite it when one party wins a substantial national
majority, largely due to members of the map-drawing party facing unexpectedly close races.
The argument of the paper proceeds as follows. First, in accordance with conventional
wisdom and past research, we show that both bipartisan and partisan-controlled legislatures draw
fewer competitive districts than nonpartisan commissions. However, we next show that a
previously unacknowledged factor, national tides, strongly and negatively correlates with
competitive elections. Finally, we demonstrate how this interaction of national and district-level
competitiveness is explained by focusing on partisan gerrymanders. Specifically, partisan
gerrymanders increase close elections during wave cycles because of unanticipated competition
in states with maps drawn by a party suffering through adverse tides. This argument is advanced
through an analysis of various measures of competition in a data set encompassing the past forty
years of congressional elections.
Previous Research on Districting and Competition
Declining competition in U.S. congressional elections has worried scholars since David
Mayhew’s seminal article “The Case of the Vanishing Marginals” (1974). Mayhew does not
arrive at an explanation for the paucity of close elections and districts that he observes, but the
role of redistricting in fostering or suppressing such competition has been controversial
3
throughout subsequent work. Tufte (1973) argues that redistricting was the cause for a reduction
in marginal seats during the 1960’s. However, later studies concluded that redistricting, whether
partisan or bipartisan, has had little effect on the competitiveness of seats or the advantages of
incumbency (Glazer et al. 1987, Ferejohn 1977). Moreover, Gelman and King (1994) find that
temporally proximate redistricting, both partisan and bipartisan, leads to an increase in electoral
responsiveness in state elections, measured by the slope of the seats/votes curve. And Gopoian
and West (1984) suggest that many partisan maps in the 1980’s appear to have reduced the
security of their incumbents, from an analysis of vote margins in two electoral cycles.
Scholars took up this topic in perhaps the greatest force in the mid-2000’s, when
competitive elections were are their lowest ebb. Most commonly, research points to bipartisan
legislative agreements as reducing competition and nonpartisan commissions as encouraging it.
Carson and Crespin (2004) find evidence that courts and commissions increase competition in
years immediately following redistricting (using data from 1992 and 2002). Lindgren and
Southwell (2013) also argue that independent commissions did reduce average margin of victory
in U.S. house elections from 2002-2010. But neither of these articles distinguishes between
bipartisan and partisan legislative maps. More generally, Cain et al. (2005) find that an overall
decline in competitive elections has tracked with an increase in the number of districts drawn by
bipartisan agreement.
Yet other recent works claim much less of a role for districting in contributing to this
trend. In an examination of state legislative elections in 2000 to 2008 (and using the same
definition of competitive elections as this paper), Masket et al. (2012) find little impact of
redistricting institutions on the likelihood of a close election. And Abramowitz et al. (2006) find
that while much of the decline in competition since the 1970’s can be attributed to the increased
4
partisan polarization of districts, this is explained much more by population sorting than
intentional districting.
And the specific role of partisan districting is inducing or suppressing competition is
perhaps even more ambiguous. Hirsch (2003) focuses on Republican-drawn maps as causing
historically low competiveness in the 2002 elections. But employing data from the same election
cycle, Yoshinaka and Murphy (2011) find that partisan maps increase competition specifically in
years immediately following redistricting, but constrain their explanation to the deliberately
increased difficulty that out-party members face, rather than unanticipated close races faced by
in-party members in the face of unexpected waves.
Yet the fact that partisan gerrymanders do not always turn out as planned for the
controlling party has not gone unnoticed by scholars. Grofman and Brunell (2005), in a series of
short case studies referred to as “dummymanders”, show that many maps drawn by Southern
Democrats in the 1990’s failed to anticipate trends favoring the Republican party. Conversely,
Seabrook (2010) argues that the effects of Republican partisan maps were largely washed out by
mid-decade partisan trends in the 2000’s. However, the larger effect of “dummymanders” on
competition, and the frequency of the their occurrence, is largely yet unexamined beyond
anecdotal evidence and narrow case studies.
Why Partisan Gerrymanders?
It is easy to see why scholars have focused on contrasting bipartisan and nonpartisan
maps in their study of gerrymandering and competition; the incentives and constraints of these
institutions make it obvious what the hypothesized effect on competition should be. In the case
of commissions, it is sometimes true that the creation of balanced districts is a stated goal of
5
nonpartisan actors: the Arizona Proposition creating that state’s districting commission requires
“competitive districts are to be favored” as long as they comply with other constitutional
requirements (Adams 2005, Kang 2004). In other cases, it is a side effect of other provisions: the
Legislative Services Bureau responsible for districting in Iowa is not permitted to incorporate
incumbency or voting data in creating their proposals. On the flip side, without the added
incentive of maximizing one party’s seats, the uniform desire to protect incumbents of both
parties in the case of bipartisan maps leads in many cases to districts clearly drawn to be
noncompetitive. Figure 1 below acutely exemplifies this in the case of the 2002 California map.
But the expected effect of partisan maps on competition is less obvious; the desires of a
party to maximize seats and simultaneously protect their own incumbents work at cross-
purposes, and it is up to the individual map-maker to choose how much emphasis to place on
each goal. Cain (1985) shows how partisan mapmakers “packed and cracked” opposing party
members into a small number of safe districts, and this has become the standard strategy for
partisan gerrymanders. Yet, a broad array of tactics are available within this strategy. If too
many opposing districts are “cracked”, mapmakers risk spreading their own party’s voters too
thin, leaving many of their own incumbents at risk if political tides shift slightly against them.
But if partisans fortify all their own districts safely enough to withstand an unlikely adverse
wave, they fail to contest seats that could have been won under more typical electoral
environments. Additionally, Niemi and Deegan (1978) demonstrate formally how neutrality and
competitiveness may be at odds when the overall vote is not evenly split, even when a fair map is
desired.
The range of “aggressiveness” (i.e. willingness to risk their own party’s seats to
maximize expected seats) that partisan mapmakers chose in the 2000’s is also shown in Figure 1
6
below, which gives a visual representation of three Republican-controlled gerrymanders
(Pennsylvania, Ohio and Florida) and one bipartisan gerrymander (California) used in the 2002
elections. The x-axis in these density plot figures represents how heavily-Republican a
congressional district is (measured by Cook PVI), while the y-axis represents the share of
congressional districts in each state with that level of partisanship (PVI’s more extreme than 20
or -20 are coded as 20 and -20 respectively).
We immediately see the stark difference between the partisan maps and the bipartisan
one. The three Republican maps have a clear slightly pro-Republican peak, accompanied by
several extremely safe Democratic districts (usually majority African-American). In contrast,
California has no districts in the middle of the graph (indicating districts that would be
competitive at the presidential level), and a clear bimodal distribution of both strongly
Republican and strongly Democratic districts. In this map, we would expect very few
competitive elections unless tides in favor of one party or the other were almost historically
extremely.
[Figure 1 about here]
But although each Republican map seems to have more districts on the positive (pro-
Republican) side of 0, the aggressiveness of each map, and thus their propensity to withstand
changes in political climate, varies. In Pennsylvania, the bulk of congressional districts lie
between D+2 and R+4; Republicans in Pennsylvania drew several swing districts that they were
counting on factors like incumbency and continued close national elections to hold. Ohio, with
its peak around R+2, was also very aggressive, but slightly less so than Pennsylvania. Florida,
with a peak around R+5, is a more moderate gerrymander, with many districts reinforced from
mild swings toward the Democratic party. The differences between these partisan maps help
7
explain why Republicans lost 9 of their 24 seats in Pennsylvania and Ohio in the 2006 and 2008
elections, but only 3 of 18 seats in Florida.
This paper hypothesizes that a typical strategy employed by partisan mapmakers,
exemplified above, will be to draw seats that will be largely safe for their own incumbents under
neutral electoral conditions (when the national popular vote is closely contested), but which will
become increasingly competitive as national tides adverse to the gerrymandering party increase.
Thus, the overall effect of partisan gerrymanders on competition can only be considered in
interaction with national competition.
Hypotheses
From past research on bipartisan and nonpartisan gerrymander, and anecdotal evidence on
partisan maps and occasional resulting “dummymaders” we would expect to see the following
interactions between national tides and redistricting institutions:1
• We expect low competitiveness in districts drawn by bipartisan agreement regardless of
electoral environment.
• We expect high competitiveness among districts drawn by nonpartisan commissions
when the national electoral environment is close. As national tides increase,
competitiveness in these districts may decline or stay steady.2
• When the national electoral environment is close, we expect low competitiveness among
districts drawn by legislatures controlled by one party. As national tides increase, we
expect competitiveness to increase, particularly when tides run adverse to the
gerrymandering party.
8
Measures & Controls
For the purpose of testing these predictions on competitiveness and partisan bias, we have
assembled a data set of all congressional elections falling on a national election day from 1972
(following the first national round of post-Wesberry redistricting) through 2010. The process by
which each state was redistricted at the start of the decade was coded as Democratic-controlled,
Republican-controlled, bipartisan, nonpartisan commission, or court.3 States with three or fewer
congressional districts are designated as “small states” and not otherwise coded. It is less feasible
to draw maps to serve partisan interests, or achieve other very particular goals, in states with very
few districts, particularly those with only one. Therefore, these small states, will serve as
controls against which to measure the effects of the regimes of interest. Maps drawn by courts
serve as a separate control for which we provide no particular directional predictions.4
This analysis attempts to isolate the effects of the redistricting institution and not the
effects of the specific districts drawn. Therefore, any controls that might be endogenous to the
actual districts have been deliberately omitted (with the exception of denoting open seats in one
specification for illustrative purposes). This of course includes district-specific demographics
and partisanship, but also candidate-specific data like campaign spending and incumbency.
Although all of these factors are of course important to the outcome of a congressional race, the
controls that are included (statewide presidential vote, region and statewide demographics, and
redistricting institution) are causally prior, and thus the exclusion of district-specific factors
should not contaminate the coefficients testing the theory.5 Errors are clustered by district
interacted with decade to account for serial autocorrelation within districts.
To get a complete picture of the effects of redistricting on competition, we employ four
measures of competitiveness: the competitiveness of (1) the state as a whole; (2) the
9
congressional district’s relative to the nation; (3) the national electorate in an individual election
year; and (4) the individual congressional race in a specific year and district.
The first of these measures is Statewide Competition, a measure of the competitiveness of
the state, with lower values indicating swing states and higher values indicating ideologically
extreme states. It is the absolute value Statewide Presidential Vote, defined the difference
between the average statewide GOP presidential vote margin and the average national GOP vote
margin over the previous two elections for a given district in a given year. Thus, Statewide
Presidential Vote is a rough measure of a state’s ideology in a given election cycle. E.g. for the
California 1st district in 2006, this variable takes a value of -13, because California voted 13%
more Democratic than the nation in the 2000 and 2004 presidential elections.
The second measure, District Level Presidential Competition, is the amount by which
presidential results in each district deviate from the national average. It is the analogous to
Statewide Competition (though scaled the same as Cook’s PVI), for the individual congressional
district, with lower values indicating swing districts at the national level. It is used as a
dependent variable in the next section.
The third competitiveness measure is National Tides, derived from National GOP Vote
Margin, the GOP margin in the nationwide congressional popular vote in a given cycle. For
example, the variable takes a value of -7.9 for all 2006 data points, because Democrats won the
national congressional popular vote that year by 7.9%. National Tides, the magnitude of tides
without regard to their direction, is the absolute value of National GOP Vote Margin, with
greater values indicating a stronger wave. Note that in our data set, eleven of the 20 elections in
the data set saw a national popular vote advantage for the Democrats of greater than 5%, only
two elections saw so great a national advantage for Republicans (1994 and 2010). So our ability
10
to test our predictions with respect to the interaction of Republican waves on Democratic
gerrymanders will largely be limited to the anecdotal.
The final measure, Close Race, is a dummy variable coded as 1 if an individual
congressional election was won by less than 10 points.6 As we are mainly interested in how
often congressional elections are competitive, this is the dependent variable throughout most of
the analysis. Overall, 14% of congressional elections are Close Races under this definition.
Demographic Competitiveness
In response to the scholarship on the competition created by legislative versus non-
partisan drawn maps, we might first ask whether gerrymandering institutions influence the
inherent partisan balance of districts, independent of waves in individual election cycles. To do
this, we regress District Level Presidential Competition, for all districts immediately following a
national districting year from the 1970’s through the 2000’s, on various gerrymandering controls.
We also include a control for the overall partisanship of the state, Statewide Competition, since
this might influence both the districting institution and the resultant districts. Lower values for
both “competition” variables indicate a swing state or swing district.
The results of this analysis are shown below in Table 1. The first column isolates
nonpartisan institutions only, and its coefficient is negative and significant, indicating that
nonpartisan maps indeed tend to draw more swing districts. Additionally, the positive
coefficient for Statewide Competition indicates that districts with closer partisan balance tend to
occur more in swing states. But controlling for state-level partisanship, nonpartisan commissions
do also draw more balanced districts. The second column also includes controls for Democratic,
Republican, bipartisan, and court-drawn maps (with small states as the excluded category).
Under this specification, Republican, bipartisan, and court-drawn maps all create fewer swing
11
districts that the control or nonpartisan maps; curiously, Democratic maps show no difference
from the control. However, this is likely an artifact of the global trends in the South during the
first half of this period, during which the region tended to vote consistently more Republican at
the presidential level than the congressional level, and had almost universally Democratic state
legislatures. This allowed Southern Democrats to draw many districts that would appear very
competitive when looking at presidential vote (how Statewide Competition is measured), but
were de facto safe Democratic seats at the congressional level. We can see that when the South
is excluded from the analysis in column 3 of Table 1, Democratic maps appear to draw just as
few swing districts as Republican and bipartisan maps.7 From this perspective, the tendency of
past literature to group partisan and nonpartisan gerrymanders together seems appropriate, and
that both do indeed produce fewer competitive districts. Conversely, the reforms intended to
produce more competitive districts under nonpartisan regimes, such as the commission in
Arizona or the bureaucracy in Iowa, have succeeded.
[Table 1 about here]
The Pseudoparadox of Competition
As discussed above, many of the recent claims about declining congressional
competition, in both the media and the literature, occurred in the wake of an era of parity at the
national level. Between 1996 and 2004, no party won a majority of the national popular vote in
congressional elections, nor did any party win that popular vote by more than 5 points, or win
more than 54% of congressional seats. This is actually something of a historical anomaly. Of
the twenty election cycles from 1972 to 2010, one party failed to win by at least 5-points in the
popular vote only seven times, included these five consecutive years. And during a time when
the country appeared so evenly divided, it would be intuitive to expect that many individual races
12
would also be close, but less than 11% of races during this era were decided by 10 points or less
(our definition of Close Races).
But this era of parity was immediately followed by three consecutive “wave” elections
from 2006 through 2010. And despite the national electorate clearly favoring one party in these
three years, the number of race that were closely contested rose to 15%. Recently, it does not
seem that close national competition has lead to greater competitiveness at the local level, and
this phenomenon is born out looking further into the past; this trend extends back at least as long
as the equal-population standard has been applied to congressional districts. Figure 2 below
shows the correlation of the proportions of Close Races in each cycle 1972-2010 with National
Tides; as the national popular vote gets closer, the number of competitive races tends to decline.
[Figure 2 about here]
As a more rigorous test, Table 2 below shows the effect of National Tides on the
percentage of Close Races (each data point is an individual race, with races clustered by state
crossed with decade) from 1972-2010, when we control for the competitiveness of individual
states by including Statewide Competition. In the first column, we see that the competiveness of
a state has very little effect on its propensity toward competitive congressional elections.8
However, the positive coefficient on National Tides (significant at p<.01), indicates that extreme
wave elections do tend to create more close races. The second column excludes races in the
South, with no effect on the National Tides coefficient, indicating the phenomenon cannot
merely be explained by increased competitiveness of the Republican party in the South in the
most recent decades, or the creation of majority-minority districts in those states. The third
column addresses the argument that competition has steadily declined over time, by including a
Year variable, as well as the argument that competition increases in years immediately following
13
redistricting (e.g. Yoshinaka and Murphy (2011); Hetherington et al. (2003)); the Gerrymander
Year dummy variable takes a value of 1 for election years ending in “2”. The coefficients for
these variables are slightly in the expected direction, but neither is significant, and neither
mediates the effect of National Tides on the number of close elections. Finally, we see in the
Figure 2 that 1974 appears to be an outlier both in terms of wave strength and number of close
races. The fourth column of Table 2 shows that the effect of National Tides on Close Races is
still significant at p<.05 even when this cycles is excluded.9
[Table 2 about here] Effects of Gerrymandering
So what explains this “pseudo-paradox” that less national competitiveness correlates with
greater local competition in congressional races? Figure 3 below shows that, far from being a
universal phenomenon, the pseudo-paradox appears limited only to states with partisan
gerrymanders. When we isolate only the partisan maps, the negative effect of national
competitiveness on local competitiveness is strengthened (and significant at p<.02). But the
magnitude of National Tides has no effect on competitive elections under bipartisan maps, which
follows if these maps drew districts safe enough for both parties to withstand strong tides in
either direction. Moreover, the coefficient for the nonpartisan maps is in the opposite direction
of partisan maps, although not significant due to the high variance from the small sample size;
this would also follow if such maps tended to draw many “naturally” competitive districts.
[Figure 3 about here]
Yet it is also possible that these observed differences are merely the result of the types of
states that tend to adopt these varying institutions. I.e perhaps swing states tend to adopt
nonpartisan regimes or extreme states adopt bipartisan regimes. So we also run a probit analysis
14
of Close Race on National Tides and the various gerrymander dummies including the Statewide
Competition control. We also assess the slope of the effect of tides by interacting National Tides
with the gerrymander dummies. As before, the unit of analysis in the first four columns Table 3
below is individual congressional races from 1972-2010, clustered by year crossed with decade.
The last two columns employ OLS on units of statewide means as described in Footnote 5, with
very similar results.
[Table 3 about here]
The first specification does not include any controls for redistricting institutions. For the
second and third specifications, the excluded category is the small states. The third specification
includes a dummy variable for whether an election was an open seat; although this is potentially
endogenous to the gerrymander, we include it merely to show that the results remain the same
with its inclusion, and thus retirements do not explain the results with respect to other variables.
Open seats do drastically increase competition (34% of open seat races are close, compared to
11% of races including an incumbent), but do not wipe out the effects of national tides or the
differences between redistricting regimes. Figure 4 below shows the probit coefficients for the
redistricting variables interpreted for neutral state and national electoral conditions.
[Figure 4 about here]
From the coefficients in all specifications, we see that bipartisan maps and partisan maps (from
both parties) create fewer close elections than elections in small states (the excluded category),
when controlling for state ideology and national tides. Conversely, nonpartisan commission
maps create more competitive elections, with an effect size large enough to be significant despite
the small sample size. From the figure, this analysis estimates that while only 8%-11% of
elections will be close races under neutral electoral conditions under partisan and bipartisan
15
maps, 21% of races will be close under nonpartisan commission maps.
Note that in all these specifications, the effect of National Tides is positive and
significant. We can test how this effect varies under different gerrymanders in two ways. First,
we can include interaction terms nested into the model run on the entire data set. This is shown
in column 4 of Table 3, which adds the interactions of each regime with tides. Consistent with
Figure 3, we see that the “pseudo-paradox” effect of tides is again largely explained by
gerrymandering institutions. When applied to bipartisan and nonpartisan gerrymanders, the
effect of National Tides is not significant, but it is still positive and significant when applied to
Democratic and Republican-drawn maps.10
As second way of analyzing this interacted effect would be to measure the effect of tides
on the data subsetted by redistricting regime. Table 4 shows the slope coefficient for the National
Tides variable when the probit from the first column of Table 3 is run only on subsets of districts
drawn under a particular institution. Again, the results conform to our predictions. We see a
large positive coefficient Republican partisan gerrymanders, (and a smaller, yet still significant,
one for Democratic maps) suggesting that partisans draw maps to protect their own seats
assuming a neutral environment, but suffer a backlash when tides go against them, leading to
many close races as national tides increase. In contrast, bipartisan gerrymanders, with safe
districts are drawn so as to be resilient to partisan tides, show no significant effect of tides, while
nonpartisan commissions show a negative but insignificant coefficient due to small sample size.
The difference between these coefficients for partisan and bipartisan maps is significant at p<.02.
In support of our hypotheses, competition at the district level increases as tides increase under
partisan maps, but not bipartisan or nonpartisan maps, controlling for statewide partisanship.
[Table 4 about here]
16
Differences by Party
The results above suggests that partisan maps induce greater competition as national waves
increase, but the tell us nothing about the direction of that wave. Yet we hypothesize that we
should only observe greater competition under waves adverse to the map drawing party (i.e.
Republican maps under Democratic waves such as 1974 and 2008, and Democratic maps under
Republican waves like 1994). And specifically, we would hypothesize that the slope of the
National Tides coefficient for the subset of Democratic-drawn maps would be greater when
Republicans with the national popular vote than when Democrats win (and conversely for
Republican maps). Table 5 below confirms this hypothesis for Democratic maps, with the
difference in responsiveness to tides under Democratic and Republican waves for Democratic
gerrymanders significant at p<.01. Unfortunately, performing the same analysis for Republican
maps would involve drawing conclusions based on a very small number of elections11; the
difference in responsiveness by wave direction for Republican maps is not significant.
[Table 5 about here]
[Table 6 about here]
Although the sample size of elections, particularly Republican-wave elections, limits our
ability to demonstrate the different responses to waves under Democratic and Republican
regimes in a statistically robust way, we can at least observe these differences anecdotally. Table
6 depicts the percentage of close races in states with Democratic maps in the 1990’s and 2000’s,
and in states with Republican maps in the 1970’s and 2000’s.12 Elections with strong opposing
tides, where we would hypothesize a greater number of close races, are shown in italics.
17
With one exception, the results conform to our expectations. In both the 1990’s and the
2000’s, Democratic maps saw more close races in the two Republican wave elections (1994 and
2010), than other years during their respective decades. Republican maps also saw the greatest
number of close elections in post-Watergate election of 1974, and many more close election in
the Democratic waves of 2006 and 2008 than the previous two cycles. There were a high
number of wave elections under Republican maps during the 2010 Republican wave, likely due
to an number of races won by Democrats in 2008 turning flipping back to the GOP in those
states.13 Contrary to Yoshinaka and Murphy (2011), partisan gerrymanders do not increase
competition in years immediately following redistricting, but rather in years when tides turn
against the party in control.
Conclusion
The results in this paper provide evidence of both the limitations of partisan
gerrymanders and the effectiveness of nonpartisan reforms, results derived even in the absence of
information about what lines were actually drawn and how those lines specifically affected the
resulting campaigns. As hypothesized, seats gerrymandered by nonpartisan commissions appear
significantly more competitive than those drawn by bipartisan legislatures, whether
competiveness is measured by district partisan demographics or by vote margins in congressional
elections. Additionally, while partisan maps appear noncompetitive at the demographic level,
they generate greater competition in actual elections as national tides increase. Further, the
“pseudo-paradox” of greater national competition inducing less local competition, specifically
under partisan maps, suggests that partisan mapmakers are drawing lines to create safe seats
18
particularly in anticipation of close national conditions, somewhat oblivious to or discounting of
the possibility of uncertain future tides.
This work fits into a broader framework aimed at assessing the impact of gerrymandering
institutions, particularly moving toward bipartisan commissions, on voter welfare under different
representational interests. While the most recent election results may depict partisan
gerrymandering at its most powerful and insidious, a wider historical view presents a somewhat
different picture. Thus, these results suggest why Democrats may have reason to be hopeful
about their chances to retake control of the House at some point during the next decade, despite
the counter-majoritarian outcome in 2012 in which Republicans lost the national popular vote
but won the majority of seats. Because the national vote was extremely close, we might expect
very few close races in states with partisan control of redistricting (mostly Republican control in
the current decade). However, these same maps might see a great deal of competition and
turnover if Democrats can win a decent popular vote margin in the future, suggested by the fact
that President Obama won five Pennsylvania districts currently held by Republicans in 2012.
Gerrymandering has historically worked very for the controlling party when the national vote
was evenly split, but led to greater and often unexpected competitiveness as the national margin
increased. There is no reason to expect these same trends will not continue in the foreseeable
future.
19
References
Abramowitz, Alan I., Brad Alexander, and Matthew Gunning. 2006. "Incumbency, redistricting,
and the decline of competition in US House elections." Journal of Politics 68(1): 75-88.
Adams, Florence P. 2005. “Minorities and Representation in the New Millennium.” In
Redistrciting in the New Millennium, 155-79. New York: Rowman & Littlefield.
Barone, Michael and Richard E. Cohen. The Almanac of American Politics. Washington:
National Journal Group, Bi-annual Publication.
Cain, Bruce E., Karin MacDonald and Michael McDonald. 2005 “From Equality to Fairness:
The Path of Political Reform since Baker v. Carr.” In Party Lines, 6-30, Thomas E.
Mann and Bruce E. Cain, ed. Washington: Brookings Institution.
Cain, Bruce E. 1985. “Assessing the Partisan Effects of Redistricting.” American Political
Science Review 79: 320-33.
Carson, James L. and Michael H. Crespin. 2004. “The Effect of State Redistricting Methods on
Electoral Competition in United States House Races.” State Politics and Policy
Quarterly 4(4): 455-69.
Ferejohn, John A. 1977. “On the Decline of Competitiveness of Congressional Elections.”
American Political Science Review 71: 166-76.
Gelman, Andrew and Gary King. 1994. “Enhancing Democracy through Legislative
Districting.” American Political Science Review 88(3): 541-559.
Glazer, Amihai, Bernard Grofman and Marc Robbins. 1987. “Partisan and Incumbency Effects
of 1970s Congressional Redistricting.” American Journal of Political Science 31(3):
680-707
20
Grofman, Bernard and Thomas L. Brunell. 2005. “The Art of the Dummymander: The Impact
of Recent Redistrictings on the Partisan Makeup of Southern House Seats.” In
Redistrciting in the New Millennium,183-200. New York: Rowman & Littlefield.
Gopoian, David L. and Darrell M. West. 1984. “Trading Security for Seats: Strategic
Considerations in the Redistricting Process.” The Journal of Politics 46(4): 1080-96.
Hirsch, Sam. 2003. “The United States House of Unrepresentatives: What Went Wrong in the
Latest Round of Congressional Redistricting.” Election Law Journal 2(2): 179-216.
Hetherington, M. J., Larson, B., & Globetti, S. 2003. “The redistricting cycle and strategic
candidate decisions in US House races.: Journal of Politics 65(4): 1221-1234.
Kang, Michael S. 2004 “The Bright Side of Partisan Gerrymandering.” Cornell Journal of Law
and Public Policy 14: 443.
Lindgren, Eric and Priscilla Southwell. 2013. “The Effect of Redistricting Commissions on
Electoral Competitiveness in U.S. House Elections, 2002-2010.” Journal of Politics and
Law 6(2): 13.
Mann, Thomas E., and Bruce E. Cain. 2005. Party Lines: Competition, Cartisanship, and
Congressional Redistricting. Washington: Brookings Institution.
Masket, Seth E., Jonathan Winburn, and Gerald C. Wright. 2012. “The gerrymanderers are
coming! Legislative redistricting won't affect competition or polarization much, no matter
who does it.” PS Political Science and Politics 45(1): 39.
Mayhew, David R. 1974. “Congressional Elections: The Case of the Vanishing Marginals.”
Polity 6(3): 295-317.
McDonald, Michael P. 2004. “A Comparative Analysis of Redistricting Institutions in the
United States, 2001-02.” State Politics and Policy Quarterly 4(4): 371-395.
21
Moxley, Warden. 1973. Congressional Districts in the 1970s. Washington: Congressional
Quarterly.
Niemi, Richard G. and John Deegan, Jr. 1978. “A Theory of Political Districting.” The
American Political Science Review 72(4): 1304-23.
Seabrook, Nicholas R. 2010. “The Limits of Partisan Gerrymandering: Looking Ahead to the
2010 Congressional Redistricting Cycle.” The Forum: 8(2), Article 8.
Tufte, Edward R. 1973. “The Relationship between Seats and Votes in Two-Party Systems.”
The American Political Science Review 67(2): 540-54.
Yoshinaka, Antoine and Chad Murphy. 2011. “The Paradox of Redistricting: How Partisan
Mapmakers Foster Competition but Disrupt Representation.” Political Research
Quarterly 64(2): 435-47.
22
Figure 1. Partisan Balance of Districts in Four States, 2002 (Density Plot)
23
Figure 2. Close Races by Year
24
Figure 3. Close Races by Year and Gerrymandering Institution
25
Figure 4. Estimated Probability of Competitive Race by Redistricting Institution in Swing State in Tied Popular Vote Election
Notes: Entries represent the estimated probability of a race being won by less than 10% under various redistricting institutions, taken from model 2 in Table 3, if Statewide Competition and National Tides are set to 0. Error bars represent the 95% confidence interval of the probit coefficient.
26
Table 1. Effects of Gerrymandering Regime on District Level Competition
No South
District Level Presidential Competition (1) (2) (3)
Statewide Competition 0.31*** 0.32*** 0.32***
(0.028) (0.028) (0.036)
Democratic Gerrymander - 0.43 2.16**
(0.55) (0.90)
Republican Gerrymander - 1.78*** 2.11**
(0.66) (0.79)
Bipartisan Gerrymander - 1.15* 1.83**
(0.61) (0.75)
Court Gerrymander - 1.43*** 2.16***
(0.50) (0.73)
Nonpartisan Gerrymander -2.47** -2.01 -1.76
(1.26) (1.28) (1.38)
Constant 6.74 5.44 4.87
(0.30) (0.58) (0.78)
Observations 1,740 1,740 1,237 R-squared 0.069 0.076 0.065 Notes: DV is the absolute value of a district’s deviation from the national average in previous presidential elections, with lower values indicating swing districts. ** = p<.05, *** = p<.01
27
Table 2. Probability of Close Race Controlling for Statewide and National Competitiveness
No South No 1974
Pr(Close Race) (1) (2) (3) (4)
Statewide Competition -0.003 0.000 -0.001 -0.003
(0.003) (0.003) (0.003) (0.003)
National Tides 0.019*** 0.017*** 0.017*** 0.013**
(0.005) (0.006) (0.005) (0.006)
Year - - -0.002 -
(0.002)
Redistricting Year - - 0.004 -
(0.041)
Constant -1.18 -1.16 3.32 -1.15
(0.048) (0.055) (4.37) (0.050)
Observations 8,700 6,185 8,700 8,265 Notes: Entries are probit coefficients. Standard errors clustered by district interacted with decade. * = p< .10, ** = p<.05, *** = p<.01
28
Table 3. Probability of Close Race controlling for Redistricting Institution and State and
National Electoral Trend (Non-South Congressional Races 1972-2010)
No South W/Open
OLS Statewide
Means
OLS Statewide
Means Pr(Close Race) (1) (2) (3) (4) (5) (6)
Statewide Competition -0.0024 0.00030 -0.0025 -0.0021 -0.00022 -0.00012
(0.0031) (0.0036) (0.0032) (0.0031) (0.00083) (0.00084)
National Tides 0.021*** 0.019*** 0.020*** 0.035*** 0.0077*** 0.013***
(0.0045) (0.0054) (0.0046) (0.012) (0.0013) (0.0032)
Democratic Gerrymander -0.21*** -0.25** -0.21*** -0.13 -0.063*** -0.022
(0.063) (0.099) (0.065) (0.11) (0.017) (0.027)
Republican Gerrymander -0.15** -0.11 -0.12* -0.25** -0.046** -0.028
(0.071) (0.081) (0.073) (0.13) (0.018) (0.026)
Bipartisan Gerrymander -0.18** -0.21** -0.15** 0.014 -0.052*** 0.0025
(0.071) (0.081) (0.073) (0.13) (0.018) (0.029)
Court Gerrymander -0.024 -0.032 -0.018 0.058 -0.0071 0.00079
(0.060) (0.080) (0.062) (0.10) (0.016) (0.023)
Nonpartisan Gerrymander 0.36*** 0.34** 0.37*** 0.60*** 0.094** 0.16***
(0.13) (0.13) (0.13) (0.17) (0.043) (0.034)
Democratic Gerry*Tides - - - -0.014 - -0.0069**
(0.013)
(0.0035)
Republican Gerry*Tides - - - 0.012 - -0.0035
(0.014)
(0.0035)
Bipartisan Gerry*Tides - - - -0.031** - -0.0092**
(0.015)
(0.0036)
Court Gerry*Tides - - - -0.012 - -0.0011
(0.012)
(0.0029)
Nonpartisan Gerry*Tides - - - -0.046* - -0.013*
(0.027)
(0.0078)
Open - - 0.78*** - - -
(0.045)
Constant -1.07 -1.05 -1.20 -1.16 0.17 0.13
(0.075) (0.091) (0.078) (0.11) (0.021) (0.027)
Observations 8,700 6,185 8,700 8,700 1,000 1,000 R-squared - - - - 0.071 0.080 Notes: Entries in columns 1-4 are probit coefficients. Standard errors clustered by district interacted with decade. * = p< .10, ** = p<.05, *** = p<.01
29
Table 4. Coefficient for effect of National Tides on Close Races by redistricting institution (Congressional Races 1972-2010)
Institution Coeff. SE n Democratic .017** (.008) 2633 Republican .041*** (.010) 1552 All Partisan .027*** (.007) 4185 Bipartisan -.001 (.010) 2246 Nonpartisan -.010 (.027) 210 Small State .024 (.015) 598 Overall .019*** (.005) 8700 Notes: Entries are probit coefficients for the National Tides variable taken from Model 1 of Table 3, run under a subset of the data for each redistricting institution. * = p<.10, ** = p<.05, *** = p<.01
Table 5. Coefficient for effect of National Tides on Close Races under Democratic Gerrymanders (Congressional Races 1972-2010)
Institution Coeff. SE n Democrats win Popular vote .012 (.009) 2029 Republicans win popular vote .092*** (.023) 604 All elections .017** (.008) 2633 Notes: Entries are probit coefficients for the National Tides variable taken from Model 1 of Table 3, run under a subset of the data for each Democratic gerrymanders. * = p<.10, ** = p<.05, *** = p<.01
30
Table 6. Percent Close Races by Year and Gerrymandering Party
Close Races in Seats Drawn by Democrats
National GOP Year
Vote Margin
Seats % Close
1992
-5.3
119 16% 1994
6.9
119 20%
1996
-0.3
119 14% 1998
0.9
119 8%
2000
0.3
119 7%
2002
4.6
108 11% 2004
2.6
76 8%
2006
-7.9
63 5% 2008
-10.5
63 6%
2010
6.6
63 14%
Close Races in Seats Drawn by Republicans
Nat'l GOP Year
Vote Margin
Seats % Close
1972
-5.3
119 18% 1974
-16.6
119 28%
1976
-10.8
119 17% 1978
-8.7
119 17%
1980
-2.7
119 12%
2002
4.6
92 7% 2004
2.6
124 6%
2006
-7.9
137 12% 2008
-10.5
137 12%
2010
6.6
137 15%
31
Endnotes 1 Authors (2012) employs a simulation model to generate these hypotheses in a more rigorous
manner. Figures summarizing the results of these simulations are included in the Supplemental
Appendix. All of these predictions are consistent with existing literature on the effects of
bipartisan and nonpartisan institutions, and the implications of “dummymandered” partisan
maps. 2 It is possible that if many “naturally” competitive districts are drawn, these districts may
become non-competitive if tides favoring one party are strong enough. But the number of
nonpartisan commission states in our data set may be too small to draw strong conclusions on
this question. 3 Codings, listed in Supplemental Appendix A1, are mostly drawn from CQ’s Congressional
Districts in the 1970’s and subsequent volumes in this series. They were compared with the
codings from research such as Glazer (1987) and McDonald (2004) for robustness and
supplemented by other sources where the descriptions are ambiguous. 4 A handful of states are coded with more than one designation. This occurs most frequently
where a map is chosen by a court from among those proposed by the political parties. In such
cases, the congressional districts in this state are designated as being drawn both by a court and
by the party of the map chosen. 5 Because controls are statewide measures, the data set essentially describes the proportion of a
state elections that are close, weighted by the size of the delegation. An alternative specification
would be to make each of these proportions a data point, weight them by delegation size (so that
the impact of each individual election was still the same), and cluster by state interacted with
decade (yielding 800 data points and 160 clusters). As the DV would now be a proportion
(rather than a binary outcome), yet not properly described by a binomial distribution, it is more
difficult to determine an appropriate specification for the data formatted in this manner. I have
included such an analysis, where the proportion of close races is analyzed using OLS in Table 3,
alongside the probit models with individual districts as data points; the results are very similar
across models. 6 A binary variable for close races in assessing competition is used frequently in the literature
(e.g. Masket et al. 2012, Carson & Crespin 2004), and is chosen here because using continuous
vote margin leads results to be very sensitive to imputation decisions in unopposed races. I have
32
also tested an alternate definition of Close Race that additionally includes all instances of
turnover, with very similar results, shown in the supplemental appendix. About 18% of all
elections are Close Races under this alternate definition. 7 South is defined as the former Confederacy plus Oklahoma. All analyses in this article have
been checked excluding this region with substantively identical results; all results available from
author. 8 This lies in stark contrast with Table 1 showing that swing states tend to draw more
demographically competitive districts (as opposed to district that generate close election). 9 As an alternative to the Close Race dummy, this analysis can also be run with the overall vote
margin as the dependent variable. Using this DV, the effect of National Tides is negative and
significant at p<.05 when unopposed races are excluded, and significant at p<.10 when
unopposed races are included and counted as a 100% vote margin. Full results available from
author. 10 The interpretation is produced by adding the negative interacted coefficients for bipartisan and
nonpartisan maps to the positive uninteracted coefficient, producing a nonsignificant effect.
When the nonsignificant interacted coefficients for partisan maps are added to the significant
uninteracted coefficient, the sum remains significant and positive. 11 Due to the lack of Republican waves prior to 1994, and the lack of Republican gerrymanders
in the 1990’s, this essentially only analyzes the difference between the 2010 wave and the
smaller Republican victories in 2002 and 2004. Results in Supplemental Appendix. 12 Data for Democratic maps for the 1970’s and 1980’s are excluded because those decades had
no Republican wave elections at the congressional level. Data for Republican maps for the
1980’s and 1990’s are excluded because very few states had Republican-drawn districts during
that period (less than 40 total districts in any year). The number of seats changes during the
2000’s due to mid-decade redistricting in Texas and Georgia. 13 This outlier suggests that competitiveness may correlate not just from absolute tides, but tides
in one election relative to tides in the previous election. Result from regression models are
robust to inclusion of an auto-regressive term for national tides in the previous election cycle
(available from author on request); although this term does have an independent significant
effect of competition, it has no effect on the on the interactions of tides and districting regimes.