803811

research-article20182018

RAP0010.1177/2053168018803811Research & PoliticsGoedert

Research Article

Democratic incumbent resilience
in the post-1980 Senate

Research and Politics
October-December 2018: 1­–9
© The Author(s) 2018
Article reuse guidelines:
sagepub.com/journals-permissions
https://doi.org/10.1177/2053168018803811
DOI: 10.1177/2053168018803811
journals.sagepub.com/home/rap

Nicholas Goedert

Abstract
In the period since the Reagan revolution that disrupted decades of Democratic control of the United States Senate,
Democratic senators have proved remarkably resilient when running for reelection. Over the past 600 Senate elections
since 1980, Republican incumbents have been defeated at more than twice the rate of Democrats, but won open seat
elections at a significantly higher rate. This partisan disparity also appears in incumbent vote share, and is statistically and
substantively robust to a range of model specifications accounting for existing theories. A series of explanations drawn
from existing research explored in this paper fail to explain this trend, suggesting the need for future research targeting
partisan differences in perceptions of parties in Senate campaigns.

Keywords
US Senate, legislative elections, US political parties, American political behavior

Introduction
On the surface, the immediate electoral future of the
Democratic Party in the United States Senate appears
dreary, with Democrats forced to defend seats in 2018 in
ten states won by Donald Trump in the preceding presidential election. Moreover, the allocation of two Senate seats
regardless of state population, giving disproportionate
power to small states, seems naturally stacked against the
Democratic party. But the uncanny resilience of incumbent
Senate Democrats in hostile territory in the past 36 years
suggests their prospects are not nearly so bleak. Indeed, following the “Reagan revolution” that swept Republicans
into office in 1980, Democrats have been defeated at only
half the rate of Republicans (8% vs. 16%), whereas
Republicans has been almost symmetrically successful in
winning 60% of open Senate contests regardless of the
incumbent party.
Given these apparent disparities between the parties, it
seems wise to include “party” as a crucial variable when
analyzing recent Senate election results, and to explore
what sort of underlying behavior might create these disparities. But elections to the US Senate have been subject
to far less study than either presidential elections or elections to the US House of Representatives. And while many
of these studies have focused on why Senate elections
appear much more competitive than House races, they

rarely incorporate party into the larger story. Incumbency
advantage has been found to be both smaller in the case of
Senate elections compared to House elections
(Ansolabehere and Snyder 2002), and declining in recent
years (Jacobson, 2015). As this note demonstrates, the difference in partisan success is not only observed in partisan
turnover, but also in overall vote share, and is robust to
model specification and inclusion of all controls typically
seen as accounting for Senate race competition. Thus, this
note does not attempt to argue for one specific explanation,
nor advance a new theory about partisan differences, but
merely broaches an important empirical finding suggesting the need for more targeted research on partisan differences in Senate elections.
In exploring the divergent success of the two parties, I
employ a data set of all Senate elections held on a national
general election day from 1982 to 2016. The data set is 611
data points, which amounts to six elections for each seat
plus a handful of special elections.1 This period can be
characterized by a high level of parity between the parties,

Virginia Tech, USA
Corresponding author:
Nicholas Goedert, Department of Political Science, Virginia Tech, 529
Major Williams Hall, 220 Stanger Street, Blacksburg, VA 24061, USA.
Email: ngoedert@vt.edu

Creative Commons Non Commercial CC BY-NC: This article is distributed under the terms of the Creative Commons
Attribution-NonCommercial 4.0 License (http://www.creativecommons.org/licenses/by-nc/4.0/) 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 (https://us.sagepub.com/en-us/nam/open-access-at-sage).

 2 

Research and Politics 

Figure 1.  Senate seat turnover percentage by time period, 1948–2016.
Table 1.  Senate turnover by party, 1982–2016.
Won by
 
Democratic incumbent
Republican incumbent
Democratic open seat
Republican open seat
Democratic-held total
Republican-held total

Democrat Republican Total Turnover %
219
39
37
14
256
53

19
210
30
43
49
253

238
249
67
57
305
306

8%
16%
45%
25%
16%
17%

with each party holding narrow majority control of the
Senate during nine of the 18 elections. In contrast,
Democrats often held large Senate majorities during the
previous post-WWII era (1946–1980), but were defeated
for reelection in similar numbers, shown in in Figure 1. The
whole sample consists of 305 seats held by Democrats
(including Democrat-leaning Independents), and 306 seats
held by Republicans. Democrats won 309 of these races,
and Republicans won 302. The distribution of open and
incumbent seats and the turnover rate in each category are
shown in Table 1. Both the difference in turnover between
Democratic incumbents and Republican incumbents and
the inverse disparity between Democrat-held open seats
and Republican-held open seats are statistically significant
at p < 0.01 and p < 0.02, respectively.

One might anticipate that this difference is merely due
to Republicans getting unlucky in marginal elections, with
no significant difference in the underlying distribution of
vote share across these elections. But Table 2 reveals that
Republicans incumbents have also significantly underperformed Democrats by three percentage points across the
last 36 years of elections (column 1; p < 0.01) controlling
for statewide partisanship. Moreover, this difference, in
both vote share and likelihood of turnover, remains robust
despite accounting for many theories used to explain
incumbent success in past studies, such as regional differences, strategic retirements, fundraising, and challenger
quality.2 Each theory is discussed below with reference to
this analysis. Equivalent analyses for open seats races,
revealing no significant effects of party, are shown in
Online Appendix Table D.3

Definition of variables in Table 2
1.  Incumbent percent adjusted (dependent variable,
DV). This is the incumbent party’s percentage of the
vote, but set to a ceiling and floor of 75% and 25%,
respectively.4
2.  State partisanship: State Republican advantage.
This is a measure of how much more Republican
the state is than the national average, based on the
two most recent presidential elections. The measure

 3

Goedert 
Table 2.  Effect of party on incumbent vote share and likelihood of turnover: Races involving senate incumbents, 1982–2016.
Incumbent party vote % (OLS)
 
Republican incumbent
Statewide partisanship
Incumbent extremism

(1)

(2)

(3)

(4)

(5)

(6)

(7)

−3.22**
(0.78)
0.26**
(0.028)
—

−2.16**
(0.55)
0.16**
(0.025)
−5.17**
(1.84)
0.42*
(0.17)
−0.84**
(0.31)
−2.13**
(0.15)

−2.47**
(0.57)
0.15**
(0.025)
−4.06*
(1.94)
0.48**
(0.18)
−1.23**
(0.40)
−2.05**
(0.16)

−2.32**
(0.66)
0.10**
(0.030)
−0.91
(2.17)
0.58**
(0.19)
−2.07**
(0.47)
−1.98**
(0.16)

0.70**
(0.23)
−0.029**
(0.011)
−0.75
(0.72)
−0.037
(0.085)
−0.31
(0.16)
0.80**
(0.16)

0.76**
(0.27)
−0.030*
(0.012)
−0.27
(0.78)
−0.10
(0.096)
−0.29
(0.20)
0.99**
(0.18)

1.23**
(0.41)
−0.038*
(0.019)
−2.20
(1.34)
−0.11
(0.16)
−0.58
(0.33)
1.62**
(0.32)

−1.19
(0.64)
−2.71**
(0.69)
−3.48**
(1.35)
−0.098**
(0.018)
−0.20**
(0.031)
−0.23
(0.54)
−1.83**
(0.63)
1.69*
(0.72)
0.19
(0.76)
−3.81**
(1.02)
81.3
(2.16)
Excluded
Excluded
487
0.643

−1.2
(0.65)
−2.62**
(0.71)
−3.82**
(1.41)
−0.080**
(0.019)
−0.22**
(0.033)
−0.20
(0.53)
−1.52*
(0.66)
—

−0.76
(0.67)
−2.28**
(0.74)
−3.08*
(1.44)
−0.055**
(0.020)
−0.15
(0.24)
−0.32
(0.53)
—

0.94**
(0.34)
0.62*
(0.29)
0.96
(0.49)
0.0076
(0.0073)
0.028*
(0.014)
0.51*
(0.25)
0.47
(0.27)
—

1.71**
(0.53)
1.27**
(0.49)
1.94*
(0.78)
0.0052
(0.0099)
−0.011
(0.17)
0.69
(0.37)
—

—

—

—

—

81.0
(2.47)
Included
Excluded
487
0.644

80.0
(3.51)
Included
Included
487
0.710

0.94**
(0.32)
0.66*
(0.27)
1.16**
(0.42)
0.0029
(0.0065)
0.022
(0.012)
0.37
(0.23)
0.52*
(0.25)
−0.27
(0.35)
0.059
(0.32)
0.73
(0.45)
−5.32
(0.93)
Excluded
Excluded
487
—

Incumbent terms

—

Log (incumbent spending)

—

Log (challenger spending)

—

Challenger quality
Local elected office

—

US Representative/
Statewide Office
Governor or Senator

—

Primary opposition %

—

Congressional districts

—

Senate control

—

South

—

Midterm election

—

Presidential control

—

Midterm*presidential control

—

Constant
 
Year fixed effects
State fixed effects
Observations
R2

Likelihood of turnover (Probit)

—

59.8
(0.54)
Excluded
Excluded
487
0.159

—

—

—

—

—

—

−6.81
(1.21)
Included
Excluded
487
—

−8.38
(2.44)
Included
Included
330
—

Standard errors are in parentheses.
*p < 0.05, **p < 0.01.
OLS: ordinary least squares.

used in the regressions, Statewide partisanship, is
the State Republican advantage when the incumbent is a Republican, and the inverse of this when
the incumbent is a Democrat.5
3.  Incumbent extremism. This is the absolute value of
an incumbent’s DW-NOMINATE score in the term
immediately prior to the election year, with lower
values indicating more moderation.

4.  Terms. This is the number of complete terms that
the incumbent had served prior to the election.
5.  Campaign spending. I include variables for logged
spending of both the challenger and incumbent parties, adjusted for state size and set to a floor of
USD$1000. Past research suggests that challenger
spending should be the more important of the two
when an incumbent runs.

 4 

Research and Politics 
6.  Challenger quality. This is scaled from 0 to 3, with
0 indicating the challenger had no prior elected
experience. Challengers are coded “3” if they were
current or former Governors or Senators; “2” if
their highest office was US Representative or other
statewide office; and “1” for lower offices. The
regressions show the effect of each individual category, but results are substantively identical if this is
modeled linearly or dichotomized.
7.  Primary opposition. This is the percentage of the
primary vote that was won by non-incumbents.
8.  Senate control, Presidential control, Midterm.
These are dependent variables indicating, respectively, if the incumbent’s party controlled the
Senate, the presidency, or if the race was contested
during a midterm election.
9.  Congressional districts. This is the number of congressional districts that are in the state as a measure
of relative population size.
10.  Fixed effects. Fixed effects for year and state are
used in some specifications.6

Fluke elections.  One possible hypothesis is that the disparity
is due to just a few anomalous election years. For example,
if there were only one or two particularly pro-Republican
years in which a large number of Democrats retired, or one
or two very pro-Democratic years with few retirements, a
theory about persistent trends throughout the era would be
less plausible. But this does not appear to be the case; as
shown in Table 2 (columns 3 to 4), the partisan difference
remains as statistically and substantively significantly large
when year fixed effects are included.7
Additionally, the disparity generally holds regardless of
whether the election year was overall favorable to the
Democrats, Republicans, or neither. In the four biggest
Senate Democratic wave cycles (1986, 2000, 2006, 2008),
the party picked up seats primarily by defeating Republican
incumbents (5–7 every cycle). But in four of the five biggest Republican waves (1994, 2002, 2004, 2010),
Republicans gained by winning a large number of seats
from retiring Democrats, but defeated only one or two
incumbents. And in most neutral years, the Republicans
won the majority of open seats but lost more incumbents
than the Democrats. In fact, the “flukiest” cycle of the past
36 years is the one that defies this trend: Republicans
defeated five incumbents in 2014, the only cycle where
more than two Democrats were defeated. Finally, although
Jacobson (2015) notes that incumbent advantage in House
elections has been declining in recent decades, this partisan
divide is not confined to early or late elections in the data
set. It appears in each of the 12-year “third” of the period
(shown in Figure 1).
Midterm loss. It is well established that the President’s
party tends to lose House and Senate seats in midterm

elections (e.g., Erikson, 1988), while beneficial riding of
coattails during presidential election years have been less
consistently evident (e.g., Campbell and Sumner, 1990).
Consistent with this, column 2 of Table 2 shows that, during midterms, incumbents of the President’s party perform 3.6 points worse on average than incumbents of the
opposed party, with other controls included. But the difference between Republican and Democratic incumbent
performance, both in vote share and turnover likelihood,
remains robust to this factor. Partisan control of the Senate has no significant effect. Moreover, the greatest differences in turnover rates between parties are seen during
presidential election years, not midterms, regardless of
the party in power (shown in Online Appendix C).8 Additionally, there is no significant effect of midterms years or
presidential control on likelihood of turnover, as shown in
column 5 of Table 2.
Ideological congruence. Canes-Wrone et al. (2002) suggest
that more extreme members of Congress gain less electoral
support. But this alone does not appear to be a significant
factor in explaining the partisan asymmetry in vote or turnover. In most specifications in Table 2, the effect of incumbent ideological extremism is in the expected direction but
not significant; in no case does it substantially alter the
effect of party on vote share or turnover probability. Interacting party with both statewide partisanship and ideological extremism (shown in Online Appendix Table E) increases
the effect of party when extremism is set to 0 while showing
an interaction effect suggesting a decline in party difference
as extremism increases, although this effect is only significant in some model specifications.9 And this cannot be
explained merely by an asymmetric distribution of these
variables across parties. Online Appendix C breaks down
races by statewide partisanship and ideological congruence,
showing that the partisan difference in turnover rates persists across friendly and unfriendly states, and among ideological moderate and extreme Senators.
Regional explanations/Southern trends. Another possible
explanation is that the disparity is due to regional partisan
shifts in the country during this period, perhaps combined
with the timing of retirements. For example, one could
posit that many Senate seats in conservative Southern states
would only remain Democratic as long as incumbents from
an earlier era held the seat; when those incumbents retired,
many Southern seats immediately shifted to the Republicans (e.g., Black, 2003). But Table 2 shows the partisan
disparity remains after including a control for the Southern
region, with no significant effect of interacting this control
with incumbent party (shown in Online Appendix E). Outside the South, Republican incumbents are defeated at
higher rates than Democrats in every other region of the
country, so a purely regional explanation is implausible
(shown in Figure 2 below).10

 Goedert 

5

Figure 2.  Senate seat turnover percentage by region, 1982–2016.

Weak freshmen/Democratic retirements. If Republicans
were winning a disproportionate number of open seats in a
certain cycle, that may create a larger number of weak GOP
freshmen running for reelection six years later. Yet this still
does not explain the disparity between reelection rates by
party, which persists after controlling for incumbent terms
in office. Although freshman senators of both parties lose
more often than very senior senators of their party, both
junior Republicans and senior Republicans lose at roughly
double the rate of Democrats of similar seniority (20% vs.
9% turnover rate for first-term senators; 12% vs. 6% for
senators running for a fourth term or more).
A related explanation might be that Democrats retire
more readily than Republicans when endangered. There is
some evidence that candidates choose retirement strategically (e.g., Choi, 2017; Highton, 2011; Carson, 2005), but
little research that this decision is asymmetric by party. I
analyzed the effect of seniority and state and candidate ideology on the likelihood of retirement, with results in Table
3. Partisanship, Terms, and Extremism are operationalized
as in Table 2, while “Incumbent Party Wave” is the
Incumbent’s Party’s advantage in the House popular vote
that cycle.
This table shows that senior senators of both parties are
more likely to retire than junior ones, and senators in ideologically hostile states are slightly more likely to retire than
those in friendlier states. However, Republicans are not significantly less likely to retire than Democrats. And none of

these factors affect Republican retirement decisions significantly differently from Democrats. Moreover, this alternate
explanation is strongly contradicted anecdotally: 30 of 36
Democrats won while running in states at least 15 percentage points more Republican than the nation, compared to
five of nine incumbent Republicans running in states that
were as extremely Democratic.
There is one intriguing partisan asymmetry in Table 3:
Democratic retirements are significantly more sensitive to
waves than Republican retirements. However, a deeper
examination of this phenomenon shows that Democrats are
not more likely to retire during adverse waves; rather,
Republicans are more likely to retire during favorable
waves.11 This might suggest more effective coordination of
retirements by the Republican party, but this is likely a
more meaningful explanation of Republican success in
open seats than the Democratic incumbent advantage.
Challenger quality and fundraising. Several authors have
argued that Senate campaigns can be divided into categories of intensity: “low key” and “hard fought,” largely
based on the viability of the challenger (e.g., Westlye, 1991;
Krasno, 1994; Kahn and Kenney, 1997, 1999). When the
campaign fails to attract a high-quality challenger, the
dynamics of the Senate campaigns resemble those of a typical House campaign, and incumbents are almost always
reelected. But in the case of high intensity campaigns, with
quality challengers and high spending, incumbents become

 6 

Research and Politics 

Table 3.  Effect of qualities of incumbent on probability of
retirement, 1982–2016.
Probability of open seat

(1)

(2)

(3)

Republican incumbent

−0.118
(0.115)
—

—

−0.315
(0.392)
−0.015
(0.008)
0.007
(0.011)
0.067
(0.701)
0.808
(0.865)
0.182**
(0.050)
−0.007
(0.076)
−0.048**
(0.015)
0.061**
(0.022)
−1.149
(0.303)
611

Statewide partisanship

−0.012*
(0.005)
—

Statewide partisanship*GOP

—

Incumbent extremism

—

Incumbent extremism*GOP

—

Incumbent terms

—

Incumbent terms*GOP

—

Incumbent party wave

—

Incumbent party wave*GOP

—

−0.016
(0.181)
—

−0.773
(0.080)
611

−1.462
(0.181)
611

Constant
Observations

0.716
(0.395)
—
0.181*
(0.037)
—

Notes: Entries are probit coefficients. Standard errors are in parentheses.
DV is assigned as “1” if seat was open, as “0” if incumbent was nominated for reelection.
*p < 0.05, **p < 0.01.
GOP: Republican Party; DV: dependent variable.

quite vulnerable, more resembling a national campaign.
Other authors focus specifically on challenger experience
(e.g., Squire, 1989; Lublin, 1994). Table 2 measures the
effect of both challenger experience and fundraising on
incumbent success.
Although the independent effect of challenger quality in
the post-1980 Senate data is consistent with existing
research (25% of incumbents facing experienced challengers lose, compared to 4% of incumbents facing inexperienced challengers), this cannot explain the partisan
difference, as the results are robust to the inclusion of this
control. There is also no significant difference between the
overall quality of challengers faced: 46% of Democrats
faced experienced challengers, compared with 47% of
Republicans, and the rate of turnover was about double for
Republicans for each category of challenger.
The concept of campaign “intensity” also meshes well
with findings that Senate election outcomes are more sensitive to challenger campaign spending than incumbent
spending (Jacobsen, 1980; Abramowitz, 1988, 1991). While
incumbent spending is less variable, high challenger spending may be a signal of a high-intensity campaign (e.g.,
Goodliffe, 2007). Indeed, Jacobson (2006) finds that greater
challenger spending tends to induce greater incumbent

spending, probably indicative of high campaign intensity.
The results in Table 2 align with this past research on campaign intensity, suggesting challenger spending is more
important to outcomes than incumbent spending; the somewhat perverse effect of incumbent spending is likely
explained by incumbents being forced to spend more against
high-quality challengers. But the partisan difference, both
with respect to turnover rates and vote share, is also robust
to controlling for incumbent and challenger spending. The
total incumbent spending advantage is also almost identical
by party; the median spending ratio in favor of the incumbent is 2.8-to-1 for Republican incumbents, compared to
2.7-to-1 for Democrats.12
Constituency size. Some have argued that incumbent vulnerability is related to constituency size (Abramowitz,
1988; Hibbing and Brandes, 1983), with incumbents
defeated more often in populous states because they are
less personally in-touch with a large number of their constituents. But any claim that the observed partisan difference can be explained by constituency size is probably
implausible for two reasons. First, this relationship is far
from a universal finding. Krasno (1994) concludes there is
no evidence for a relationship between constituency size
and incumbent performance in all but a handful of very
large states, while Lascher (2005) argues that small constituencies may lead to more competitive elections through
higher quality challengers. Second, this past research would
probably predict a partisan effect in the reverse direction
from what we observe: incumbents would be safer in small
states (generally more Republican) and more vulnerable in
large states (generally more Democratic), and thus on average we would expect to see more vulnerable Democratic
incumbents. Moreover, Table 2 shows the partisan disparity
is robust to controlling for state size, operationalized by
number of CDs.
Primary opposition or divisiveness.  Kenney and Rice (1984)
find that increased divisiveness of a Senate primary election
(as measured by the winner’s primary vote percentage) may
carry over into reducing the candidate’s vote in the general
election. And indeed, Table 2 shows that primary opposition
significantly correlates with reduced general election vote
share among incumbents (though it does not significantly
affect turnover likelihood). However, the partisan difference
remains substantively and significantly robust when including this control. Given the effect of Tea Party candidates on
the recent outcomes in several seats, one might also hypothesize that primary opposition would differently affect
Republican and Democratic candidates. However, there is
no significant difference between the effect that primary
opposition has on Republican incumbents compared to
Democrats, or any significant difference within subsets of
primary strength (see Online Appendices C and E).

 7

Goedert 
Economic factors. Lublin (1994) finds that national economic conditions are a significant predictor of Senate
incumbent vote share. I tested the effects of a variety of
national economic indicators, including change in disposable income, unemployment, and stock market growth, on
both incumbent vote share and turnover likelihood. None
of these indicators significantly correlated with turnover,
and only unemployment rate significantly correlated with
vote share; more importantly, the independent effect of
party remained statistically and substantively robust to the
inclusion of all economic indicators (results in Online
Appendix F, columns 5 and 6).
Partisan issue competence.  Throughout much of the 20th
century, an apparently persistent public preference for
divided government existed in which Republicans more
often won the presidency but Democrats won majorities
in Congress. Jacobson (1990) explained this by arguing
that voters’ choice of party control of each branch is based
on perceived differences between the parties. He noted
that in late 20th century surveys, voters judged Republicans more competent in dealing with issues that affect the
nation as a whole, such a defense, taxation, and inflation,
while judging Democrats more competent on issues that
affect small groups and localities, such as unemployment,
agriculture, education, and problems affecting minorities.
He also notes that voters prefer presidents who will focus
on national problems, but legislators who will put the
interests of their district ahead of the national interest.
This explanation relying on party issue competence has
diminished since the Republican revolution of 1994, but it
could also apply to preferences over incumbency and
open seats. If voters prefer Republicans when dealing
with national issues, ideology may be more important in
judging a Republican incumbent, while seniority may be
more important when judging a Democrat, seen as strong
in delivering local or distributional benefits. More
recently, Grossman and Hopkins (2016) argue that Republicans campaign more ideologically, while Democrats
campaign through providing interest group benefits.
Grimmer (2013) finds that Senate incumbents of both parties in ideological unfriendly states may emphasize appropriations over position taking, but the data explored here
suggests that Democrats may be asymmetrically successful in doing so.
Although this theory is intuitively promising, an initial
exploration based on available election and survey data
yields contradictory and inconclusive results. This theory
would suggest that state partisanship and incumbent ideology should be more important to vote outcomes involving
Republican incumbents, while perhaps factors like seniority would be more important to Democratic outcomes. But
models interacting party with partisanship, ideology, and
terms (shown in Online Appendix E) yield little significant
evidence for this pattern.13

Additionally, I briefly explored this idea through two
survey series targeting Senate elections: the National
Election Studies Senate Study conducted in 1988–1992,
and the Cooperative Congressional Election Study (CCES)
from 2008 to 2012, with contradictory results. In the earlier
Senate Study, voters seeking local benefits were significantly more likely to vote for Democratic incumbents compared with voters concerned about national issues, and all
voters were more likely to identify local activity as a
strength of Democratic incumbents running for reelection
than Republicans, providing some initial support for the
theory. But the more recent CCES revealed no evidence
that voters weighed ideology more heavily when evaluating
a vote for Republican incumbents (shown in Online
Appendix Tables H–J.) Moreover, while these studies
remain the best existing survey resources on Senate elections, each has significant weaknesses. The National
Election Studies Senate Study is now 25 years old, and covers only a set of low-competition election cycles atypical of
the larger era. The CCES, designed largely for study of the
lower house, is more poorly adapted to Senate research in
both its sample and choice of questions.

Discussion
This note uncovers an approximately 3-percentage-point
difference in vote share and an 8-point difference in turnover rate between Democratic and Republican Senate
incumbents persisting over more than three decades. While
such an effect may seem superficially small, it is equal to or
larger than effects observed by other canonical factors studied in US Senate elections.14 Additionally, the effect is of
substantive importance to government outcomes. The gap
in reelection rates observed here suggests a net gain of 3–4
seats for the Democrats, enough to frequently swing control over such a closely contested body. Yet this possibly
pivotal difference has remained unexplored, further suggesting a need for additional study targeted at Senate elections, campaigns, and voter perceptions of parties.
Beyond the potential of additional individual-level
research, the upcoming 2018 midterm elections also provide a possibly pivotal test of the theory for several reasons.
First, although the partisan patterns in the late 20th century
appear to have been replicated in the first decade of the
21st, results in the last two cycles call into question whether
those trends are still relevant today. As mentioned above,
2014 represented a clear exception in which five incumbent
Democrats were defeated. And despite only Republican
incumbents being defeated, 2016 presented a poor test
overall, with no Democrats running in strongly Republican
states. The year 2018 presents a clearer contrast, with ten
Democrats running in states won by the most recent
Republican presidential candidate. On top of the typical
benefit of running in a midterm election against the
President’s party, the patterns explored here suggest that

 8 
these Democratic incumbents may be able to overcome
unfavorable terrain, perhaps due to an ability to craft a
moderate image independent from their national party, in a
way that both Republican incumbents and all candidates
running in open seats have been unable to. For almost 40
years, the brand Democratic incumbents presented to voters has resisted overall partisan trends in a way the
Republican brand has not. Even with the backdrop of a
divisive Republican president, 2018 presents an unusually
intimidating challenge to that brand. Whether the resilience
of Democratic Senate incumbents proves to be a historical
anomaly or an enduring but unstudied fact in American
politics may also hang in the balance.
Declaration of conflicting interest
The author declared no potential conflicts of interest with respect
to the research, authorship, and/or publication of this article.

Funding
The author received no financial support for the research, authorship, and/or publication of this article.

Supplemental materials
The supplemental files are available at http://journals.sagepub.
com/doi/suppl/10.1177/2053168018803811
The replication files are available at https://dataverse.harvard.edu/
dataset.xhtml?persistentId=doi%3A10.7910%2FDVN%2FZOC
PYH

Notes
 1.  Six races where an incumbent was defeated in a primary
are counted as open seats; two races were excluded where
an incumbent was defeated in a primary but won reelection
as an independent, with results robust to different coding
choices.
  2.  Although Table 2 separates incumbents from open seats, the
partisan difference in incumbent vote share and turnover is
also robust to nesting open seats within the same model;
result available from author.
  3.  Online Appendix D shows the effect of state partisanship is
larger in the case of open seats than incumbent races, with
no significant party difference in any open seat specification. This suggests that the Republican advantage in open
seat turnover can be explained simply by state partisanship,
with a greater number of states favoring Republicans at the
national level.
 4.  Approximately 6% of the sample lies in these extreme
ranges. Running all analyses on actual incumbent party performance yields substantive results that are almost identical,
but does nothing to adjust for unopposed races.
  5.  Although an imperfect measure of underlying voter preferences, averaging across recent presidential vote for federallevel partisanship is commonly accepted practice when such
measurement is required across states and time periods (e.g.,
Ansolabehere et al., 2001).
  6.  State fixed effects cannot be used in analyzing open seats due
to lack of within-state variation.

Research and Politics 
  7.  The complete results of including year and party interacted
fixed effects are shown in Online Appendix G.
  8.  Although the difference in turnover is greatest in presidential election years, there is no significant difference in the
party disparity in vote share when the data is broken down
into midterm and presidential cycle subsets. In both cases,
Republicans perform 2–3 points worse than Democrats with
other controls included; see Online Appendix F, columns 3
and 4.
 9.  The sign of the interacted GOP*Extremism coefficient is
positive, mostly counteracting the negative coefficient for
uninteracted extremism. This could be interpreted to suggest
that more voters are inclined to reward or punish the individual moderation or extremism of Democratic incumbents
than Republicans, although the inconsistent and marginal
significance of this interaction warns against any strong conclusions without further evidence.
10.  Online Appendix F shows the model from Table 2, column
2 broken down into South and non-South subsets. The difference between Democratic and Republican incumbent
vote share remains significant in the non-South subset, with
no significant or substantive difference shown in the South
subset.
11.  The effect of the wave variable entirely disappears when the
quartile of cycles most favorable to the incumbent are truncated from the data set, but increases when the least favorable
quartile are truncated. When the wave variable is included in
analysis with incumbent vote share as a DV, the substantive
effect of party is reduced by about 20% while remaining significant at p < 0.01 (results available from author).
12.  Online Appendix F, column 7 additionally shows that the
party difference is robust to an alternate definition of the
spending variables, including one control for the ratio of
incumbent to challenger spending and a second control for
the sum of overall spending, as a representation of campaign
intensity.
13.  Columns 2–4 of Online Appendix E show results including only interactions relevant to an existing theory, varying
the inclusion of state and year fixed effects. Columns 5 and
6 include interactions with all controls included in Table 2
(with the exception of spending interactions excluded in column 6).
14.  For example, Canes-Wrone et al. (2002) find that a two SD
change in incumbent ideology corresponds to a 2–6% change
in vote share, while Squire (1989) finds that a one major category increase in quality increases challenger vote share by
approximately 1.2%.

References
Abramowitz A (1988) Explaining Senate election outcomes.
American Political Science Review 82(2): 385–403.
Abramowitz A (1991) Incumbency, campaign spending, and the
decline of competition in US House elections. The Journal of
Politics 53(1): 34–56.
Ansolabehere S and Snyder J (2002) The incumbency advantage
in US elections: An analysis of state and federal offices,
1942–2000. Election Law Journal 1(3): 315–338.
Ansolabehere S, Snyder J and Stewart C, III (2001) Candidate
positioning in US House elections. American Journal of
Political Science 45(1): 136–159.

 Goedert 
Black E (2003) The Rise of Southern Republicans. Cambridge,
MA: Harvard University.
Campbell J and Sumners J (1990) Presidential coattails in Senate
elections. American Political Science Review 84(2): 513–524.
Canes-Wrone B, Brady D and Coogan J (2002) Out of step, out of
office: Electoral accountability and House members’ voting.
American Political Science Review 96(1): 127–140.
Carson J (2005) Strategy, selection, and candidate competition in US
House and Senate elections. Journal of Politics 67(1): 1–28.
Choi J (2017) Leaving office: The US Senator’s representation,
ideological adoption, and strategic retirement. International
Journal of Social Science Studies 5(10): 17–27.
Erikson R (1988) The puzzle of midterm loss. The Journal of
Politics 50(4): 1011–1029.
Goodliffe J (2007) Campaign war chests and challenger quality in
Senate elections. Legislative Studies Quarterly 32(1): 135–156.
Grimmer J (2013) Appropriators not position takers: The distorting effects of electoral incentives on congressional representation. American Journal of Political Science 57(3):
624–642.
Grossmann M and Hopkins D (2016) Asymmetric Politics:
Ideological Republicans and Group Interest Democrats.
Oxford: Oxford University Press.
Hibbing J and Brandes S L (1983) State population and the electoral success of US Senators. American Journal of Political
Science 27(4): 808–819.
Highton B (2011) The influence of strategic retirement on the
incumbency advantage in US House elections. Journal of
Theoretical Politics 23(4): 431–447.

9
Jacobson G (1980) Money in Congressional Elections. New
Haven, CT: Yale University Press.
Jacobson G (1990) The Electoral Origins of Divided Government.
San Francisco, CA: Westview.
Jacobson G (2006) Campaign spending effects in US Senate elections. Electoral Studies 25(2): 195–226.
Jacobson G (2015) It’s nothing personal: The decline of the
incumbency advantage in US House elections. Journal of
Politics 77(3): 861–873.
Kahn K and Kenney P (1997) A model of candidate evaluations
in senate elections: The impact of campaign intensity. The
Journal of Politics 59(4): 1173–1205.
Kahn K and Kenney P (1999) The Spectacle of US Senate
Campaigns. Princeton, NJ: Princeton University.
Kenney P and Rice T (1984) The effect of primary divisiveness
in gubernatorial and senatorial elections. The Journal of
Politics 46(3): 904–915.
Krasno J (1994) Challengers, Competition, and Reelection:
Comparing Senate and House Elections. New Haven, CT:
Yale University.
Lascher E (2005) Constituent size and incumbent safety. Political
Research Quarterly 58(2): 269–278.
Lublin DI (1994) Quality, not quantity: Strategic politicians in US
Senate elections, 1952–1990. The Journal of Politics 56(1):
228–241.
Squire P (1989) Challengers in US Senate elections. Legislative
Studies Quarterly 14(4): 531–547.
Westlye M (1991) Senate Elections and Campaign Intensity.
Baltimore, MD: Johns Hopkins University.