Content uploaded by Brendan Pablo Montagnes
Author content
All content in this area was uploaded by Brendan Pablo Montagnes on Mar 20, 2014
Content may be subject to copyright.
The Presidential Race for Office and the Persistent
Electoral Bias It Creates: Evidence from Entry and
Exit of Senators
Yosh Halberstam∗B. Pablo Montagnes†
September 13, 2009
Draft
Abstract
In this paper, we compare senators first elected during midterm elections with
those first elected during presidential elections and find them strikingly different:
the cohort of senators first elected during presidential elections is consistently more
ideologically extreme and party disciplined than the cohort first elected midterm.
This result is surprising in light of empirical evidence suggesting that the electorate
in presidential elections is more ideologically moderate and less partisan than the
electorate in midterm elections. Furthermore, we find that senators who are ousted
or retire from office sometime around presidential elections are significantly more
ideologically moderate and vote more independently than those who exit around
midterm elections. Together, these two empirical regularities suggest that the rela-
tively more moderate electorate in presidential elections generates a more extreme
and polarized Senate. These findings suggest that holding concurrent races for of-
fice is not outcome neutral and raise policy questions about the timing of elections
and ballot initiatives. Our empirical approach is robust to econometric specification
and outliers and can be extended to examining models of electoral competition and
voting behavior.
∗Department of Economics, Northwestern University, 2001 Sheridan Road, Evanston, IL 60208, USA;
E-mail: y-halberstam@northwestern.edu.
†Department of Managerial Economics and Decision Science, Kellogg School of Manage-
ment, Northwestern University, 2001 Sheridan Road, Evanston, IL 60208, USA; E-mail: p-
montagnes@kellogg.northwestern.edu.
1
1 Introduction
The laws governing the timing of races for public office vary across electoral systems.
While predetermined term lengths for elected federal offices fix the timing of general
elections in the United States, the number of simultaneously contested offices in a given
election cycle is variable; specifically, the office of president is contested every four years,
whereas congressional races are held every two years. The presence of an additional
concurrent race for office during a presidential election cycle is what sets it apart from
midterm elections. Can this institutional detail have a systematic and long-term effect
on the type of candidates that prevail?
In this paper, we provide strong empirical evidence that the presence of a presidential
race for office affects the manner in which citizen preferences are aggregated. Moreover,
we show that electoral outcomes are biased in a particular and counterintuitive way. Our
empirical results suggest that the two electoral environments—midterm and presidential
elections—aggregate the preferences of citizens differently: we find that senators first
elected during presidential elections are more ideologically extreme than senators first
elected midterm. Conversely, we find that senators who are ousted or retire from the Sen-
ate sometime around presidential elections are more ideologically moderate than those
who exit midterm. We also find extremism highly associated with party loyalty. Thus,
while empirical evidence suggests that the electorate is relatively moderate in presidential
elections, the electoral environment returns a more ideologically extreme and disciplined
Senate overall. In addition to the theoretical implications for models of voting and elec-
toral competition, demonstrating that midterm and presidential elections systematically
produce significantly different outcomes may have implications for politics and policymak-
ing in general, and the timing of elections and ballot initiatives in particular. Awareness
of such persistent differences may aid lawmakers in establishing electoral rules by bringing
attention to the possible significance of holding multiple races for office at the same time.
There are two main reasons for holding concurrent races for office. First, since the
variable costs of running an election—hiring monitors and polling staff, printing ballots,
delivering equipment, securing polling locations—do not vary much with the number
2
of offices being contested, the marginal cost of holding an additional race for office is
relatively low. The fixed costs, however, can be substantial. A recent special election for a
single congressional seat in Illinois was estimated to cost over $3.5 million or $33 per vote.1
With all 435 seats in the House and one-third of the 100 seats in the Senate contested
every general-election year, the possible savings are considerable. A second reason is the
cost reduction for prospective voters. Holding races for different offices simultaneously
decreases the time and effort required for citizens to turn out and vote, and mitigates the
cost of acquiring information about policy issues, parties and candidates.
While the direct effects of holding concurrent races for office are unambiguously pos-
itive, there are a variety of indirect effects that make the overall net benefit of such
elections less obvious. The behavior of voters, candidates, parties, the media and other
players may differ systematically between presidential and midterm elections. Citizens
may have strategic and informational concerns that alter their voting behavior in each
environment. In response, parties may implement their own set of changes. In addition,
holding many races for office simultaneously can alter the parties’ strategic landscape. For
example, differences in fundraising between the two environments can affect the candidate
selection process. Thus, if changes in voter or party behavior are widespread, they can
lead to persistent differences in electoral outcomes.2
In this paper, we identify the total net of all such indirect effects on electoral outcomes.
In other words, we recover the differential effect of holding a presidential race for office
on the selection of senators. We demonstrate consistent and significant differences in
outcomes as measured by ideology and party discipline between midterm and presidential
elections. If different electoral environments create persistent outcome bias, then one must
tak, the resulting implications must be taken into account when assessing the fitness
and suitability of possible electoral systems or reforms. For example, the savings of
1F.N. D’Alessio, The Associated Press State & Local Wire, March 4, 2009,
http://www.huffingtonpost.com/2009/03/05/board-wants-mail-in-speci_n_172157.html.
2An evaluation of the normative implications of such differences in the aggregation of preferences
depends on the underlying mechanism that generates them and the measure of the degree to and manner
in which they differ. In the concluding section, we suggest a possible mechanism, and provide an example
of the direct positive effects of these differences, but leave out a discussion of the normative concerns.
3
holding local elections concurrently with federal elections should be evaluated in light of
such possible bias. Apart from the normative and positive policy concerns, identifying
persistent bias that stems from variation in an institutional detail provides guidance for
theoretical models of voting and electoral behavior. We discuss some previous models that
take into account differences between midterm and presidential elections in the concluding
section.
Our new empirical approach is our second contribution in this paper. The voting
behavior of citizens and the electoral decisions of candidates are linked. The inclusion
of an ideologically appealing candidate may encourage citizens to turn out and vote.
Similarly, a particular distribution of citizen preferences may motivate a candidate of
a particular bent to run for office. The possibility that the distribution of voters can
influence the distribution of candidates and vice versa generates an endogeneity problem
for an econometrician testing models of voting and electoral competition. Our empirical
method solves this problem and enables a broad set of data to be applied to such models for
hypothesis testing. To purge the aforementioned relationships from endogeneity, we utilize
the exogenous oscillation between presidential and midterm elections in the United States.
While significant variation in voter turnout between midterm and presidential elections
is observed, the electoral institutions and the rules governing these elections remain the
same. In both midterm and presidential elections, congressional offices are contested
by two major parties and outcomes are determined by plurality rule. Consequently, the
marginal effect of a presidential race for office on features that are present in both electoral
environments is identified.
In this paper, we proceed as as follows. In the next section, we describe the electoral
environment, our basic approach and our general results. We also review the relevant
literature. In section 3, we describe our data. In section 4, we present our results formally,
establishing their significance. We conclude with a discussion of the result in light of a
variety of theories, and outline an explanation that we expand upon in a companion paper.
4
2 Midterm and Presidential Elections
In the United States, all federal-office elections occur concurrently and are held on the first
Tuesday following the first Monday in November during even-numbered years.3Members
of the House of Representatives, senators, and presidents serve terms of two, six and
four years, respectively; these varying term lengths create two staggered federal electoral
cycles. Senators are elected in one of two electoral environments: either they are elected
concurrently with presidents or they are elected in the absence of a presidential race
during midterm-election years.4As a result, at any given time, the Senate contains two
dominant groups of senators: those who are first elected to the Senate during presidential
elections and those who are first elected during midterm elections between presidential
election years. The existence of these two electoral environments is an artifact of the
variation in term length between senators and the president. There was no explicit intent
by the framers of the constitution to create electoral environments that would differ in any
significant way; in fact, senators were originally chosen by state legislatures, leaving the
selection process explicitly separated from the electoral environment. Thus, the specific
environment—presidential or midterm elections—in which senatorial candidates run for
office is not intended to favor any one type of candidate over another.
The literature on multiple election environments primarily focuses on the impact of
presidential coattails on wins down-ticket, and individual ticket-splitting. Research on the
coattail effect—the relationship between the popularity of a presidential candidate and
the winning prospects of other candidates from the same party—attempts to identify the
seat-vote relationship or identify differences in the electorate. (See Besley and Preston
(2007), Campbell (1993) and Coate and Knight (2007) for a review of this literature.) For
example, Campbell (1993) has examined the differences in the size and composition of the
3The sole exceptions to this rule are special elections, which are held to fill the remaining terms of
offices vacated prematurely.
4The oscillating electoral environments create three classes of senators and three distinct senatorial
electoral cycles. At the national level, there are five electoral cycles: the two year House cycle, the four
year presidential cycle, and the three classes of senatorial six year cycles. For statewide elections, there
are three electoral cycles, the presidential cycle and the two senatorial cycles. Because there are three
classes of six-year terms and the two senators from the same state belong to different classes, electoral
cycles for junior and senior senators will be two or four years apart.
5
voting population in midterm and presidential elections, and the success of the president’s
party in each environment. However, previous studies have failed to look at the types of
candidates elected in each environment. Our results suggest that presidential coattails or
other mechanisms might have consequences beyond success for the president’s party and
may affect the type of senators elected from both parties as reflected in their ideological
positions and voting behavior.
A more recent strand of literature has attempted to make the connection between the
electorate and the ideology of senators elected. A leading example is the work of Aldrich,
Brady, de Marchi, McDonald, Nyhan, Rohde, and Tofias (2008). In a book chapter, the
authors demonstrate a link between constituent demographics and the ideological position
of senators. They demonstrate consistent and strong relationships between ideology scores
of senators and demographic characteristics of states. However, since their data is a
broad measure of state population, they do not connect the ideology of senators to the
preferences of voters in particular and to electoral environments in general.
Since we focus on senators’ entry and exit electoral environments, we refer to cohorts
of senators as follows: the presidential-entry cohort refers to senators first elected during
a presidential election year, while the midterm-entry cohort refers to senators first elected
during a midterm election year. Conversely, the presidential-exit cohort refers to senators
who leave office during a presidential election year, whereas the midterm exit cohort refers
to senators who leave office during a midterm election year.5
We compare the ideologies of senators first elected during midterm elections with
those first elected during presidential elections. In other words, we examine how electoral
environments affect the process in which citizen preferences are aggregated and repre-
sented. Our finding is that the cohort of senators elected during presidential elections is
more ideologically extreme than the cohort elected during midterm elections. This re-
sult is surprising in light of empirical evidence suggesting that the electorate’s ideological
composition is relatively more moderate in presidential elections.6
5Because the electoral cycles of a junior and senior senator in a given state are two or four years apart,
a state’s Senate delegation may or may not display both electoral environments.
6Surprisingly, the literature comparing the electorate in midterm and presidential elections is somewhat
lacking, and direct comparisons of the electorates are missing from the literature (see Leighley and Nagler
6
In addition to the electoral-environment results, we obtain an analogous set of results
by segmenting senators according to their exit environment. Once again, counter to what
empirical evidence on the electorate would suggest, senators who leave office sometime
around presidential elections are significantly more ideologically moderate than those who
leave office during midterm elections. Together, these two sets of results suggest that the
relatively more moderate electorate in presidential elections generates a more extreme and
polarized Senate.
Studying the relationship between electoral environments and outcomes solves some
of the problems associated with testing models of electoral competition and voting be-
havior. Our methodology relates variation in electoral environments to outcomes, and
not variation in characteristics of the electorate to outcomes. This allows us to examine
a wide range of electoral outcomes at the national level, does not restrict our analysis to
a particular time or locality and does not require us to collect data on voter preferences
or individual characteristics. Because our focus is on aggregate results, we do not need to
be concerned about unobserved elements of competition, such as the position of defeated
candidates who did not win office. Our focus is on establishing a relationship between
the types of senators elected and the two electoral environments, and not on direct tests
of electoral-competition models. A discussion of how the results established in this paper
relate to various models of voting and electoral competition follows in the conclusion.
(2007)). However, given the substantial variation in turnout between presidential and midterm elections,
we can use results on the likelihood of voting and ideology to interpolate the differences. In Palfrey and
Poole (1987), the authors find the likelihood to vote highly correlated with the ideological extremism of the
voter. Leighley and Nagler (2007) find additional indirect evidence that voters are more strongly partisan
than non-voters. Two recent working papers lend further indirect evidence by highlighting the ideological
differences between voters and citizens in general. In a recent working paper, Shor (2009) estimates
citizen preferences in a left-right ideological space and uncovers a unimodal distribution, whereas Herron
and Bafumi (2007) estimate voting preferences and find a bimodal distribution with low density about
the mean.
7
3 Data
Our dataset comprises senators who were elected to the senate from 1966 to present.7For
each senator we gather biographical information from the CQ Congressional Collection.
We combine these data with Poole and Rosenthal’s DW-NOMINATE dataset. Using
information about the senators’ entry and exit environments and party identification, we
employ a non-parametric methodology to compare the DW-NOMINATE scores of senators
first elected during presidential elections to their midterm counterparts. We perform this
comparison for each congress in our 40 years of data, and run an identical analysis in
which we use a senator’s exit environment as the classification criterion.
3.1 Senate Composition and Flows
Our dataset consists of all senators who faced federal elections for the first time between
1966 and 2006. For each senator we gather biographical information from the CQ Con-
gressional Collection. These data include party affiliation as well as the starting and
ending dates for service in the Senate. We use these dates to construct two classification
variables: the first indicates the entry environment of each senator and the second, if
applicable, the exit environment. We exclude senators who were appointed to fill a va-
cated seat, unless they were subsequently elected during regular federal elections. In our
analysis of exit electoral environments, we do not distinguish between incumbent senators
who lose and those who do not seek reelection; however, we exclude senators who leave
office due to death or who leave office before the end of their term.8Finally, we preclude
7This period is characterized by the emergence of two major parties following the decline of Southern
Democrats that ensued the ratification of the Voting Rights Act in 1965 and the signing of the Civil
Rights Act of 1964 (see Gelman, Park, Shor, Bafumi and Cortina(2008) for further reading).
8This is done with the strategic concern of the senator in mind and in light of our focus on the effects
of electoral environments. If a senator anticipates an unfavorable electoral environment, her best course
of action might be to step down rather than seek reelection. (See Diermeier, Keane and Merlo (2005) for
more on this topic.) This classification will present an interpretive problem only if there is a systemic
difference in the retirement decisions of senators facing different electoral environments that does not
depend on how favorable an environment is but, nonetheless, varies with it. For example, if senators
have a strong preference for serving one term or one and three terms, their retirement decision will be
correlated with their electoral environment. However, having performed appropriate testing, we know
this not to be the case.
8
Table 1: Senate Electoral Composition
Democrats Republicans
Entry Exit Entry Exit
Midterm 45 21 54 40
Presidential 58 35 64 41
Total 103 56 118 81
from our analysis senators who were not affiliated with the Democratic or Republican
parties during both their entry into and, if applicable, exit from the Senate.
In Table 1, we summarize the data by party identification with respect to our classi-
fication criteria.9Since our data on exiting senators is right-censored, we observe a sub-
stantially smaller number of incumbents who leave the Senate than the number of new
senators who enter. Over the course of the data, 221 new senators are elected and 137
incumbents exit the Senate; 45 percent of turnover in the Senate occurs during midterm
elections.
Since we do not track senators who were elected before 1966, our dataset contains only
a handful of senators in the initial years. We begin with seven senators in the 90th congress
and reach over 80 by the 97th. In Figure 1a, we display for each congress in our data the
composition of the Senate with respect to the entry classification. For example, of 100
senators in the 109th congress, 93 were elected during regular federal elections in 1966 or
later, 55 were elected during a presidential election, and the rest during midterm elections.
The remaining seven senators were either elected before 1966, appointed or specially
elected and did not subsequently compete in regular federal elections or were affiliated
with a third party.10 Overall, the Senate maintains a relatively balanced proportion of
senators who are elected during midterm elections to those elected during presidential
elections. We detail this feature of the Senate in Figure 1b. In this graph, we display
for each party the number of new senators entering the Senate following each election.
9Senators who switch parties during their service are coded using their initial party affiliation for entry
and final party affiliation for exit.
10Ted Kennedy and Robert Byrd were elected before 1966 and are still serving in the Senate, there was
one Independent senator and four appointed senators who did not face federal elections.
9
2
2
25
5
58
8
85
5
511
11
1113
13
1318
18
1812
12
1225
25
2521
21
2133
33
3316
16
1638
38
3817
17
1734
34
3417
17
1736
36
3621
21
2139
39
3919
19
1947
47
4723
23
2350
50
5022
22
2251
51
5125
25
2549
49
4922
22
2243
43
4324
24
2441
41
4124
24
2442
42
4228
28
2845
45
4526
26
2644
44
4427
27
2741
41
415
5
58
8
814
14
148
8
819
19
1913
13
1324
24
2411
11
1124
24
2419
19
1929
29
2917
17
1735
35
3530
30
3048
48
4828
28
2849
49
4930
30
3048
48
4821
21
2140
40
4025
25
2541
41
4124
24
2439
39
3925
25
2540
40
4023
23
2347
47
4730
30
3049
49
4927
27
2750
50
5025
25
2544
44
4421
21
2146
46
4627
27
2751
51
5190
90
9091
91
9192
92
9293
93
9394
94
9495
95
9596
96
9697
97
9798
98
9899
99
99100
100
100101
101
101102
102
102103
103
103104
104
104105
105
105106
106
106107
107
107108
108
108109
109
10990
90
9091
91
9192
92
9293
93
9394
94
9495
95
9596
96
9697
97
9798
98
9899
99
99100
100
100101
101
101102
102
102103
103
103104
104
104105
105
105106
106
106107
107
107108
108
108109
109
109congress
congress
congressDemocrats
Democrats
DemocratsRepublicans
Republicans
Republicanspresidential
presidential
presidentialmidterm
midterm
midtermGraphs by firstparty
Graphs by firstparty
Graphs by firstparty
(a) Composition
2
2
26
6
63
3
38
8
88
8
89
9
910
10
102
2
22
2
25
5
511
11
115
5
52
2
26
6
61
1
16
6
65
5
58
8
81
1
13
3
35
5
58
8
85
5
55
5
52
2
28
8
810
10
1015
15
153
3
32
2
22
2
24
4
43
3
36
6
610
10
109
9
94
4
42
2
28
8
87
7
790
90
9091
91
9192
92
9293
93
9394
94
9495
95
9596
96
9697
97
9798
98
9899
99
99100
100
100101
101
101102
102
102103
103
103104
104
104105
105
105106
106
106107
107
107108
108
108109
109
10990
90
9091
91
9192
92
9293
93
9394
94
9495
95
9596
96
9697
97
9798
98
9899
99
99100
100
100101
101
101102
102
102103
103
103104
104
104105
105
105106
106
106107
107
107108
108
108109
109
109Democrats
Democrats
DemocratsRepublicans
Republicans
Republicans
(b) Flow
Figure 1: Senate Entry Environment
10
Although we observe variation in the number of entrants across congresses, its correlation
with the alternating electoral environments is negligible.
Analogous to the description on entry environments, in Figures 2a and 2b, we display
the exit decomposition of the Senate and exit flow of incumbents. Similar to the entry
decomposition, the ratio of incumbents who leave the Senate sometime around a midterm
election to those who leave sometime around a presidential election remains stable. How-
ever, unlike the entry data, the exit data contains fewer senators at both tailends. The
diminishing number of exiting senators toward the end of our data produces the triangu-
lar bar chart we observe in Figure 2a. Here, too, the exit flow of senators appears to be
uncorrelated with the alternating exit environments.
3.2 DW-NOMINATE Scores
As a measure of a senator’s ideology we use Poole and Rosenthal’s DW-NOMINATE
scores.11 The data employed for estimating these scores consists of (nearly) all individual
roll call votes in United States congressional history. DW-NOMINATE scores are esti-
mates derived from a dynamic weighted nominal three-step estimation procedure, which
was created by Poole and Rosenthal in the 1990s. An iterative Maximum Likelihood
estimation is employed to recover each legislator’s ideal point and roll call midpoints of a
spatial model in a random utility framework. The points are placed in a common space
and constrained to lie within a unit hyperspace. The point estimates are robust to con-
cerns about strategic voting, logrolling and time-variant ideal points.12 We employ the
first dimension of DW-NOMINATE scores, which captures the ideology of senators in the
liberal-conservative (or left-right) space; a higher score is associated with a more conserva-
tive voting record. For example, in the 109th congress, the Senate voted on 645 roll calls;
the average DW-NOMINATE score for a Democrat was -0.428, and for a Republican,
11In our robustness checks in the Appendix we employ alternative ideology measures and achieve similar
results. For further reading see Poole and Rosenthal (2000).
12
In Table 7 in the Appendix we employ W-NOMINATE scores, which are a static version of DW-
NOMINATE scores, and rule out a potential concern that the persistence of our results is driven by
uncommon voting behavior in a particular year.
11
2
2
25
5
57
7
77
7
79
9
99
9
915
15
1512
12
1221
21
2113
13
1327
27
2719
19
1930
30
3014
14
1425
25
2515
15
1526
26
2614
14
1425
25
2521
21
2130
30
3022
22
2231
31
3122
22
2231
31
3116
16
1626
26
2616
16
1621
21
219
9
916
16
1611
11
1114
14
146
6
611
11
116
6
69
9
93
3
33
3
35
5
56
6
612
12
1212
12
1218
18
1813
13
1323
23
2314
14
1423
23
2311
11
1124
24
2417
17
1730
30
3019
19
1940
40
4022
22
2241
41
4119
19
1939
39
3919
19
1929
29
2916
16
1628
28
2818
18
1827
27
2714
14
1425
25
2518
18
1830
30
3015
15
1528
28
2816
16
1626
26
2610
10
1021
21
2112
12
1219
19
199
9
916
16
1690
90
9091
91
9192
92
9293
93
9394
94
9495
95
9596
96
9697
97
9798
98
9899
99
99100
100
100101
101
101102
102
102103
103
103104
104
104105
105
105106
106
106107
107
107108
108
108109
109
10990
90
9091
91
9192
92
9293
93
9394
94
9495
95
9596
96
9697
97
9798
98
9899
99
99100
100
100101
101
101102
102
102103
103
103104
104
104105
105
105106
106
106107
107
107108
108
108109
109
109congress
congress
congressDemocrats
Democrats
DemocratsRepublicans
Republicans
Republicanspresidential
presidential
presidentialmidterm
midterm
midtermGraphs by lastparty
Graphs by lastparty
Graphs by lastparty
(a) Composition
1
1
11
1
11
1
14
4
46
6
62
2
22
2
22
2
26
6
65
5
57
7
74
4
44
4
42
2
26
6
63
3
32
2
23
3
34
4
42
2
22
2
23
3
39
9
95
5
54
4
44
4
43
3
36
6
64
4
46
6
65
5
53
3
37
7
79
9
992
92
9293
93
9394
94
9495
95
9596
96
9697
97
9798
98
9899
99
99100
100
100101
101
101102
102
102103
103
103104
104
104105
105
105106
106
106107
107
107108
108
108109
109
109110
110
11092
92
9293
93
9394
94
9495
95
9596
96
9697
97
9798
98
9899
99
99100
100
100101
101
101102
102
102103
103
103104
104
104105
105
105106
106
106107
107
107108
108
108109
109
109110
110
110Democrats
Democrats
DemocratsRepublicans
Republicans
RepublicansGraphs by lastparty
Graphs by lastparty
Graphs by lastparty
(b) Flow
Figure 2: Senate Exit Environment
12
0.458.
The dynamic weighting of roll calls in the DW-NOMINATE estimation procedure
affords the scores cardinality. In other words, while information on scores alone can-
not indicate the exact number of roll calls on which one senator voted differently from
another, increasing disparity between their DW-NOMINATE scores suggests that the un-
derlying voting records that generated them are increasingly different. In this paper, we
focus on differences in electoral outcomes. For our purposes, DW-NOMINATE scores are
particularly useful as they are derived from realized voting records of senators and not
ideological preferences reported by them or other surveyed groups.13 Importantly, the
use of DW-NOMINATE scores allows us to connect directly electoral environments to the
spatial model framework that is central to theories of electoral competition and voting.
We merge the DW-NOMINATE data with the CQ Congressional Collection data. An
observation in our dataset consists of time variant and invariant variables. A senator’s
biographic information as well as her original entry environment and (if applicable) her
exit environment, do not vary by congress; however, the measure of a senator’s ideology
does vary by congress and is a function of her voting behavior. For a given senator, the
time variant components are congress number, length in office, upcoming and past elec-
toral environments, and DW-NOMINATE scores; the time invariant variables are party
labels, dummies for entry electoral environments, dummies for exit electoral environments
(if applicable), characteristics of the entry environment, and starting and ending years (if
applicable).
4 Results
We employ a non-parametric method to identify the presence of systematic ideological
differences between senator cohorts who differ with respect to the electoral environments
in which they enter or exit the Senate. The objective of this section is to document
13Several alternatives to DW-NOMINATE scores that are occasionally used in empirical research, most
notably ADA scores, are produced by partisan lobby groups and rely on voter behavior on a particular
set of votes that each group chooses.
13
thoroughly two empirical regularities observed in the United States Senate; one related
senator entry and the other related to their exit.14
4.1 Electoral Bias in Entry
For each congress in our dataset, we partition the Senate into four groups per our de-
scription in the data section. We first segment senators by their initial party affilia-
tion—Democrat or Republican—and then further divide the senators by electoral envi-
ronment—midterm or presidential —in which they first won a Senate election. Following
our exclusion criteria, a senator in our entry analysis must belong to one of the following
four groups: presidential-entry Democrats, midterm-entry Democrats, presidential-entry
Republicans or midterm-entry Republicans.
Using DW-NOMINATE scores, for every congress in our dataset, we compute the
average ideology scores for each of the four groups of senators. We report these results in
Figure 3a. In this graph, each point corresponds to the average DW-NOMINATE score
of senators in a given group for a given congress. For example, for the congress ending in
1990 there are two points associated with each party in the Senate: one for the midterm-
entry cohort and one for the presidential-entry cohort. The average DW-NOMINATE
score for midterm-entry senators was -0.3 and 0.27 and for presidential entry senators,
-0.347 and 0.34, for Democrats and Republicans, respectively.
Two notable patterns emerge in Figure 3a. While the first regarding the slopes of the
connecting lines is not of little import, it is the second pattern concerning the ordinal
raking of the lines that is novel.15 The line connecting the set of points that are asso-
ciated with the average ideology scores of presidential-entry Democrats (Republicans) is
consistently below (above) the connecting line associated with the average ideology scores
of midterm-entry Democrats (Republicans). In other words, the average voting behavior
14Our results are robust to alternative ideology scores, such as ADA, weighted DW-NOMINATE and
W-NOMINATE. See the Appendix for testing under these specifications.
15The first pattern, which is not the subject of this paper, is well-documented in the literature: over the
past 40 years, Democrats have become relatively more liberal, and Republicans, more conservative, the
overall effect being increasing ideological polarization between Democrats and Republicans in congress in
general and in the Senate in particular.
14
-.5
-.5
-.5-.4
-.4
-.4-.3
-.3
-.3-.2
-.2
-.2.1
.1
.1.2
.2
.2.3
.3
.3.4
.4
.4.5
.5
.51970
1970
19701980
1980
19801990
1990
19902000
2000
20002010
2010
20101970
1970
19701980
1980
19801990
1990
19902000
2000
20002010
2010
2010Democrats
Democrats
DemocratsRepublicans
Republicans
Republicanspresidential
presidential
presidentialmidterm
midterm
midtermDW-NOMINATE
DW-NOMINATE
DW-NOMINATEyear
year
yearGraphs by firstparty
Graphs by firstparty
Graphs by firstparty
(a) Ideologies by Party
-.1
-.1
-.1-.05
-.05
-.050
0
0.05
.05
.05.1
.1
.1.15
.15
.15DW-NOMINATE
DW-NOMINATE
DW-NOMINATE1970
1970
19701980
1980
19801990
1990
19902000
2000
20002010
2010
2010year
year
yearDemocrats
Democrats
DemocratsRepublicans
Republicans
Republicanspresidential-midterm
presidential-midterm
presidential-midterm
(b) Normalized Ideological Differences
Figure 3: Ideologies of Midterm versus Presidential Entry Cohorts
15
of a Democrat senator who first enters the Senate following a victory in a presidential
election is consistently more liberal than that of the average Democrat senator who is
elected to the Senate for the first time during midterm elections. Similarly, in each of
the 20 congresses we display in the figure, the voting records of Republicans whose entry
into the Senate follows presidential elections are relatively more conservative than those
of Republicans who enter following midterm elections. To evaluate this new pattern more
clearly we neutralize the other one concerning the long-run trend in congressional ide-
ology. In Figure 3b we display the average ideological differences between presidential-
and midterm-entry senator cohorts alone. To produce this figure we subtract the average
DW-NOMINATE scores for midterm cohorts from the scores of the presidential cohorts.
We do this for both parties. The zero reference line represents the normalized ideology
scores of midterm-entry cohorts. There is no instance in which the connecting lines cross
the reference line and in most congresses the ideological differences are substantial.16
4.2 Electoral Biases in Exit
We now perform an analogous analysis of senators’ exit environments. Similar to the entry
methodology, for each congress in our dataset we partition the Senate into four groups.
We first segment senators by their final party affiliation—Democrat or Republican—and
then further divide the senators by the electoral environment—midterm or presidential
—in which they leave the Senate. Consequently, a senator in our exit analysis must belong
to one of the following four groups: presidential-exit Democrats, midterm-exit Democrats,
presidential-exit Republicans or midterm-exit Republicans.
As before, using their DW-NOMINATE scores, for each congress in our dataset, we
compute the average ideology score of senators in each of the four groups. We report
these results in Figure 4a. In the analysis of exit environments the pattern that arises is
16While the focus of our analysis is on the original entry environment of senators, one might be con-
cerned that the original electoral environment proxies for upcoming electoral environments and that these
electoral concerns are generating the observed differences. In Table 6 in the Appendix, we address these
concerns by including dummies for a senator’s upcoming and most recent electoral environments. We
find our main results unaffected by immediate electoral concerns, which in turn have no statistically
significant effect on observed voting records.
16
-.5
-.5
-.5-.4
-.4
-.4-.3
-.3
-.3-.2
-.2
-.2.1
.1
.1.2
.2
.2.3
.3
.3.4
.4
.4.5
.5
.51970
1970
19701980
1980
19801990
1990
19902000
2000
20001970
1970
19701980
1980
19801990
1990
19902000
2000
2000Democrats
Democrats
DemocratsRepublicans
Republicans
Republicanspresidential
presidential
presidentialmidterm
midterm
midtermDW-NOMINATE
DW-NOMINATE
DW-NOMINATEyear
year
yearGraphs by lastparty
Graphs by lastparty
Graphs by lastparty
(a) Ideologies by Party
-.2
-.2
-.2-.1
-.1
-.10
0
0.1
.1
.1.2
.2
.2.3
.3
.3DW-NOMINATE
DW-NOMINATE
DW-NOMINATE1970
1970
19701980
1980
19801990
1990
19902000
2000
2000year
year
yearDemocrats
Democrats
DemocratsRepublicans
Republicans
Republicanspresidential-midterm
presidential-midterm
presidential-midterm
(b) Normalized Ideological Differences
Figure 4: Ideologies of Midterm versus Presidential Exit Cohorts
17
Table 2: Weighted Least Squares
Independent Variable: DW-NOMINATE
Entry Exit
Democrat Republican Democrat Republican
Presidential dummy -0.0547* 0.0890** 0.136** -0.171**
(0.022) (0.001) (0.000) (0.000)
Constant -0.257** 0.319** -0.356** 0.439**
(0.000) (0.000) (0.000) (0.000)
Observations 697 742 379 504
Robust p values in parentheses
** p<0.01, * p<0.05
even more pronounced than the one observed for entry environments.17 For both parties
the midterm-exit cohorts are significantly and consistently more ideologically extreme
than the presidential-exit cohorts. Figure 4b illustrates the normalized version of these
differences.
In Table 2, we report entry and exit regression results employing a Weighted Least
Squares (WLS) approach.18 In this regression we account for the fact that DW-NOMINATE
scores are estimates and are, therefore, a noisy signal of the actual unobserved ideology of
senators. To internalize the inaccuracy we weigh each estimate by its corresponding boot-
strapped standard error. Consequently, in the least squares minimization problem, less
accurate estimates will be less significant than more accurate and informative estimates.
We report the regression results for each class of senators for the entire period.19 The
17In Figure 8 in the Appendix we replicate our main results using the weighted DW-NOMINATE
scores. We follow the DW-NOMINATE model specification and compute the weighted average of the
ideology estimates of both dimensions using the weights provided by Poole and Rosenthal. The results
follow a similar pattern to those using the first dimension of DW-NOMINATE scores. The ideologi-
cal interparty average distance between presidential-entry cohorts is 81 percent greater than between
midterm-entry cohorts, whereas midterm-exit cohorts are nearly three and a half times more differenti-
ated than presidential-exit cohorts.
18See Table 7 in the Appendix for similar results using alternative ideology measures.
19We are aware that the DW-NOMINATE estimates in this regression are not independent across
senators and across congresses; however, any potential bias is unlikely to be correlated with electoral
environments and for precision purposes is second-order.
18
presidential dummy regressor is equal to one for senators whose entry (or exit) is during
presidential elections. Consequently, the coefficients on the constant represent the average
ideology scores for midterm cohorts. Overall, the regression results indicate that for both
Democrats and Republicans the average presidential entrant is twenty to thirty percent
more ideologically extreme than the average midterm entrant. The differences between
exit cohorts are starker: an average senator who exits during a presidential election is
likely to be thirty to forty percent more ideologically moderate than a senator who exits
during midterm elections.20
One way to get a sense of the magnitude of these differences is to compare the in-
terparty ideological differences across electoral environments. In our period of analysis,
the average difference between Democrats and Republicans in DW-NOMINATE scores
is roughly 0.65. Relative to the interparty ideological difference, the presidential entry
cohorts are 11 percent ideologically further apart and the midterm-entry cohorts are 11
percent ideologically closer together. Overall, the ideological interparty average distance
between the presidential entry cohorts is 25 percent greater than between midterm-entry
cohorts, whereas midterm-exit cohorts are 63 percent more ideologically differentiated
than presidential-exit cohorts.
4.3 Nesting Entry in Exit
A strong observation in our results is that the differences between presidential and midterm-
exit cohorts are substantially greater than the differences between the entry cohorts.21
This is true for both Republicans and Democrats. This observation has important impli-
cations for the ideological composition of the Senate overall. While senators who enter
during presidential elections are more extreme than those who enter during midterm elec-
20Notethat the one-sided nulls—presidential dummy>0 for Democrat entrants and Republicans who
exit and vice versa for Democrats who exit and Republican entrant—are rejected at the 5 percent signif-
icance level as indicated by the reported p-values.
21If senators predominately serve an odd numbers of terms, then entry and exit effects are not be
independent of each other. In Table 5 in the Appendix we address this concern by including both a
dummy for entry and a dummy for exit concurrently. If anything, the results seem stronger when we
control for entry and exit simultaneously. This strongly suggests that entry and exit effects are not driven
by the length of tenure in the Senate.
19
-.3
-.3
-.3-.2
-.2
-.2-.1
-.1
-.10
0
0.1
.1
.1.2
.2
.2DW-NOMINATE
DW-NOMINATE
DW-NOMINATE1970
1970
19701980
1980
19801990
1990
19902000
2000
2000year
year
yearDemocrats entry
Democrats entry
Democrats entryRepublicans entry
Republicans entry
Republicans entryDemocrats exit
Democrats exit
Democrats exitRepublicans exit
Republicans exit
Republicans exitentry: presidential-midterm; exit: midterm-presidential
entry: presidential-midterm; exit: midterm-presidential
entry: presidential-midterm; exit: midterm-presidential
Figure 5: Nesting Ideology
tions and those who exit during presidential elections are more moderate than those who
exit during midterm elections, the ideological disparity in exit is significantly greater than
in entry. In Figure 5 we establish this relationship by nesting the differences in entry
cohorts within the differences in exit cohorts. We do this by collapsing Figure 3b with
the inverse of Figure 4b; we subtract the midterm averages from the presidential averages
in entry and vice versa in exit. Figure 5 not only confirms our conjecture as a global
statistic in our data, but also establishes its regularity over the past forty years. Beyond
the ordinal ranking, the differences between entry and exit are highly correlated at 0.928;
thus, greater ideological differences in entry are associated with even greater ideologi-
cal differences in exit. With extremists entering and moderates leaving office regularly
during presidential elections, the overall composition of the Senate may radicalize and
become more partisan over time, which is consistent with empirical evidence on Senate
polarization.
20
4.4 The Presidency and Party Loyalty
In this subsection we present further evidence of electoral bias generated by the presi-
dential race for office. We show that senators first elected during midterm elections are
less party disciplined than senators first elected during presidential elections. Conversely,
senators who exit during midterm elections more frequently vote along party line than
those who exit during presidential elections. Thus, not only are the senators who enter
during presidential elections more ideologically extreme, they are also more loyal to their
respective parties. This result is no less surprising than the ideological differences we
find among senator cohorts—a priori we do not expect to observe systematic differences
between senators first elected (ousted) during midterm elections versus those who enter
(exit) during presidential elections.
A second point we address concerns the utility framework specified in the DW-NOMINATE
algorithm. The specification of a spatial component in the model used to estimate sen-
ators’ ideal points implies a certain behavior: one votes for the alternative that is most
closely aligned with her position. Given this assumption and that Democrats’ positions
are likely to be more liberal than those of Republicans, an extremely liberal senator is
less likely to vote along Republican party lines than a senator that is moderately lib-
eral. Thus, our results suggest that the DW-NOMINATE model is robust to the spatial
specification—extreme senators are also more loyal to their party.
We use unity scores as a proxy for party loyalty. A unity vote is defined as a roll
call during which at least 50 percent of Republicans vote against at least 50 percent of
Democrats. A unity score is the percent of votes cast with one’s party from the total unity
roll calls on which the voter cast her vote; thus, a higher score implies greater party loyalty
(e.g., a score of 100 implies that on all unity roll calls in which a senator participated
she voted party line). The literature has also referred to these scores as a measure of
accountability to one’s constituents; thus, a lower score implies greater accountability.
Unlike DW-NOMINATE scores, which are derived from a parametric model, unity scores
are free of misspecification concerns.
We first plot in Figure 6a the long-term trend in unity scores for midterm and pres-
21
65
65
6570
70
7075
75
7580
80
8085
85
8590
90
90average Unity score
average Unity score
average Unity score1970
1970
19701980
1980
19801990
1990
19902000
2000
20002010
2010
2010year
year
yearEntry
Entry
Entry60
60
6070
70
7080
80
8090
90
90100
100
1001970
1970
19701980
1980
19801990
1990
19902000
2000
2000year
year
yearpresidential
presidential
presidentialmidterm
midterm
midtermExit
Exit
Exit
(a) Unity by Environment
0
0
05
5
510
10
1015
15
1520
20
20average Unity score
average Unity score
average Unity score1970
1970
19701980
1980
19801990
1990
19902000
2000
2000year
year
yearentry
entry
entryexit
exit
exitentry: presidential-midterm; exit: midterm-presidential
entry: presidential-midterm; exit: midterm-presidential
entry: presidential-midterm; exit: midterm-presidential
(b) Nesting Unity
Figure 6: Party Discipline
22
idential cohorts for both entry and exit. Similar to the ideology plots, the population
of senators has become increasingly more party disciplined over time, yet the systematic
differences between the presidential and midterm cohorts persist throughout our data.
Unlike ideology, party loyalty is vertically differentiated and unity scores are placed on a
common scale for both Democrats and Republicans. To get a better sense of the magni-
tude of these differences, in Figure 6b we plot the normalized unity scores for entry and
exit. As with the ideological nesting plot, we subtract midterm averages from presidential
averages in entry and vice versa in exit. The predominant pattern that emerges is that
presidential entry (exit) cohorts display greater (less) loyalty toward their respective par-
ties than midterm-entry (exit) cohorts. As in the case of ideology, the results for exit are
more pronounced than those for entry. On average, a senator who is first elected during
midterm elections is 20 percent more likely to vote against his party than one who enters
during presidential elections. Similarly, senators who exit during presidential elections are
51 percent more likely to vote against their party than their midterm counterparts.
4.5 Ruling Out Outliers
Since we presented our main results establishing significant and persistent bias in elec-
toral outcomes using averages, a main concern might be that our results are driven by
outliers. Since averages are sensitive to outliers, we now perform an analogues empirical
analysis using cohort medians rather than cohort averages. To construct Figure 7, we
first identify the median senator in each cohort of every congress. We then subtract the
DW-NOMINATE score of the midterm cohort median senators from the presidential co-
hort median senators to produce the analogous normalized ideology figures we presented
using averages. As with the averages, a similar pattern emerges: the presidential entry
(exit) cohort medians are systematically more ideologically extreme (moderate) than the
midterm-entry (exit) cohort medians. Thus, there is little evidence that our results us-
ing averages are a product of outliers. If anything, outliers have resulted in mitigating
the differences that our work aims to capture. In Table 3 we report the results from a
Weighted Least Absolute Deviations (WLAD) regression that speaks to this point. The
23
-.2
-.2
-.2-.1
-.1
-.10
0
0.1
.1
.1.2
.2
.2DW-NOMINATE
DW-NOMINATE
DW-NOMINATE1970
1970
19701980
1980
19801990
1990
19902000
2000
20002010
2010
2010year
year
yearDemocrats
Democrats
DemocratsRepublicans
Republicans
Republicanspresidential-midterm
presidential-midterm
presidential-midterm
(a) Entry
-.4
-.4
-.4-.2
-.2
-.20
0
0.2
.2
.2.4
.4
.4DW-NOMINATE
DW-NOMINATE
DW-NOMINATE1970
1970
19701980
1980
19801990
1990
19902000
2000
2000year
year
yearDemocrats
Democrats
DemocratsRepublicans
Republicans
Republicanspresidential-midterm
presidential-midterm
presidential-midterm
(b) Exit
Figure 7: Normalized Senate Cohort Medians
24
Table 3: Weighted Least Absolute Deviations
Independent Variable: DW-NOMINATE
Entry Exit
Democrat Republican Democrat Republican
Presidential dummy -0.0800** 0.0940** 0.128** -0.249**
(0.000) (0.001) (0.000) (0.000)
Constant -0.231** 0.298** -0.325** 0.500**
(0.000) (0.000) (0.000) (0.000)
Observations 697 742 379 504
p values in parentheses
** p<0.01, * p<0.05
results are more striking than the ones produced by the WLS regression; the presidential
entry (exit) cohort median ideology is approximately thirty to forty (forty to fifty) percent
more extreme (moderate) than that of midterm-entry (exit).
5 Conclusion
We find strong evidence that senators first elected during presidential-election years are
ideologically more extreme than their counterparts elected during midterm elections. We
take this as evidence that these two environments systematically affect the type of can-
didates elected to office or their behavior in the Senate. In addition to this result, we
find even stronger evidence suggesting that the environment in which a senator leaves
office is also correlated with her policy positions in the Senate. Here, we find that sen-
ators who are more moderate leave office during presidential-election years, and more
extreme senators exit during midterm elections. Together, these two facts suggest that
the presidential-election season returns a more extreme and polarized Senate.
Formal literature on the interaction between contemporaneous elections is somewhat
lacking. With the important exception of Alesina and Rosenthal (1989), (1995) and
(1996), most of the theoretical modeling has been informal. This literature has tended to
25
view down-ticket-voting in light of presidential politics, and has primarily been preoccu-
pied with midterm decline. There are two broad themes in this literature. The first views
midterm elections as a reversion to the mean in terms of presidential support. For example,
in the surge and decline models (see Angus Campbell (1960) and James Campbell(1991)),
the major difference between the two electorates is the presence of presidential partisan
voters. Thus, midterm elections are distinguished from presidential elections by their lack
of voters in support of the president’s party. Conversely, another strand of the literature
characterizes midterm elections by the presence of voters who vote against the president’s
party (see Erikson (1988), Kernell (1977) and Tufte (1975)). Both approaches, however,
fail to provide an account of why we might observe consistent differences in ideology by
electoral environment across parties and over time.
There are three main reasons why these models are ill-suited for explaining the facts
we uncover in this paper. First, they do not directly consider down-ticket races, but
instead focus on the effect of presidential politics on voter decision making. Second, the
models are purely partisan and do not consider races in light of a spatial setting. And
third, it is difficult to incorporate spatial competition into models that posit that voting
decisions in congress are based on the perception of the presidency.
A model that accounts for the effects of multiple simultaneous elections in a spatial
setting is Alesina and Rosenthal’s balanced government model. In their model, voters
attempt to balance the policy produced by congress and the president by electing a divided
government. This produces a more moderate policy outcome, which better reflects the
preferences of voters. Our results on electoral entry and exit environments are difficult
to reconcile with such a model, unless voters are systematically electing senators who are
extreme in the opposite direction of the presidential preference. Furthermore, presidential
candidates’ preferred by such split-ticket voters would have to be sufficiently ideologically
extreme such that the balancing senators from the opposing party would necessarily be
even more extreme than their midterm counterparts. Preliminary evidence (expanded in
Halberstam and Montagnes (2009b)) suggests that the opposite is true: states that vote
for Democratic presidential candidates elect more-liberal candidates during presidential
26
elections then during midterm elections.
An explanation that we explore in our related paper is that in environments with
simultaneous elections, voters who are only partially informed about all the races for
office rely on party labels to make inferences about the positions of candidates. In our
model, voters have prior beliefs about party platforms, but do not observe candidate
positions in all races for office. Upon observing candidates in one race, voters update their
beliefs about parties and candidates in other races. This updating introduces a rational
contagion effect that alters the competitive landscape in races in which voters do not
observe candidate positions. A relatively more moderate candidate in the observed race
creates a coattail effect for other members of her party running for office. This enables
relatively more extreme and less electorally viable candidates to win. Alternatively, a
relatively more extreme candidate reduces the range of electorally attainable positions for
her ticket.
A point that we have emphasized throughout the paper is that DW-NOMINATE
scores correspond to real-world voting records of senators. The differences in the ideology
of senators will, thus, be reflected in their voting records and ultimately in the types
of bills passed by Congress. To illustrate the connection, we provide in table 4 the
breakdown of one important roll call vote chosen from the Americans for Democratic
Action (ADA) set of critical roll calls from 1984. The vote was on an amendment to a
bill that appropriates $700 million in federal grants to states to provide health benefits
to the long-term unemployed. The amendment was rejected by 57 to 39 votes. With 14
Democrats joining 43 Republicans in voting nay, and 11 Republicans joining 28 Democrats
in voting yay, voting on this amendment was not entirely partisan. The total number
of senators voting against the majority of their party was 25 out of 96 or about 26
percent (11 out of 54 or 20 percent for Republicans and 14 out of 42 or 33 percent for
Democrats). However, if we restrict our attention to the set of senators in our data and
look at the differences between electoral environments, we find that the moderation of
the midterm-entry cohort is reflected in a much greater propensity to cross party lines
and vote against the majority of the party. For Democrats, a 53 percent majority of
27
Table 4: 1984 ADA Senate Vote 3
Entry Exit
Midterm Presidential Midterm Presidential
Vote Yay Nay Yay Nay Yay Nay Yay Nay
Democrats 8 9 13 3 7 2 8 7
Republicans 5 16 5 22 3 15 4 18
the midterm-entry cohort voted with the Republican majority, whereas only a 19 percent
minority of the presidential-entry cohort did the same. Similarly, for the Republicans,
close to one-quarter of the midterm-entry cohort voted with the Democrats, while over
80 percent of the the presidential-entry cohort voted along party line. We find a similar
pattern when we tabulate the votes on this this bill with respect to one’s exit environments.
Consistent with our broader results, the midterm-exit cohorts are significantly less likely
than presidential cohorts to vote independently from their party. Among Democrats,
nearly half the presidential-exit cohort crossed party lines, while only 22 percent of the
midterm-exit cohort followed suit. Although counterfactual predictions about the type of
legislation produced and passed by the Senate are difficult to make, since the agenda of
the Senate is endogenous to the particular makeup of the Senate, this example illustrates
that a Senate comprised of moderates from the midterm cohort would be more likely to
have cross-partisan voting patterns if faced with the same set of voting decisions.
Finally, our results suggest that the timing of multiple elections has significant effects
on the type of senator elected as reflected by her voting behavior. The normative implica-
tions of such a result depend on the model of policy formation that one employs. However,
an awareness of the result might inform debates over policy issues such as the timing of
an election for multiple offices. When studying electoral institutions, the temptation is to
look at elections in isolation. our result suggests that simultaneous elections can impact
one another’s electoral outcomes, beyond the immediate electoral success.
28
References
[1] Abramowitz, Alan I. 1988. “Explaining Senate Election Outcomes.” American Polit-
ical Science Review 82(2): 385-403.
[2] Aldrich, John, Michael Brady, Scott de Marchi, Ian McDonald, David Rohde and
Michael Tofias. 2008. "Party and Constituency in the U.S. Senate, 1933-2004." in
Why Not Parties? Nathan W. Monroe, Jason M. Roberts, and David Rhode, eds.,
University of Chicago Press.
[3] Alesina, Alberto, and Howard Rosenthal. 1989. “Partisan Cycles in Congressional
Elections and the Macroeconomy.” American Political Science Review 83(2): 373-
398.
[4] Alesina, Alberto and Howard Rosenthal. 1995. Partisan Politics, Divided Govern-
ment, and the Economy. Cambridge University Press. Cambridge.
[5] Alesina, Alberto, and Howard Rosenthal. 1996. “A Theory of Divided Government.”
Econometrica 64(6): 1311-1341.
[6] Ansolabehere, Stephen, James M. Snyder and Charles Stewart. 2001. “Candidate
Positioning in U.S. House Elections.” American Journal of Political Science 45(1):
136-159.
[7] Bafumi, Joseph and Michael Herron. 2007. "Preference Aggregation, Representation,
and Elected American Political Institutions." Dartmouth University. Working paper.
http://www.dartmouth.edu/~herron/RepInstitutions.pdf
[8] Besley, Timothy and Ian Preston. 2007. “Electoral Bias and Policy Choice: Theory
and Evidence.” Quarterly Journal of Economics 122(4): 1473-1510.
[9] Burden, Barry C., Gregory A. Caldeira and Tim Groseclose. 2000. “ Measuring the
Ideologies of U.S. Senators: The Song Remains the Same.” Legislative Studies Quar-
terly 25(2): 237-258.
29
[10] Callander, Steven and Catherine H. Wilson. 2007. “Turnout, Polarization, and Du-
verger’s Law.” Journal of Politics 69(4): 1047-1056.
[11] Campbell, Angus. 1960. “ Surge and Decline: A Study of Electoral Change.” Public
Opinion Quarterly 24(3): 397-418.
[12] Campbell, James E. 1991. “ The Presidential Surge and its Midterm Decline in Con-
gressional Elections, 1868-1988.” Journal of Politics 53(2): 477-487.
[13] Campbell, James E. 1993. The Presidential Pulse of Congressional Elections. 2nd ed.
University Press of Kentucky.
[14] Coate, Stephen and Brian Knight. 2007. “Socially Optimal Districting: A Theoretical
and Empirical Exploration.” Quarterly Journal of Economics 122(4): 1409-1471.
[15] Diermeier, Daniel, Michael Keane and Antonio Merlo. 2005. “A Political Economy
Model of Congressional Careers.” American Economic Review 95(1): 347-373.
[16] Erikson, Robert S. 1988. “The Puzzle of Midterm Loss.” Journal of Politics 50(4):
1011-1029.
[17] Gelman, Andrew, David Park, Boris Shor, Joseph Bafumi, Jeronimo Cortina. 2008.
Red State, Blue State, Rich State, Poor State: Why Americans Vote the Way They
Do. Princeton University Press.
[18] Halberstam, Yosh and B. Pablo Montagnes. 2009. “Party Labels and Information:
The Implications of Contagion in Coelection Environments.” Northwestern Univer-
sity. Working Paper.
[19] Kernell, Samuel. 1977. “Presidential Popularity and Negative Voting: An Alterna-
tive Explanation of the Midterm Congressional Decline of the President’s Party.”
American Political Science Review 71(1): 44-66.
[20] Leighley, Jan E. and Jonathan Nagler. 2007. “ Who Votes Now? And Does It Matter?”
presented at the Annual Meeting of the Midwest Political Science Association, April,
2007.
30
[21] Palfrey, Thomas R. and Keith T. Poole. 1987. “The Relationship between Informa-
tion, Ideology, and Voting Behavior.” American Journal of Political Science 31(3):
511-530.
[22] Poole, Keith T. and Howard Rosenthal. 2000. Congress: A Political-Economic His-
tory of Roll Call Voting. Oxford University Press, USA.
[23] Shor, Boris. 2009. "All Together Now: Putting Congress, State Legislatures, and
Individuals in a Common Ideological Space" University of Chicago. Working paper.
http://home.uchicago.edu/~bshor/research/ccap.pdf
[24] Tufte, Edward R. 1975. “Determinants of the Outcomes of Midterm Congressional
Elections.” American Political Science Review 69(3): 812-826.
31
Appendix
Table 5: Entry and Exit Multicolinearity
(a) DW-NOMINATE
Democrats Republicans
Presidential entry -0.0286* 0.103**
(0.0134) (0.0262)
Presidential exit 0.116** -0.116**
(0.0136) (0.0257)
Constant -0.382** 0.337**
(0.0120) (0.0252)
Observations 380 388
Robust standard errors in parentheses
** p<0.01, * p<0.05
(b) Weighted DW-NOMINATE
Democrats Republicans
Presidential entry -0.0982** 0.216**
(0.0283) (0.0406)
Presidential exit 0.152** -0.179**
(0.0271) (0.0397)
Constant -0.242** 0.168**
(0.0245) (0.0378)
Observations 380 388
Robust standard errors in parentheses
** p<0.01, * p<0.05
32
Table 6: Immediate Electoral Concerns
(a) DW-NOMINATE
Democrats Republicans
Presidential entry -0.0503** 0.0467**
(0.0119) (0.0170)
Presidential next election -0.00527 -0.0310
(0.0162) (0.0219)
Presidential previous election -0.0153 -0.0121
(0.0158) (0.0199)
Constant -0.309** 0.318**
(0.0131) (0.0162)
Observations 696 750
Robust standard errors in parentheses
** p<0.01, * p<0.05
(b) Weighted DW-NOMINATE
Democrats Republicans
Presidential entry -0.131** 0.101**
(0.0229) (0.0269)
Presidential next election -0.0155 -0.0196
(0.0322) (0.0349)
Presidential previous election -0.0220 0.0130
(0.0299) (0.0314)
Constant -0.157** 0.130**
(0.0260) (0.0264)
Observations 696 750
Robust standard errors in parentheses
** p<0.01, * p<0.05
33
Table 7: Alternative Ideology Measures
(a) ADA
D R D R D R
Presidential entry 5.338** -4.679** 4.138* -6.210**
(1.240) (1.105) (1.686) (1.593)
Presidential exit -6.467** 4.894** -14.64** 9.980**
(1.694) (1.635) (1.502) (1.570)
Constant 72.03** 20.67** 69.16** 22.24** 79.18** 16.32**
(0.899) (0.852) (1.550) (1.471) (1.004) (1.015)
Observations 1335 1413 840 908 743 823
Robust standard errors in parentheses
** p<0.01, * p<0.05
(b) W-NOMINATE
D R D R D R
Presidential entry -0.0780** 0.0718** -0.0582 0.0639
(0.0230) (0.0268) (0.0311) (0.0396)
Presidential exit 0.0795* -0.142** 0.189** -0.231**
(0.0316) (0.0395) (0.0279) (0.0365)
Constant -0.553** 0.404** -0.537** 0.432** -0.671** 0.523**
(0.0160) (0.0202) (0.0280) (0.0377) (0.0194) (0.0271)
Observations 655 697 427 466 376 420
Robust standard errors in parentheses
** p<0.01, * p<0.05
(c) Weighted DW-NOMINATE
D R D R D R
Presidential entry -0.133** 0.108** -0.0967** 0.118**
(0.0220) (0.0256) (0.0299) (0.0382)
Presidential exit 0.0790** -0.122** 0.174** -0.240**
(0.0303) (0.0377) (0.0283) (0.0367)
Constant -0.171** 0.128** -0.169** 0.141** -0.305** 0.281**
(0.0158) (0.0182) (0.0283) (0.0349) (0.0203) (0.0286)
Observations 696 750 430 474 379 427
Robust standard errors in parentheses
*** p<0.01, ** p<0.05
34
-.2
-.2
-.2-.1
-.1
-.10
0
0.1
.1
.1.2
.2
.2.3
.3
.3weighted DW-NOMINATE
weighted DW-NOMINATE
weighted DW-NOMINATE1970
1970
19701980
1980
19801990
1990
19902000
2000
20002010
2010
2010year
year
yearDemocrats
Democrats
DemocratsRepublicans
Republicans
Republicans
(a) Entry
-.5
-.5
-.50
0
0.5
.5
.5weighted DW-NOMINATE
weighted DW-NOMINATE
weighted DW-NOMINATE1970
1970
19701980
1980
19801990
1990
19902000
2000
20002010
2010
2010year
year
yearDemocrats
Democrats
DemocratsRepublicans
Republicans
Republicans
(b) Exit
Figure 8: Weighted DW-NOMINATE
35
