Presidential Leadership and Congressional Seat...
Transcript of Presidential Leadership and Congressional Seat...
Presidential Leadership and Congressional Seat Change:
Evidence from US House Elections (1912-2006)
Hanna Rutkowski Honors Thesis
Department of Political Science Northwestern University
Advisor: Prof. Laurel Harbridge May 15, 2011
Abstract
In the existing body of literature on congressional elections, there is a well-documented trend wherein the president’s party loses seats during midterm elections. It is also clear from the data that there is great variation across presidents in the number of congressional seats lost during the midterms. Some of the existing literature attempts to explain this variation through the theories of Economic Punishment, Systematic Punishment and Presidential Coattails. This thesis integrates those theories with a typology of Presidential Leadership developed by Stephen Skowronek. Through statistical analysis, primarily using a panel data analysis of district level data on congressional elections, this thesis shows that there is in fact variation across presidential leadership type in the size of the midterm decline, and demonstrates how Skowronek’s typology of Presidential Leadership interacts with the already established mechanisms of midterm decline.
Contents List of Figures and Tables p. 1 Introduction (or Preface, if any) p. 2 Acknowledgments (if any) p. 5 Chapter 1. Literature Review p. 6 Chapter 2. Theory and Hypotheses p. 14 Chapter 3. Data and Methodology p. 22 Chapter 4. Results: Aggregate Seat Change p. 35 Chapter 5. Results: Panel Data Models p. 39 Chapter 6. Conclusions p. 61 Bibliography p. 65
1
List of Figures and Tables
Figures:
Figure 1 Graphical Representation of Economic Punishment in On-Year Elections p. 42 Figure 2 Graphical Representation of Economic Punishment in Midterm Elections p. 45 Figure 3 Graphical Representation of Presidential Coattails in On-Year Elections p. 51 Figure 4 Graphical Representations of Presidential Coattails in Midterm Elections p. 53 Tables:
Table 1: Average Seat Change for Different Groups of Presidents p. 25 Table 2: Key Statistics By President p. 36 Table 3: Economic Punishment in On-Year Elections p. 41 Table 4: Economic Punishment in Midterm Elections p. 44 Table 5: Systematic Punishment in On-Year Elections p. 46 Table 6: Systematic Punishment in Midterm Elections p. 47 Table 7: Presidential Coattails in On-Year Elections p. 50 Table 8: Withdrawn Coattails in Midterm Elections p. 52 Table 9: Economic and Systematic Punishment in On-Year Elections p. 55 Table 10: Economic and Systematic Punishment in Midterm Elections p. 56 Table 11: All Three Theories in Midterm Elections p. 59
2
Introduction
The American political system recently experienced a midterm election of historical
proportions. President Obama himself has gone so far as to call the electoral response to his party
a “shellacking (CNN, 2010).” And while the Democratic Party’s seat loss in the 2010
congressional elections was extraordinary in magnitude, the loss of seats itself is not a new
phenomenon. The president’s party historically suffers from a weakening in support during
midterm elections. That decline however, is not uniform in magnitude. For example, in the 1958
midterm elections, President Dwight D. Eisenhower’s Republican party lost a total of 48 seats in
the House of Representatives, while 4 years later, in 1962, John F. Kennedy’s Democrats lost
only 4 House Seats. In very rare instances (1934, 1998, and 2002 from my sample beginning in
1912), the president’s party actually gains seats at the midterm.
There is clearly variation in the number of seats lost by each president. If that variation is
systematic, it is valuable to understand which factors make the greatest impact. An
understanding of when and how the president’s party loses or gains seats in congressional
elections can be incredibly important for legislative strategy. The majority party may prioritize
certain legislative initiatives in the months leading up to an election if it fears a marked decline
in seat control. The minority party may similarly attempt to stall certain of the majority’s
initiatives if it feels it will soon take power itself. Both majority and minority may take increased
measures to counteract factors like a poor economic climate if they have been shown to be
significant determinants of their electoral success. The extent to which congressional agents will
undertake any or all of the above can be informed by an understanding of the very mechanisms
that frequently lead to seat loss of the president’s party.
3
It seems nearly impossible to predict the exact magnitude of a swing in congressional
control, as an incredible amount of local and national nuance goes into the outcome of any
particular race. Valuable insight can be gleaned however from a thorough understanding of the
various factors shown to predispose the president’s party to lose or gain seats. Economic growth
is frequently mentioned as an important factor, as many have theorized that the president’s party
will be punished for a poor economic climate (Abramowitz 1985, Levitt 1994). The theory of
surge and decline, where a popular president is able to bring other members of his party into
office on his coattails, and then loses some of those seats in the following midterm, is another
(Born 1990, Campbell 1991, Levitt 1994). A final theory concerns the systematic punishment of
the president’s party (Levitt 1994). This thesis will argue that presidential leadership is an
important piece of this puzzle, both on its own and in conjunction with other factors.
The analyses that follow will approach the concept of presidential leadership through the
lens of political time, as defined by Stephen Skowronek. They will attempt to demonstrate that
presidents in certain leadership moments may be able to soften the declines suffered by their
parties in off-year elections (midterms). Leaders who can convince the country that they are on
the right track may be able to muster continued support for their party, even when they are not at
the top of the ticket. On the other hand, presidents perceived to struggle might face harsh public
referenda, and pass that frustration onto their parties through the midterm cycle. For Skowronek,
the success or failure of presidential leadership depends on more than personal charisma; it also
depends on the specific conditions they inherit in political time.
Why does this matter? The past two years illustrate the fact that the modern president
may have difficulty enacting his legislative priorities, even with dominant majorities in both
houses of Congress. A reversal in control of the legislature may be enough to derail his agenda
4
entirely. Any president who can minimize his party’s midterm loss will have a tremendous
advantage in this realm of politics. As such, the impact of leadership on the outcome of
congressional elections could have great bearing for the strategic leader of the future. It could
illuminate ways in which the public evaluates the president and inform both the president’s
communication with the voters as well as his relationship and responsibility to his party. At the
same time, it is important to understand the limits of leadership’s ability to influence electoral
outcomes. If Skowronek is correct that presidents are constrained in their influence by the
political time in which they take office, there may be important implications for how they
prepare for and anticipate congressional elections.
This thesis will begin with a review of the existing literature, first addressing those
theories that seek to explain the general pattern of seat loss for the president’s party in midterm
elections. The literature review will briefly introduce the work done by Skowronek in the field of
presidential leadership. Chapter 2 is a theoretical section that elaborates on Skowronek’s theory
in the context of the hypotheses I have developed and will test in this thesis. I then provide an
overview of the data I will analyze and introduce the statistical models I have developed to test
my hypotheses. Chapters 4 through 6 present the results of my analyses, along with implications
and conclusions.
5
Acknowledgements
I would like to thank my adviser Laurel Harbridge for her tireless efforts to help me
throughout the course of this project. I could not have done it without our weekly meetings and
her thoughtful comments on everything I produced. I would also like to thank my friend Lauren
Gramza for the moral support she provided throughout the year- it is always nice to have
someone else in the same boat with whom to commiserate. I want to also thank the MMSS
program for giving me the tools to complete this project. I should note that it was Professor
Daniel Galvin’s class on The American Presidency that first inspired this thesis three years ago.
Finally, I want to thank my parents for the support they’ve provided throughout this process, and
especially my mother for staying up late to read over a final draft.
6
Chapter 1
Literature Review
Though a pattern of presidential coattails and midterm loss in congressional elections has
been substantiated empirically, the magnitude of these effects varies across presidents, and it is
less clear what explains the variation. Although this thesis explores the variation across
presidents, I begin with an overview of existing literature on midterm loss. Three prominent
theories concerning midterm loss for the president’s party in congressional elections are: the
theory of surge and decline through presidential coattails, the theory that the president’s party is
punished for national economic performance, and the theory that the president’s party is
systematically punished, regardless of observable factors. All of these theories have room to
interact with the literature on presidential leadership, and in fact, there is a body of literature
concerning how association with the president and his policies impacts the electoral fortunes of
in-party representatives. My thesis will operate within the intersection of these areas of literature,
with a particular emphasis on presidential leadership style as defined by Stephen Skowronek.
One of the foremost theorists in the area of presidential surge and midterm decline in
congressional elections, James Campbell, explores the difference in seat change in on-year
elections, when there is a simultaneous presidential vote, compared with off-year, or midterm,
congressional elections (Campbell 1991). He hypothesizes that a winning president has a
positive effect on his party’s share of the vote in on-year elections, but that the president’s party
will suffer in the following midterm election cycle. He offers one possible explanation for this
decline in the presence of presidential coattails in on-years and their absence in midterms,
suggesting that popular presidential candidates boost turnout and voting for their party, but when
7
they are absent from the ticket in the following election, these voters return to the opposing party
or abstain from voting altogether (Campbell 1991).
Coattails work in the following way. When the president runs at the top of the ticket in
on-year elections, his personal popularity attracts support that can “spill over” to his party’s
congressional candidates (Campbell 1991, 477). This increase in vote for congressional
representatives, which Campbell argues should be proportional to the president’s margin of
victory, is what is sometimes referred to as the presidential surge. According to Richard Born,
some of the surge is due to the fact that even those people with lower political involvement vote
in presidential elections, casting straight tickets for the party of their preferred presidential
candidate (Born 1990). These voters then abstain in the midterm elections, and those
congressional representatives helped into office by the surge see a decrease in their vote share.
Born also notes that positive popularity is a weaker motivating factor than negative
reactions to a president’s policies. While the personal appeal of a presidential candidate may
draw his enthusiastic supporters in an on-year election, at the following midterms, it is easier to
mobilize a galvanized opposition then a satisfied support base (Born 1990). Both cases: where
uninvolved voters are drawn into a presidential election and vote straight ticket for his party, and
where the president finds it harder to mobilize his supporters in the midterms than the opposition
finds it to fight unpopular policies, can lead to a midterm decline.
In a second explanation, Campbell (1990) explores the idea that midterm elections serve
as a public evaluation of the administration’s performance, a theory further explored by Levitt
(1994). While Campbell broadly explores this effect by measuring whether the decline in off-
years exceeds the surge in on-years, indicating that more seats are lost than those gained due to
the surge, Levitt takes a more nuanced approach. He distinguishes between punishment of the
8
president’s party based purely on economic conditions, and a broader measure of systematic
punishment that occurs independent of coattails, the economy and presidential approval.
The economic punishment theory states that the president’s party benefits from strong
economic performance in presidential election years, but at the midterms, voters use the
economy as a proxy for evaluating the administration’s performance, and punish the president’s
party for below average growth (Levitt 1994). As they cannot vote the president out of office
himself at the midterms, voters take their frustrations about the economy out on members of his
party in congress, where those in the House of Representatives are up for reelection every two
years. Alan Abramowitz further explores this theory in a 1985 paper. He directly tests whether
economic conditions had an impact on the public’s evaluation of the president, and then how that
evaluation impacts the president’s party in midterm elections (Abramowitz 1985). He finds that
economic conditions have effects on presidential evaluation that vary by year, but that in general,
expectations about the future were most salient in the public’s evaluations (Abramowitz 1985).
Systematic punishment refers to the practice wherein voters punish the president’s party
regardless of how the administration has performed on quantifiable criteria like economic
performance (Levitt 1994). The president may even have high approval ratings, but be punished
simply because voters seek a mediating position between the extremes at the midterms, and vote
for candidates that are more moderate (Levitt 1994). Because the president puts a visible face on
his party’s platform and attempts to publically advance his agenda, voters may see the opposing
party as less extreme, or a weaker legislative majority for the president’s party as an important
check on his agenda.
Levitt departs from the traditional time series approach employed by Campbell in his use
of district level panel data, greatly expanding his sample size. His analysis uses deviations in
9
economic growth (measured from the average annual growth in GNP) as a predictor of party
control in individual districts over time (Levitt 1994). His model finds support for all three
theories of midterm decline (coattails, economic punishment, and systematic punishment). It
predicts that economic punishment accounts for a one percent loss of seats for the president’s
party, compared to two percent for withdrawn coattails and two percent for systematic
punishment (1994). The story on systematic punishment is particularly interesting in that Levitt
(1994) finds statistical support for systematic punishment of the incumbent president’s party in
on-year as well as midterm elections. These findings are consistent with the results of
Campbell’s model, though of a much smaller magnitude. Levitt (1994) addresses this
discrepancy in effect size in the introduction to his paper, and argues that existing models predict
effects so large that they far outstrip the actual midterm decline. This provides another reason
why it may be interesting to segment the data by president, as it may provide a more nuanced
picture of the important factors.
Aside from an indirect measure through presidential approval ratings, most empirical
models of midterm decline do not account for aggregate differences in presidential leadership.
There are a number of papers that take a case-study approach to a single president, and his
relationship to midterm elections during his tenure. One such study is by Brady, Cogan, Gaines,
and Rivers (1996). These authors focus on the 1994 congressional elections, when the
Republican Party took control of the US House of Representatives for the first time in 40 years.
They found that while Democrats saw a huge loss of seats, these losses were regionally and
ideologically concentrated. They also found that in districts where the vote for Clinton in the
1992 presidential election was lowest, Republicans successfully ran on anti-Clinton campaigns,
painting the democratic candidates in the unpopular president’s image. They hypothesize that
10
because Clinton ran on a moderate platform and then moved to the left once elected, he left more
moderate and conservative Democratic representatives exposed to defeat (Brady 1996)
Another case study of the effect of presidential leadership is a study of the Eisenhower
congressional cohorts, conducted by Kevin Price (2002). Price sets out to examine the effects of
a particular leadership type defined by Stephen Skowronek, which I will explore later:
preemptive leadership. He focuses on three partisan effects of this leadership style: ideological
characteristics of members of congress, voting behavior, and electoral dynamics, the last of
which is most interesting for my analysis (Price 2002). Price looks at the changing overall
partisan affiliation of congress during each two-year period in Eisenhower’s tenure, as well as
what kind of incumbent candidates were more likely to win, those closer or farther from Ike’s
moderate rhetoric.
What Price finds is that Eisenhower was eventually successful in bringing some like-
minded moderate Republicans into office, but that those gains were temporary, and did not
largely change the makeup of congress. While moderates were advantaged during his tenure, the
move to the center left more conservative Republicans vulnerable to the galvanized efforts of the
Democratic opposition, and Ike’s party lost those seats. Price also notes an important fact of
preemptive leadership. By co-opting some of the opposition’s agenda in their third-way approach
to politics, Preemptive leaders actually legitimize that agenda, here FDR’s New Deal social
programs, and lengthen the opposition’s tenure as the dominant party in politics (Price 2002).
While empirical studies like those offered by Brady (1996) and Price (2002) provide an
important window into the effect of individual presidential leadership on their party’s midterm
success, there is a gap in the literature when it comes to systematically examining this aggregate
relationship across presidents. In order to bridge this gap, this thesis combines insights from the
11
theories of midterm loss with theoretical perspectives on presidential leadership. In his book The
Politics Presidents Make: Leadership from John Adams to George Bush, Stephen Skowronek
develops an influential typology for presidential leaders. He divides presidents into four groups
depending on the vulnerability of the dominant political party, and the president’s affiliation with
that party (Skowronek 1993).
Based on this typology, it is possible to classify the modern presidents into the four
categories put forth by Skowronek. The classification contains two main categories of political
regimes. For Skowronek, a political regime is distinctly partisan, and yet goes far beyond the
agenda advanced at a national convention. In this sense, regimes provide an ideological
undercurrent that defines “the government’s basic commitments of ideology and interest”
(Skowronek 1993, p. 9). For example, when Franklin Roosevelt took power in the midst of the
Great Depression, his New Deal philosophy fundamentally altered the relationship between the
government and the public. Those ideological commitments would persist in the political
discourse for nearly 50 years, throughout the course of the major Democratic regime of the 20th
century. Despite periods when Republicans controlled the presidency, the idea that the federal
government had a responsibility to provide for the people’s basic security was never far from the
forefront of politics.
The first split in Skowronek’s typology occurs based on whether regimes are resilient or
vulnerable. When a regime first begins to dominate the political discourse, the ideology is strong
and the regime is considered resilient. Presidents at the helm during periods of resilience
generally have far more license for change. As Skowronek says: “A president’s authority hinges
on the warrants that can be drawn from the moment at hand to justify action and secure the
legitimacy of the changes effected” (Skowronek 1993, p 18). Simply put, a president in a
12
resilient regime has stronger warrants to draw on. As time continues to pass, the regime begins to
fracture. The unified message about government commitments that brought the regime to power
begins to deteriorate. The process of actually legislating causes difference to come out within the
party structure as different coalition members have different priorities. The public may become
disenchanted with certain aspects of the regime’s message. These factors contribute to a general
weakening of the once resilient regime, until it becomes vulnerable and eventually makes way
for the rise of a new dominant ideology (Skowronek 1993).
Within the group of vulnerable regimes, there are two types of presidents: the
Reconstructive leaders who are opposed to the prevailing ideology, and the Disjunctive leaders
who are affiliated and are in turn vulnerable themselves. Within the resilient group, there are
Preemptive leaders who are opposed to the ideology, and Articulative leaders who are affiliated
and must find ways to innovate within the limits of the regime’s message. Together, these
different types of presidents form waves in political time. Each wave, which persists from the
very beginning of a regime to its end, contains at least one of each type of president, and often
more than one Articulative or Preemptive leader (Skowronek 1993). For example, the
Democratic wave that began with the election of FDR in 1932 encompassed one Reconstructive
president (FDR), three Articulative leaders (Truman, Kennedy, and Johnson), two Preemptive
leaders (Eisenhower and Nixon), and one Disjunctive president (Carter).
This thesis will employ Skowronek’s classification of presidents in the kind of empirical
analysis used by Campbell and Levitt. It will test the aggregate effect of presidential leadership
on congressional elections, particularly in midterm elections, where the president’s party
traditionally suffers. Analyzing the various types of presidents will allow me to explore not only
13
average midterm losses for each type, but also to consider the mechanisms of midterm decline in
the context of presidential leadership.
14
Chapter 2
Theory and Hypotheses
Skowronek employs his typology to discuss recurring trends in political time, presenting
a limited number of illustrative examples. As he notes early on, the idiosyncratic nature of a
relatively small sample of individual actors make generalization difficult (Skowronek 1993, p.
4). It is challenging to parse out general trends in a sample of very individualized actors.
Essentially, the small sample size makes it difficult to determine whether the apparent patterns in
the data are evidence of a wider trend or simply a statistical anomaly peculiar to the available
cases. This is one limitation of this thesis, but if significant results emerge, they may still be
compelling. This theory section will begin with a general description of the way I chose to
approach the existing literature, and will then include an in-depth analysis of each type of
president as well as my hypotheses for empirical study.
My first step in developing a workable theory was to consider the characteristics of each
type of president in Skowronek’s typology in the context of existing theories of midterm decline.
This required an intimate understanding of each leadership type. The specific constraints and
warrants for authority faced by each type of president have important implications for how they
will be judged in the public, and for how that judgment may manifest in their party’s electoral
fortunes. By definition, regime change as considered by Skowronek involves a switch in the
dominant partisan discourse, a change in the way the public perceives “the government’s basic
commitments of ideology and interest” (Skowronek 1993, p. 9). The nature of where presidents
fall in their respective regime’s trajectory has some initial implications for congressional seat
change. For instance, one would expect presidents at the lead end of a new regime to bring with
them an increase in congressional power. This same logic implies that those presidents who
15
come to power at the tail end of a political regime will see a dampening in their party’s
congressional influence. In the zero-sum game of what is an essentially two-party system, for a
new dominant regime to arise, the previous regime must fall out of power.
While suggestive, this first brush with the theory does not go deep enough. While we can
expect those presidents who come first in a new regime - Reconstructive leaders - to experience
congressional gains, and those presidents who come at the tail end of a faltering regime-
Disjunctive leaders - to experience losses, we can say little about the magnitude of those gains
and losses relative to the leaders who come between. Perhaps Disjunctive leaders come to power
at the end of a long decline in their party’s influence, and merely lose the final seats necessary
for a change in dominance. Reconstructive leaders may simply get the ball rolling before turning
the regime over to their successors, who then make larger gains. This shallow look also obscures
the way in which these temporal patterns interact with established mechanisms for decline. It is
here that a deeper look into Skowronek’s theory becomes important.
i. Reconstructive leaders
In The Politics Presidents Make, Skowronek says that Reconstructive presidents “shared
the most promising of all situations for the exercise of political leadership” (Skowronek 1993, p.
37). A prime example of a Reconstructive leader is Franklin Roosevelt, whose New Deal
ideology served as a radical change in course from the Republican discourse of the previous
decades and persisted as an underlying principle for American politics through the regime’s
demise under Jimmy Carter. Because these presidents come to power in moments of great
change, they do not always need to articulate a specific path for the future; at times, it is enough
that they merely do something different. They come to power at a time when “pre-established
commitments of ideology and interest have, in the course of events, become vulnerable to direct
16
repudiation as failed or irrelevant responses to the problems of the day” (Skowronek 1993, p 39).
Past failures give these presidents enormous warrant to disrupt the status quo, and perhaps, will
allow them some leeway in terms of accountability. This brings us to my first hypothesis:
1. Reconstructive leaders will be insulated from systematic and economic punishment,
relative to other leadership types. They will have large coattails, but see smaller declines
in the following midterms.
It is my hypothesis that the repudiative position of Reconstructive leaders will insulate
them from the same kind of economic and systematic punishment suffered by other presidents.
Because their predecessors have failed so miserably, I believe the public will allow these
presidents some flexibility in the actual success of their initiatives. With past failures fresh in
their mind, it may not take an unequivocal success to appease the voters’ wishes. The mere
departure from past disappointment may be enough to be considered progress. As Skowronek
says: “Roosevelt succeeded despite stunning (legislative) defeats because he remained
throughout the sponsor of an alternative to a bankrupt past” (Skowronek 1993, p 28).
Regarding coattail effects, these presidents will be seen as the bearer of a new and fruitful
message, the leader responsible for returning to the values lost under the previous regime
(Skowronek 1993, p. 37). As the recognizable face of change, these presidents should be able to
bring sweeping changes to the makeup of congress, carrying in new representatives under the
auspices of change. In addition, I do not expect that they will see the kinds of declines usually
associated with big coattails. More than any other type, these presidents should be able to argue
that they need solid congressional support in order to make the changes necessary for the future
of the country. Having just suffered the trauma associated with regime failure, the voters will
feel an urgent need for change that later presidents will struggle to regain.
17
ii. Articulative leaders
For those presidents affiliated with a resilient regime - the articulators, or orthodox
innovators - conditions are less favorable. These presidents are tasked with carrying forth the
message set forth by their reconstructive predecessors. They must find a way to make the
message their own while preserving the legitimacy of the regime. Any change they make is a
movement away from the founding message from which they draw their authority. They must
find a way to move forward while controlling the fissures that naturally occur in their base
(Skowronek 1993). An illustrative example of an Articulative president is Lyndon Johnson, who
modified FDR’s New Deal ideals in his Great Society Programs, while also adding innovations
of his own. As Skowronek says, these presidents are “ministers to the faithful” and have an
obligation to adhere to principles and act in a certain manner (Skowronek 1993, p. 41). They
must do all this while preventing the stagnation of their agenda, a daunting task for any leader.
This brings us to my second hypothesis:
2. Articulative leaders will have big coattail effects and see large corresponding midterm
losses. They will also experience systematic and economic punishment, but to a lesser
degree than the disjunctive leaders who follow.
Articulative presidents do just that: articulate the message of the regime. As such, I
hypothesize that their presence on the ticket in on-year elections will help carry their party to
electoral gains. A compelling message brought forth by a strong leader may help tie factions of
the party together in time for elections, as the president’s presence reminds voters of the reason
they put the regime in power in the first place. He serves as a national face for the party, both in
the media and on the ballot. The pomp and circumstance of national conventions during
presidential election years is all about party unity. However, during the midterm election that
18
follows, each candidate seeks an individual identity that will resonate with their voters, seeking
distance from the national party establishment if they believe that will be rewarded in the polls.
The consistent message of the previous election will be lost, allowing the fractures in the party to
come forth and potentially reversing the electoral gains of the previous vote.
I expect that these Articulative presidents will also experience more economic and
systematic punishment than Reconstructive leaders, but less than the Disjunctive leaders who
will close out the regime. They will not be insulated in the same way as their predecessors; the
regime will have reached a point in its progression where the public will expect concrete results.
It will no longer be enough to simply take the country in some nebulously defined new direction,
and voters will look to the performance of the economy as an indicator of administration success.
If growth is down, these president’s policies will appear to be at fault. Voters may also begin to
tire with more of the same message, resulting in a moderate level of systematic punishment for
these types of leaders.
iii. Disjunctive leaders
Disjunctive leaders are those unfortunate presidents who fall at the tail end of a regime in
decline. Affiliated with a vulnerable message, they struggle to find any legitimate authority. To
stick to a failed message is to present oneself as a threat to the future of the nation, but to
repudiate that message is to alienate any remaining partisan support. As Skowronek says, these
presidents face the “very definition of the impossible leadership situation: a president affiliated
with a set of established commitments that have in the course of events been called into question
as failed or irrelevant responses to the problems of the day” (Skowronek 1993, p. 39).
These Jimmy Carters and Herbert Hoovers serve as memorable failures in our political
history, but for Skowronek, they have little choice in their incompetence. Without much
19
legitimate authority, we can hardly expect that these presidents should bring about electoral
gains; rather, they might expect to be systematically punished, regardless of what they actually
do. My third hypothesis follows from these facts:
3. Disjunctive leaders will have limited to non-existent coattails, and thus should see little
midterm decline attributable to withdrawn coattails. They will however, be systematically
punished in all election years, and should experience economic punishment as well.
Unpopular themselves, these unfortunate leaders should not be able to sweep in an on-
year election surge. An interesting consequence is that they may see slightly lower midterm
losses than the Reconstructive and Articulative leaders that came prior in the regime. The surge
and decline theory states that while on-year elections bring a big surge to the President’s party,
the midterm election decline has a corrective effect, bringing party divisions back in line with the
natural order. Since Disjunctive presidents do not get the surge, they should not see that
corresponding decline.
They can however, expect severe systematic and economic punishment. For Skowronek, it
does not matter what these presidents do, they will always be seen as “an easy caricature of all
that has gone wrong” (Skowronek 1993, p. 39-40). This idea corresponds greatly with the theory
of systematic punishment, where the president’s party will suffer congressional losses regardless
of coattails, approval or economic conditions. This notion fits with the image of a helpless
leader, doomed to oversee the final fracturing of a once dominant coalition.
iv. Preemptive leaders
One might easily identify Preemptive leaders in the following way: find any president
caught up in a constitutional crisis or impeachment proceedings- that president is probably a
Preempitve leader (Skowronek 1993, p 44). These opposition leaders in the midst of a resilient
20
regime often overstep their legitimate mandate for authority and find the country calling for their
removal. They are personal leaders who build their base of support across traditional ideological
divisions. They harness dissatisfied components of the dominant regime as well as a portion of
their party’s traditional base (Skowronek 1993). It is this kind of party building that gave rise to
Bill Clinton’s Third Way.
In the process of actually governing, these fragile coalitions become difficult to hold
together. Preemptive leaders can do little without angering some component of their support
base. Their centrist policies are too conservative for the left and too liberal for the right.
Eventually, the coalition fractures beyond repair, leading the public to question the very
foundations of their legitimacy and sweeping the latent regime back into power (Skowronek
1993, p. 44). My final hypothesis is as follows:
1. Preemptive leaders will experience strong coattails and withdraw coattail effects due to
the personal nature of their presidency. They will also be judged on the performance of
their administration, with a strong focus on economic punishment.
Preemptive leaders live and die by their ability to personally build a working support
base. They do not always advocate the traditional platforms of their parties, and thus cannot rely
on ideological identity alone to garner votes. This personal nature of their presidency makes
them particularly susceptible to the effects of coattails and withdrawn coattails. During on-year
elections, they may shepherd like-minded candidates into congressional office, only to see those
representatives fail to get reelected during the following midterms. They are not only the face of
the party during election years; they are the face of their unique preemptive identity. Thus,
during the midterm election cycle, like-minded representatives suffer from the absence of the
leader of both their party and ideology. While congressmen affiliated with other types of
21
presidents have the party structure to fall back on during midterms, these representatives are
without that luxury.
In addition, because they do not always have a traditional issue platform, these presidents
will be judged on the tangible performance of their administration, namely the economy. Where
other presidents might be judged on their ability to advocate for the issues their party cares
about, these presidents appeal to voters with disparate ideological leanings, and thus must
demonstrate progress in a different way. I hypothesize that if the economy does well, they will be
able to maintain support for their policies, but if growth is slow, these leaders will be punished.
22
Chapter 3
Data and Methodology
As previously discussed, if presidents are in fact constrained in their ability to influence
congressional elections by their position in political time, this will have important implications
for their electoral relationships with Congress. The previous literature on midterm decline is
lacking in that it treats all presidents the same, regardless of their position in regime history.
Critical to addressing this deficiency is a model that can incorporate both the existing theories on
midterm elections and Skowronek’s typology. Any such model requires election data and
controls similar to those employed by previous empirical studies, as well as a system for
transferring Skowronek’s broad classification system into a concrete coding scheme. I elect to
explore the data on multiple levels, first taking a more macro level approach that measures
aggregate seat change, and then a district level panel data approach that greatly increases the
sample size, allowing for a far more robust analysis.
I begin by classifying the presidents of the past century in accordance with Skowronek’s
typology. For the most part, Skowronek has done so already in his 1993 book or a follow-up
published in 2008: Presidential Leadership in Political Time. The only presidents not explicitly
classified were Harding and Taft. In these cases, I used the cross-tabulation of regime resiliency
and identity affiliation to classify each leader. Both Harding and Taft were Republican presidents
affiliated with a resilient regime, and thus are Articulative presidents. One limitation in this part
of my study concerned the classification of the last two presidents: George W. Bush and Barack
Obama. Skowronek’s classification relies on determining each president’s place in the broad
trajectory of the regime’s rise and decline. This depends on historical patterns that are difficult to
determine without any future data. It is impossible to tell whether the end of Republican
23
dominance came with the failed presidency of George W. Bush, or whether Obama’s Democratic
insurgence is just a momentary lapse during the reign of a Preemptive president and will end in
the same way as Clinton’s with the return of a Republican president. However, given the huge
seat turnover between these two presidents, I have chosen to classify George W. Bush as a
Disjunctive leader and Barack Obama as a Reconstructive president. The seat loss experienced
by Bush and his complete annihilation in terms of public opinion- his approval rating bottomed
out at 25 percent in November of 2008 (Peters, Presidential Job Approval, 1999-2011) -seem
more reminiscent of a Disjunctive leader than an Articulator.
For each empirical observation in my panel, I created variables representing which type
of president was the incumbent, as well as which type was elected (for midterms, these will be
the same). In all cases where the president left office midway through his term (FDR, Kennedy,
Nixon), I classified the replacement president according to Skowronek’s typology. These
situations present additional complications when considering coattail effects, as the presdident
elected to office in one year, is not always the same as at the following midterm. In that case,
their successor experiences the effect of their withdrawn coattails. After classifying all presidents
according to Skowronek’s typology, I turned to the next step, gathering the remaining data.
i. Data
I compiled the data necessary for my analyses from a variety of sources. The data on the
number of seats held by the president’s party during each congress- used in the aggregate seat
change models- is easily obtained from the office of the Clerk of the House of Representatives
(Office of the Clerk of the House of Representatives). For the panel models, district level
election data come from House General Elections Data, 1876-2008 (Brady et al. 2008). I chose
to focus on House elections alone rather than both the House and Senate because of the fact that
24
each Senate seat comes up for reelection only once every six years. In any given election, only
one third of the Senate seats are up for election. Because my panel data component for House
elections covers every district every two years, it could not easily be combined with data on
Senate elections. Running a model for the Senate alone is complicated by the same reality that
only one third of the seats would be included in every year. Further segmenting the data to
account for these differences makes the sample sizes for analyses of the Senate too small to
achieve meaningful statistical results. For these reasons, I elected to explore only data on House
elections, beginning in 1912, when the House reached its current total of 435 members, and
ending in 2006, the last year for which I have complete data.1
In order to test the presidential coattails hypothesis, I needed data on the popular vote for
the president in all national elections. For on-year elections, the regression includes a measure of
the two-party popular vote share obtained by the winning president. In order to capture
withdrawn coattails during midterm election year models, I used the popular vote share from the
previous presidential election. I obtained these data from The American Presidency Project at
UC Santa Barbara (Peters, Presidential Elections Data, 1999-2011).
To measure economic growth I draw on historical data on Gross Domestic Product
(GDP). There are some temporal limitations to the data on GDP, as before 1929, economists did
not measure all of the current components used to calculate GDP. Any data on GDP prior to this
year are retroactive estimates. The estimates I use in my models come from the Millenial Edition
of Historical Statistics of the United States (HSUS) (Sutch 2006). HSUS publications provide
descriptions of the various estimates that have been created to represent GDP in these earlier
1 The analyses of aggregate seat change actually go through the election of Barack Obama in 2008. District level election data was incomplete for the 2008 congressional elections however, so I elected to use only data through 2006 for the panel data component.
25
years. I chose the Millenial Edition estimates because they were the most comprehensive. The
Millenial Edition estimates are in real 1996 US dollars, and after 1929, represent actual values of
GDP. However, this data set includes only data on GDP up through 2002. After that year, data
come from National Income and Product Accounts (Bureau of Economic Analysis, 2011), and
are in 2005 dollars. Given the fact that the GDP variable is derived from a percentage growth
rate, the different base years for the data should not be an issue. For robustness, I checked growth
rates for some of the later years to ensure that indices for both base years produced consistent
estimates.
ii. Aggregate Seat Change
Before focusing on specific hypotheses, I begin with a quick look at the aggregate trends.
Using a macro level analysis of aggregate seat turnover, conditional on where the president falls
in Skowronek’s typology, this approach provides a first look at the data and is designed to
provide big picture numbers and inform later analyses. I had initially intended to run regression
models, similar to those used in the panel data section, for aggregate seat change as well, but the
small sample size proved prohibitive. Thus the results of this section will be limited to summary
statistics. I first compare the average seat turnover in the House of Representatives for each type
of president.
Table 1: Average Seat Change for Different Groups of Presidents Midterms On-Years*
Skowronek Presidential
Type Average Seat Change Standard
Deviation Average Seat Change Standard
Deviation Reconstructive -6.684
(Min: -18.4; Max: 2.97) 8.321 2.486
(Min: -.460; Max: 4.38) 1.973
Disjunctive -5.181 (Min: -12.0; Max: 1.73)
5.783 -8.941 (Min: -23.02; Max:0 .69)
10.063
Articulative -7.476 (Min: -17.7; Max: -.643)
6.433 3.024 (Min: -9.80; Max: 17.28)
8.455
26
Preemptive -7.559 (Min: -14.5; Max: 1.15)
5.497 .5133 (Min: -4.11; Max: 4.73)
2.773
*On-year elections exclude the only outlier (defined by box plots) in the sample- the 1920 Presidential election year following Woodrow Wilson’s Presidency. This data is excluded from the statistics on Preemptive leaders.
The data reveal a seemingly surprising trend. Disjunctive leaders (those who come at the
tail end of a faltering regime) lose, on average, the fewest seats of any presidential type during
midterm elections. The observation however, fits with the logic of Hypothesis 3, wherein these
leaders do not have large coattails to begin with, and thus do not experience a dramatic decline
when those coattails are later withdrawn. The surge and decline theory indicates that when a
president’s party gains seats on his coattails in a presidential election year, they should then lose
some of those seats in the following midterms. If the president does not have large coattails
during the on-year elections, then there are no gains to translate into losses during the following
midterm. The large majority of seat loss must come from other factors. Further support for this
theory about Disjunctive leaders comes in the data for on-year elections. Here the parties of
Disjunctive leaders barely ever gain seats, and on average lose 8.94 percent of the two-party
seats in congress. Given the fact that these presidents are actually losing seats in on-years, it is no
longer so surprising that they have a lower drop-off in the midterms.
Additionally of note is the fact that Articulative presidents were the only group never to
make gains in the midterms. All other categories (even Disjunctive presidents) saw at least one
president’s party make modest gains in a midterm election. This may simply be an anomaly due
to the small number of presidents covered by this study, but it is nevertheless interesting to note.
Similarly, while Articulative presidents exclusively suffer from declines at the midterms, these
same presidents had the highest average seat gain during on-year elections. The implications for
27
these trends are unclear from these simple summary statistics, and merit further analysis. It will
be interesting to see how they play out in the panel data analysis of the different theories.
iii. Panel Data Component
The second component of my analysis is a panel data study using district level election
data. Each observation consists of the two-party vote share garnered by the Democratic Party
candidate in a given district in a given election year. For instance, if the Democratic candidate
won 55 percent of the vote while the Republican won 40 percent, then the dependent variable in
these analyses would equal 55/(55+40) * 100, or 58 percent. These vote shares are then regressed
on a number of variables, some of which test specific theories, some of which are included
merely as controls. I elect to first test each theory individually, and then to run models that test
the competing hypotheses for midterm decline together. This provides a clear picture of how
each theoretical mechanism operates on its own, followed by a more comprehensive model
designed to prevent the overestimation of each individual effect. As I will explain below, at least
for on-year elections, I cannot correctly test all three theories: economic punishment, systematic
punishment, and presidential coattails, while still running separate analyses for each type of
president in Skowronek’s typology.
The nature of important variables makes it impossible to accurately test all three theories
together while still segmenting the data by presidential type. Both theories of economic and
systematic punishment attribute midterm decline to voter reflection on the performance of the
president elected at time t-l, while the theory of presidential coattails relies on the newly elected
(at time t) president’s ability to draw votes for his party’s congressional candidates. In some
cases, the presidents involved in each theory will be the same. An example is the election that
took place in 2004, when George W. Bush was both the incumbent president who would have
28
suffered systematic and economic punishment, as well as the elected president who would have
provided coattails to Republican congressional candidates. In other cases however, the
incumbent president and elected president will not be the same, nor even of the same party. Take
for example the election that occurred in 1980, when Jimmy Carter was the incumbent but
Ronald Reagan won the election. Here, the three theories attribute their effects to different
presidents. The two punishment theories test the public’s evaluation of president Carter’s
performance over the previous years, while the coattail theory refers to Reagan’s ability to help
elect Republican congressional candidates to office.
Fortunately, this is not a complication that exists during midterm elections. In these years,
the two punishment theories consider the president who has been in office the previous two years
(since t-1), and the coattail theory tests the influence of the same man in the election at time t.
Withdrawn coattails are attributed to the same president who won in the preceding on-year
election. In most2 cases, this man was president for the entirety of the time considered by the
punishment theories. Because of the complications just discussed for on-year elections, I will test
the three theories individually for both on-year and midterm elections, test the two punishment
theories together for both on-year and midterm elections, and then test all three together, but only
for midterm elections. I begin below by specifying the models.
iv. Models for Individual Theories
The first theory I test is the theory of Economic Punishment. Here, the hypothesis is that poor
economic performance over the previous year will cause voters to punish the president’s party in
midterm elections. These models are specified the same way for both on-year and midterm
election years, but there is no reason the coefficients should be constrained to be the same, so I
2 The only exceptions being years where the president died or resigned his office. In these cases, his vice president runs in the midterm elections.
29
run them separately for each type of election. I also ran each model separately for each type of
presidential leader. Equation 1 below gives the specification for Economic Punishment. If the
theory holds, the coefficient on GDP should be positive. In that case, above average growth is
rewarded while below average growth is punished through the loss of vote share.
(1) Demvoteit = α0 +ϕ1Incdemit + ϕ2Increpit + ϕ3GDPt + νit
The second theory is that of Systematic Punishment. The theory says that during
midterms, voters systematically punish the president’s party in congressional elections,
regardless of other factors like economic performance. In order to capture this effect, the model
contains a variable specifying the president’s party. If the theory holds, this variable should have
a negative coefficient during midterm elections. That would indicate that the presence of a given
party’s president in the White House diminishes that party’s vote share in congressional
elections. A negative coefficient for the on-year election model would indicate that systematic
punishment occurs not only in midterms, but in presidential election years as well. Equation 2
below specifies the test for systematic punishment. I run it separately for on-year and midterm
elections, and also run it for each group of presidential leaders. The leadership type of the
incumbent president segments both Equations 1 and 2, as these theories both test punishment of
past performance.
(2) Demvoteit = α0 +ϕ1Incdemit + ϕ2Increpit + ϕ3PresPartyt + νit
The final individual theory to test is the theory of Presidential Coattails. Unlike the
previous two theories, which required a single regression that could be run separately for on-year
and midterm elections, this model necessitates two separate equations:
(3) Demvoteit = α0 +ϕ1Incdemit + ϕ2Increpit + ϕ3PresVotet + νit
(4) Demvoteit = α0 +ϕ1Incdemit + ϕ2Increpit + ϕ3PresVotePrevt + νit
30
The first equation (3) is the specification for on-year elections. In that model, I include a
variable for the two-party share of the national popular vote garnered by the winning president:
PresVote. If the coattail theory holds, this variable should have a positive coefficient. In that
case, winning the presidency would lead to an increase in the vote share captured by that party’s
congressional candidates. The second equation (4) is the specification for midterm elections. The
variable PresVotePrev represents the share of the two-party popular vote captured by the current
president, in the previous presidential election (at time t-1). The theory of withdrawn coattails
predicts a negative coefficient for this variable. The logic is that when the president wins the
popular vote by a larger margin, they will bring to office more representatives from their party
on their coattails, and the more representatives elected in this way, the more will be vulnerable to
the effect of withdrawn coattails in the following midterms. Equations 3 and 4 are run separately
for each type of president. While the incumbent president segments the models testing economic
and systematic punishment, the elected president segments the coattail models in order to reflect
the fact that it is the presence of the president elected at time t that has the effect on
congressional elections in the same year.
v. Models Testing Individual Theories Together
I am able to test the two theories of economic and systematic punishment together for
both on-year and midterm elections. As both theories reflect an evaluation of the past
performance of the same president, they can be tested in the same model while still segmenting
the data by leadership type. Including the two theories in a single model helps to provide more
accurate estimates, as testing each theory separately might overestimate the effect of each
individual variable in turn. I run Equation 5 separately for both on-year and midterm elections,
and then for each group of presidential leaders, segmented by the incumbent president. It
31
includes both PresParty and GDP in order to test both theories simultaneously. These variables
should have effects in the same direction as predicted by the individual models above, but if
there was overestimation in those models the coefficients might now be smaller in magnitude.
(5) Demvoteit = α0 +ϕ1Incdemit + ϕ2Increpit + ϕ3PresPartyt + ϕ4GDPt + νit
Finally, I will test all theories together, but only for midterm elections. As noted above,
during on-year elections, it is impossible to test all three together while still segmenting by
leadership type (some variables refer to the incumbent president and some refer to the elected
one). Because in midterm elections, there is no change in the man holding the presidency, all
variables refer to the same leader. Equation 6 below provides the specification for this final test.
(6) Demvoteit = α0 +ϕ1Incdemit + ϕ2Increpit + ϕ3PresPartyt + ϕ4GDPt +ϕ4PresVotePrevt + νit
I estimated all the above models using OLS regression with district level fixed effects.
These fixed effects help to control for the status quo partisan leaning of each district, as well as
any other district specific characteristics that might influence the outcome of elections, but are
the same for that district in any given year. For instance, we expect the partisan leaning and thus
vote share for the Democratic candidate to be very different in Chicago verses in rural Texas.
Including fixed effects allows the model to account for these differences.
vi. Description of Variables
The models described above contain variables that are both time and district specific.
Panel data analysis allows for variation in both aspects at the same time. Variation across time is
indicated by a subscript of t, while variation across district is indicated by a subscript of i.
Demvoteit for instance, represents the share of the two-party vote gained by the Democratic
candidate in district i at time t, while GDPt in any given year is the same for each district, and
only varies across time.
32
The dummy variables Incdem and Increp measure whether the incumbent candidate in a
given observation is a Democrat or a Republican. For Democratic candidate, Incdem is coded as
1 and Increp is coded as zero (vice versa for Republican candidates). These are included as
important control variables, as incumbency is a well-established advantage for congressional
candidates (Gelman and King 1990). Given that the dependent variable is the share of the two-
party vote garnered by the Democaratic candidate, we would expect Incdem to have a positive
coefficient, and Increp to have a negative coefficient.
PresParty measures which party had control of the White House at the time of the
election and is used to test the theory of systematic punishment. It is valued at 1 if the President
is a Democrat and -1 if the President is a Republican. This allows me to combine two opposing
effects into a single variable, simplifying the functional form of the model (as opposed to using
two dummies). It is also consistent with the method used by Levitt (1994). If the president is a
Democrat, then variable PresParty takes on the value of 1 and the theory of systematic
punishment predicts a negative coefficient representing a decrease in the Democratic vote share.
If the president is a Republican, then PresParty takes on the value of -1, so a negative coefficient
actually represents a positive effect (for ϕ <0, -1(ϕ) >0). This prediction is also in line with the
theory: If presidents are systematically punished, then we expect the presence of a Republican
president to have a negative effect on the Republican candidate’s vote share. This translates into
a positive effect for the Democratic candidate’s vote share, which is equal to 1 minus the
Republican’s vote share.
The two variables PresVote and PresVotePrev are included to test the theory of surge and
decline. In on-year election models, Presvote is the share of the two-party popular vote garnered
33
by the winning presidential candidate, minus 50. With the exception of the election in 20003, this
should be positive for all presidents. It represents the strength of the president’s popularity; the
theory suggests that more popular candidates (those with a larger margin of victory) would have
larger coattails. PresVote is multiplied by -1 for Republican presidents. The logic to this is
similar to that for PresParty. A positive coattail effect for a Republican president should increase
the Republican candidate’s vote share, therefore decreasing the Democratic candidate’s vote
share.
In the midterm election models, PresVotePrev is equal to the value of PresVote in the
previous presidential election. As in the on-year models, it is multiplied by -1 for Republican
presidents. It is expected to have a negative effect, representing the effect of withdrawn coattails.
However, the withdrawn coattails of a Republican president, which would have a negative effect
on Republican candidates, should then have a positive effect on Democratic vote share. If there
is indeed a negative coefficient on PresVotePrev, then multiplying by the negative value for
Republican presidents should yield the expected positive effect on Democratic vote share.
Finally, GDP is measured as the deviation in the growth rate of Gross Domestic Product
(GDP) from the average growth rate over the entire time period studied. I first took the
difference between GDP in year t and year t-1, divided it by total GDP in year t-1, and then
subtracted the average of those values for the whole sample.4 For example, in 1998, GDP was
8.509 trillion US (1996) dollars, while GDP in 1997 was 8.160 trillion US (1996) dollars. This
3 When George W. Bush won the presidency through the Electoral College, but did not win the national popular vote. 4 A note on notation: Thus far, when I have discussed elections in time t and t-1, I have been referring to time in two-year periods. For instance, at time t in 2000, time t-1 refers to 1998. Here, for GDP calculations (script) t and t-1 refer to single year periods.
34
represents a growth rate of 4.28 percent. The average growth rate of GDP from 1912-2006 was
3.28 percent. Therefore, the value of GDP for the year 1998 is equal to 1 (4.28-3.28).
For Republican presidents, the value of GDP is multiplied by -1.The economic
punishment theory predicts a positive coefficient for GDP- the parties of presidents who preside
over years of growth are rewarded, while those who preside over decline are punished. In years
where there was a Republican president and above average growth, we expect the two-party vote
share for Republican congressional candidates to increase, and the share for Democrats to
decrease. Multiplying GDP by -1 for Republican presidents allows above average economic
growth to have a negative effect on the Democratic vote share when the coefficient on GDP is
positive, as the theory predicts. The logic in taking a deviation from the mean is that the public
has a standard in mind of acceptable economic performance, and compares performance over the
previous year with that established standard when evaluating a presidential administration. In
order to isolate economic fluctuations that were due to actual growth rather than inflation, all
GDP data is expressed in real 1996 dollars.
For all of the models, Demvote is the percentage of the two-party vote share won by the
Democratic congressional candidate in district i in year t. It is calculated by dividing the
Democrat’s percentage of the vote by the Democrat’s percentage added to the Republican’s, and
multiplying this ratio by 100. As in a previous example, if the Democratic candidate won 55
percent of the vote while the Republican won 40 percent, then Demvote would equal 55/(55+40)
* 100, or 58 percent. That concludes the methodological exposition of the thesis. I will now
present the results, beginning with the statistics for aggregate seat change.
35
Chapter 4
Results: Aggregate Seat Change
Before I begin an analysis focusing on individual elections, I explore what the broad
swings in congressional control look like empirically. By Skowronek’s formulation, there have
been at least three major regime changes during the period studied in this thesis. The first is a
wave of Republican domination that encompasses the earliest leaders (Taft is the first incumbent
president considered, in 1912), up through and including Herbert Hoover. The election of
Franklin Roosevelt in 1932 marks the beginning of the second wave, one of Democratic control.
This second wave ended with Carter’s defeat at the hands of Reagan, in 1980. And though it is
beyond the scope of what Skowronek considers, this thesis argues that a fourth wave began in
2008, with the election of Barack Obama.
Skowronek’s theory suggests that with each wave of presidents, we should see a build up
of power at the beginning of a regime’s dominance, and then a gradual decline as the regime
fractures and makes way for another to take hold. In general, this would suggest congressional
gains for those presidents who come at the beginning of a regime, but decline in seat share for
those who follow. In fact, the biggest favorable swing might be expected when a Reconstructive
president first office, with the biggest loss expected at the end of a Disjunctive leader’s tenure.
While the regressions that come later in this thesis will consider seat changes during a single
election year, here I consider the aggregate seat change over a president’s entire tenure. The
statistics in Table 2 explore these relations.
36
Table 2: Key Statistics By President Key Statistics By President
President Incoming Seat Gain Outgoing Seat Loss Total Seat Change FDR 23% N/A 6.12%
Reagan 7.92% -0.46% 4.90% Reconstructive Leaders
Obama 5.52% N/A N/A Hoover 7.12% -23% -27.90%
Carter 0.23% -7.92% -11.10% Disjunctive Leaders
W. Bush -0.34% -5.52% -10.50% Taft N/A -9.80% N/A
Harding 14.20% N/A -3.50% Coolidge N/A 7.12% 3.01%
Truman N/A -5.07% -6.81% Kennedy -4.73% N/A -5.40% Johnson N/A -1.05% -3.68%
Articulative Leaders
Bush -0.46% 2.07% -0.14% Wilson 9.80% -14.20% -28.40%
Eisenhower 5.07% 4.73% -6.04% Nixon 1.05% N/A *1.15%
Ford N/A -0.23% *-11.4%
Preemptive Leaders
Clinton -2.07% 0.46% -12.60% Notes: Elections in italics represent outliers within each presidential type N/A is listed where a president did not initially come to office through an election or did not serve their full term. Some figures for Taft come before the frame of analysis. *Indicates a potentially misleading figure- Ford may have been punished for Nixon’s actions when he left office before he could be hurt himself
There are a few important takeaways from the numbers above. First, the table indicates
that Reconstructive leaders see the biggest incoming gains when they are swept into office. On
average, these presidents gain 12.1 percent of the two-party vote share when they are first
elected. No other group of presidents comes close to this figure. Each of these gains has its
corresponding outgoing loss for a Disjunctive leader, who sees a loss of the same magnitude
when they leave office at the end of their term. Table 2 also confirms the hypothesis that
persistent regime gains come only at the beginning of each period of party dominance. With the
exception of Calvin Coolidge, no non-reconstructive president’s party controlled more seats at
the end of their tenure in office than at the beginning (Richard Nixon also technically gained
37
seats over his tenure, but given the electoral punishment suffered by Ford in the following
midterms, I think it is safe to assume he would have lose those modest gains if he had served out
his full term).
Predictably, the Disjunctive presidents fall at the other end of the spectrum from the
Reconstructive leaders. Not one of these presidents from the tail end of a regime saw their party
lose less than 10 percent of the two-party vote share in the House. Additionally of note and
slightly more surprising is the fact that Preemptive leaders also fare poorly over their tenures in
office. These presidents find themselves opposed to a resilient regime; their party is in the midst
of a period of relative weakness, and yet they seem to weaken it further through their brief foray
into power. Rather than detract from the strength of the dominant regime, these opposition
leaders seem to strengthen it further, perhaps even prolonging its time at the forefront of political
discourse. As expected, the results for Articulative presidents fall somewhere in between
Reconstructive and Disjunctive leaders. They do not generally gain over their tenure, but rather
see modest losses in seat control. Their numbers might represent the slow withering of the
dominant regime as internal fractures begin to erode its support base.
A complication in the above analysis, as well as in the panel data analysis of on-year
elections that will follow, is the fact that frequently, the incumbent president for a given election
is not the same as the president elected. This makes it difficult to parse the effects for each
individual president. It is impossible to tell, for instance, if Reconstructive presidents gain so
much when first elected because they bring with them a tidal wave of support, or if what appear
to be their gains are really the massive punishment of Disjunctive leaders. One other
complication along these lines is specific to the transition from Nixon to Ford (mentioned in the
notes for Table 2). Gerald Ford saw a massive loss of seats in his first midterm election.
38
However, it is difficult to determine for certain whether this loss came primarily from his
position as a Disjunctive leader. It seems more likely in fact that at least a portion of the huge
loss was rather retaliation against any vestige of the Nixon administration following his
resignation. In general, while these statistics on aggregate seat change are useful for determining
big picture trends, they do little to explain the mechanisms behind the large difference evident
for different groups of presidents. The more detailed panel data models that follow will attempt
to parse out these effects.
39
Chapter 5
Results: Panel Data Models
I begin the panel data analysis by testing each theory individually. Though it is possible
that this approach will overestimate the effects of each mechanism on its own, it is a valuable
first step in exploring how the theories play out for each presidential type. The panel data
approach greatly expands the number of observations compared to the aggregate seat change
analysis. For the macro level models, each election year contains only one observation, a
measure of the change in the proportion of seats controlled by the president’s party. In contrast,
each election year in the panel data set corresponds with 435 observations, one for each district
represented in the House of Representative.
Even with this increased sample size, when I segment the data in order to run models
separately for each presidential type, there is limited variation for some of the major variables.
For instance, even though there are three Disjunctive presidents (Hoover, Carter, and W. Bush)
in the sample, they only have four midterm election years between them. This means that for the
regressions for Disjunctive presidents during midterms, there will only be four unique values for
variables like GDP and PresVotePrev. Perhaps even more significant are the limitations that
come from the small number of presidents in each category of Skowronek’s typology. With only
two Reconstructive presidents throughout the sample (FDR and Reagan), it is impossible to
determine whether the results that emerge for Reconstructive leaders are simply due to
idiosyncrasies common to these two men, or evidence of a general tendency inherent to the
Reconstructive leadership type.
These realities certainly present some limitations for inference, but I do not believe they
are fatal for my analysis. The idiosyncratic nature of the individual presents provides a challenge,
40
but is also a potential benefit. Franklin Roosevelt and Ronald Reagan were two very different
men, who ruled in very different conditions, with very different priorities. While it is
undoubtedly possible that what appear to be the effects of Reconstructive leadership could
instead be the result of commonalities between these two men, the fact that they are each unique
individuals makes this less likely. In fact, given the vast range of their individual attributes,
evidence of a common trend between them, while not necessarily precise given the
aforementioned limitations, is certainly significant.
i. Economic Punishment
I test each theory separately for on-year and midterm elections, first for the entire sample
and then for leadership type. The first theory is Economic Punishment, which dictates that the
voters will judge the president and his administration based on the country’s economic
performance. If the economy performs well, they may reward the president’s party in
congressional elections. If however, economic performance fails to meet voters’ expectations,
the theory says that they will punish the president’s party for that poor performance. For
reference, Equation 1 below is the model specification for these regressions. The results for on-
year elections can be found in Tables 3, while those for midterms are in Table 4.
(1) Demvoteit = α0 +ϕ1Incdemit + ϕ2Increpit + ϕ3GDPt + νit
The first column represents the results for all observations together, and is not segmented
by presidential type. It represents a baseline by which we observe deviations in the point
estimates for each type of leader. The coefficient on the GDP variable is interpreted as the
amount at which the growth rate for gross domestic product in the year preceding the election
deviated from the average annual growth rate for GDP throughout the sample. Average annual
growth for all the years covered was 3.28 percent. For reference, in a stronger economic climate,
41
where GDP growth was one standard deviation (sd= 5.79) above average, or 9.07 percent, then
the coefficient of .412 implies that the Democratic candidate would receive an additional 2.39
percent of the two-party vote share:
Change in vote share = 0.412 (9.07 - 3.28) = 0.412 * 5.79 = 2.39
Table 3: Economic Punishment in On-Year Elections
(1) All
(2) Reconstructive
(3) Disjunctive
(4) Articulative
(5) Preemptive
Point Estimate
Point Estimate
Point Estimate
Point Estimate
Point Estimate
(t-statistic) (t-statistic) (t-statistic) (t-statistic) (t-statistic) Deviation in Growth Rate of GDP 0.412*** 0.397*** 0.674*** -0.706*** 0.234*** (-14.11) (-6.33) (-14.31) (-3.41) (-4.23) Incumbent is Republican -13.648*** -8.773* -13.107*** -14.688*** -8.185*** (-14.57) (-2.42) (-7.39) (-10.05) (-3.39) Incumbent is Democrat 15.789*** 23.471*** 20.709*** 15.165*** 23.944*** (-16.93) (-6.47) (-11.64) (-10.46) (-9.95) Constant 52.523*** 46.397*** 48.175*** 55.281*** 45.100*** (-56.14) (-12.77) (-27.66) (-39.72) (-18.84) Overall R2 0.546 0.553 0.593 0.523 0.56 N 10251 2151 1238 3470 3392
* p<0.05, ** p<0.01, *** p<0.001
Models 2-5 separate the effects by presidential leadership type. It is interesting to note
that, at least for these point estimates, Disjunctive presidents experience the most severe
economic punishment in on-year elections. This is in line with my hypothesis about these
leaders. As those presidents who come at the tail end of the regime, Disjunctive presidents can
seemingly do nothing right, and will be given no leeway when it comes to observable metrics of
performance like economic growth. The results in Table 3 for Articulative presidents are
puzzling. These presidents have a very high constant relative to the other groups, representing
the share of the two-party vote they would be expected to capture if all the other variables took
on the value of zero. Their coefficient for economic growth is in the opposite direction of what
we expect. There seems to be some strange interaction going on for these presidents, wherein a
42
very high constant is balanced by the negative coefficient on economic growth. Explaining this
anomaly is beyond the scope of this project.
I also provide a graphical representation of the results for Economic Punishment. In
Figure 1 below, I use coefficients provided by the model and the standard deviation of GDP
growth to compute vote share levels for the Democratic candidate, given average growth, growth
that is one standard deviation above average, and growth that is one standard deviation below
average. One should note that in fact, standard deviations as well as average growth levels may
vary somewhat by presidential type, and thus these calculations are primarily for interpretive
value. They should not be interpreted as predictive figures. For simplicity, I assume the House
seat was open in the election. This allows both incumbent variables to take on the value of zero.
Figure 1: Graphical Representation of Economic Punishment in On-Year Elections
There are some interesting points to note in Figure 1. While Articulative presidents
achieve 50 percent of the two-party vote with average levels of growth, Disjunctive and
48.67 52.03 51.24 46.44 46.69 48.68
54.75
45.27 44.72 45.32
58.27
44.11
0.00
10.00
20.00
30.00
40.00
50.00
60.00
70.00
Recon Disjunc Artic Preempt
On-Year Elections
One SD Above
Average
One SD Below
43
Preemptive presidents must achieve much higher levels of growth in order to gain a majority of
votes. Preemptive presidents do not even meet the 50 percent mark for growth rates that are one
standard deviation above average. While the coefficients in Table 3 are illuminating, the graph in
Figure 1 throws into sharp relief the implications the variation across presidents can have for
actually winning congressional elections.
Table 4 presents the results of regressions testing the theory of Economic Punishment in
midterm elections. Immediately, we can see that the coefficient on GDP for the whole sample is
less than half of what it was for on-year elections. In fact, this effect seems to have gotten
smaller for all presidential types except Preemptive leaders, who see the opposite change. These
presidents had the smallest effect in on-year elections, but now have the largest point estimate.
This seems to indicate that Preemptive leaders are judged particularly strongly on their economic
performance during midterms, which is line with the fourth hypothesis outlined in the theory
section. Because these leaders tend to transcend traditional party divisions when building their
coalitions, it is difficult for the public to judge them on conventional criteria such as how they
are advancing the party platform. These results give credence to the fact that economic
performance, which transcends party lines just like these presidents often do, will be a
particularly important factor in the evaluation of Preemptive leaders.
Also in line with my theory is the fact that Reconstructive leaders see the smallest
significant effect. This seems to support the idea that these presidents are given some leeway in
their concrete performance. For presidents who come to power in these pivotal moments of
political time, it may be enough to simply move the country in a different direction from their
predecessor, whose methods had been so thoroughly discredited. A final note from the data in
Table 4 is the fact that Articulative leaders do not have a significant effect for Economic
44
Punishment in midterm elections. The unusual results regarding their constant and opposing sign
for GDP that were present for on-year elections are absent during midterm elections. In general,
the data on Economic Punishment for Articulative presidents is fairly inconclusive.
Table 4: Economic Punishment in Midterm Elections
(1) All
(2) Reconstructive
(3) Disjunctive
(4) Articulative
(5) Preemptive
Point Estimate
Point Estimate
Point Estimate
Point Estimate
Point Estimate
(t-statistic) (t-statistic) (t-statistic) (t-statistic) (t-statistic) Deviation in Growth Rate of GDP 0.202*** 0.185*** 0.416*** 0.058 0.466*** (-8.77) (-4.32) (-5.35) (-1.35) (-9.35) Incumbent is Republican -10.134*** -13.683*** -8.684** -8.508*** -12.276*** (-9.96) (-8.02) (-3.28) (-3.75) (-5.21) Incumbent is Democrat 20.158*** 17.868*** 30.711*** 22.327*** 19.320*** (-19.88) (-10.78) (-11.62) (-9.88) (-8.25) Constant 49.612*** 51.882*** 42.180*** 47.656*** 51.473*** (-48.2) (-31.38) (-16.31) (-21.25) (-22.18) Overall R2 0.535 0.477 0.628 0.534 0.534 N 10173 2158 1642 3035 3338
* p<0.05, ** p<0.01, *** p<0.001
As I did for on-year elections, Figure 2 below presents a graphical interpretation of the
effect of Economic Punishment, using coefficients and standard deviations to depict Democratic
vote share levels. Noteworthy is the fact that Reconstructive presidents still attain the share of
votes necessary to win an election, even when they preside over economic growth that is a full
standard deviation below average. This is in sharp contrast to Disjunctive presidents, who fail to
capture 50 percent of the two-party vote, even when they preside over growth that is a full
standard deviation above average. This lends further support to my theory that Reconstructive
presidents will be insulated somewhat from poor economic performance, while Disjunctive
presidents can seemingly do nothing to insulate themselves from seat loss.
45
Figure 2: Graphical Representation of Economic Punishment in Midterm Elections
ii. Systematic Punishment
The second theory I test is the theory of Systematic Presidential Punishment. Here, the
theory states that the president’s party will suffer losses in midterm elections, regardless of their
economic performance or any other observable factors. Voters simply vote against the
president’s party in midterm elections. If the theory holds, in midterm election models there
should be a negative coefficient on the PresParty variable, which is equal to 1 if the president is
a Democrat and -1 if the president is a Republican. It is less clear what the theory predicts for on-
year elections, but as the results below show (Table 5), there are some interesting results for
different types of leaders. For reference, the Equation 2 specified the model as follows:
(2) Demvoteit = α0 +ϕ1Incdemit + ϕ2Increpit + ϕ3PresPartyt + νit
52.93
44.53 47.66
54.11 51.74
41.87
51.12 50.56
39.20
48.13
0.00
10.00
20.00
30.00
40.00
50.00
60.00
Recon Disjunc Artic Preempt
Midterm Elections
One SD Above
Average
One SD Below
46
Table 5: Systematic Punishment in On-Year Elections
(1) All
(2) Reconstructive
(3) Disjunctive
(4) Articulative
(5) Preemptive
Point Estimate Point Estimate
Point Estimate
Point Estimate
Point Estimate
(t-statistic) (t-statistic) (t-statistic) (t-statistic) (t-statistic) President's Party -0.003 2.234*** -3.874*** 1.963*** -2.006*** (-0.02) (-7.23) (-8.67) (-7.95) (-8.33) Incumbent is Republican -15.007*** -9.102* -18.208*** -16.775*** -7.148** (-15.89) (-2.52) (-10.13) (-11.74) (-2.98) Incumbent is Democrat 14.469*** 23.396*** 15.820*** 13.212*** 24.276*** (-15.4) (-6.47) (-8.77) (-9.28) (-10.17) Constant 54.554*** 47.313*** 54.644*** 56.585*** 44.602*** (-58.27) (-13.14) (-31.79) (-40.79) (-18.78) Overall R2 0.539 0.556 0.554 0.527 0.565 N 10251 2151 1238 3470 3392
* p<0.05, ** p<0.01, *** p<0.001
The interpretation of the coefficient for PresParty is more straightforward than it was for
GDP in the models of economic punishment. It is simply the additional percent (or lost percent)
of the two-party vote share awarded to congressional candidates from having a president from
their party in the White House. Beginning with the pooled sample of all presidential types, it is
notable that the PresParty variable is not significant and has a very small point estimate,
indicating that there is no general pattern of systematic punishment in on-year elections.
However, the coefficients for each presidential type are illuminating. For instance, the positive
coefficient of 2.234 for Reconstructive leaders, representing an extra 2.234 percent of vote share
for the Democratic congressional candidate, suggests that these presidents are rarely punished in
on-year elections, rather, their party benefits simply from having its leader in office. The same is
true, but to a lesser extent (a smaller point estimate) for Articulative leaders, while Disjunctive
and Preemptive leaders both have significant negative coefficients. This analysis highlights the
importance of considering presidential leadership. Due to the conflicting effects on these
different types of leaders, the overall effect is insignificant, and yet, in one way or another,
47
having a president from your party in office makes a difference for congressional candidates
running under each of type of president.
One potential explanation for these results comes from where these presidents fall in the
cycle of political time defined by Skowronek’s typology. Reconstructive leaders are always
followed by a president from their own party. They preside at times when their regime is perhaps
at its most resilient, following a whole-hearted repudiation of the other party. Articulative leaders
also belong to a resilient regime, though slightly less so than the Reconstructive leaders, as their
coalitions have had time to fracture. On the other hand, a president from the other party always5
follows Preemptive and Disjunctive leaders. Hence, we would at least expect them to experience
losses in the final election in which they were incumbents. This might at least in part explain
their negative coefficients.
Table 6: Systematic Punishment in Midterm Elections
(1) All
(2) Reconstructive
(3) Disjunctive
(4) Articulative
(5) Preemptive
Point Estimate Point Estimate
Point Estimate
Point Estimate
Point Estimate
(t-statistic) (t-statistic) (t-statistic) (t-statistic) (t-statistic) President's Party -2.007*** -0.512 -2.954*** -0.654* -4.525*** (-13.71) (-1.51) (-7.05) (-2.40) (-18.43) Incumbent is Republican -11.312*** -14.188*** -6.162* -9.213*** -12.423*** (-11.14) (-8.23) (-2.35) (-4.05) (-5.54) Incumbent is Democrat 19.577*** 17.200*** 33.822*** 22.051*** 18.775*** (-19.42) (-10.35) (-12.82) (-9.76) (-8.39) Constant 50.752*** 53.334*** 39.681*** 48.148*** 52.447*** (-49.56) (-32.4) (-15.24) (-21.42) (-23.56) Overall R2 0.541 0.474 0.632 0.536 0.56 N 10173 2158 1642 3035 3338
* p<0.05, ** p<0.01, *** p<0.001
5 Note the exceptional Preemptive case of Nixon, who is coded as being “followed” by Ford, though this might be interpreted as merely a continuation of Nixon’s administration.
48
The results for midterm elections are also illuminating, and generally fit the theory. Note
that there is support for the theory of Systematic Punishment in the model containing the entire
sample (Model 1). The coefficient of -2.007 indicates that on average, congressional candidates
capture 2.007 percent less of the vote share when their own party holds the presidency. For
example, a Democratic candidate running in an average district when a Republican was president
would garner 50.75 percent of the vote, while the same candidate running in the same district
when a Democrat was in the White House would garner only 48.74 percent. While the number
2.007 might appear small, in this case it represents the difference between a candidate that wins
and one who loses.
As hypothesized, Reconstructive presidents (Model 2) appear to be insulated from this
kind of systematic punishment. The coefficient on PresParty is statistically insignificant. My
hypothesis was that in their place at the helm of a newly powerful regime, these presidents would
be given some leeway in their performance. Remember from Skowronek that these presidents
have enormous warrants for change because the previous regime is so thoroughly discredited. As
long as they do something different that their predecessor, they will be perceived as moving the
country in the right direction (Skowronek 1993, p. 38). In contrast to Reconstructive presidents,
all other presidential types suffer, but at differing degrees. As expected, the coefficient for
Disjunctive leaders (Model 3) is relatively large. These presidents, who fall at the tail end of a
vulnerable regime, are likely to see their party lose, regardless of what they actually do in office.
The size of the coefficient for Preemptive leaders is difficult to explain. These presidents
seem to be judged particularly harshly at the midterms. The large coefficient for Systematic
Punishment may be a reflection of the general dissatisfaction with these presidents’ middle
ground approach to policy. Because they integrate aspects of both party’s platforms into their
49
message, but do not fully cater to either group, these presidents often find that no one is happy
when they actually stick to that middle ground (Skowronek 1993).
In general, the results of the models on Systematic punishment demonstrate the utility of
separating the presidents by their leadership type. For the on-year elections, the size and
direction of the effect on PresParty vary across leadership type, and these varied effects cancel
each other out in the full model. While the effects are all in the same direction for midterm
elections, they still vary in magnitude, with Reconstructive and Articulative presidents seemingly
more insulated from systematic punishment.
iii. Presidential Coattails: Surge and Decline
The final theory to test individually is the theory of Presidential Coattails. Here, the
general hypothesis is that the explicit presence of the president on the party ticket in on-year
elections will bring additional voters both into the election and into alignment with his party.
Two years later, when the president is no longer physically on the ballot, some of those voters
who were pulled into the previous election will abstain, and those who were drawn to the
president’s party by his personal appeal will return to the opposing party’s ranks. Those
congressional candidates who won narrow victories because of presidential coattails will be
exposed in the following midterms, and there will be realignment back towards the opposing
party.
If the theory holds, there should be positive coefficients on PresVote in on-year elections
and negative coefficients on PresVotePrev in midterms. I present the results for on-year elections
in Table 7. The regression for the entire sample has the predicted effect (see Model 1). The
coefficient of 0.316 can be interpreted as follows: PresVote represents the surge to the winning
president. It measures how large the majority was for the winning candidate by taking their share
50
of the two-party popular vote, and subtracting 50. For instance, if a Democrat won the
presidential election with 60 percent of the vote, this would translate into a value of 10 for
PresVote. The leadership group specific coefficients are generally in line with my hypotheses.
For instance, the largest coattails appear to be for Reconstructive and Articulative presidents.
Hypothesis 1 predicts that Reconstructive presidents would have large coattails, as they
tend to be popular leaders who usher in sweeping changes to the dominant political rhetoric.
Hypothesis 3 predicts large coattails for Articulative presidents as they provide a unified
message for their party while running for president. Also as expected, there is no significant
coattail effect for Disjunctive presidents. Given their lack of personal popularity and particular
political moment, when the regime has mostly run its course, it is not surprising that these
presidents are unable to carry additional member of their party into congressional office. The
coefficient for Preemptive leaders falls somewhere in between: indicating the presence of some
coattails for these presidents, but not as large as those for the two other types.
Table 7: Presidential Coattails in On-Year Elections
(1) All
(2) Reconstructive
(3) Disjunctive
(4) Articulative
(5) Preemptive
Point Estimate
Point Estimate
Point Estimate
Point Estimate
Point Estimate
(t-statistic) (t-statistic) (t-statistic) (t-statistic) (t-statistic) President's Share of Vote in the Election 0.316*** 0.413*** 0.093 0.454*** 0.088** (-18.37) (-11.5) (-1.18) (-14.26) (-2.78) Incumbent is Republican -12.907*** -13.328*** 3.637 5.595 -16.530*** (-13.84) (-7.95) (-0.81) (-1.42) (-12.66) Incumbent is Democrat 16.111*** 14.382*** 38.695*** 39.321*** 13.521*** (-17.4) (-8.69) -8.64 (-10.01) (-10.45) Constant 52.898*** 54.403*** 31.822*** 33.422*** 55.548*** (-57.27) (-33.5) (-7.12) (-8.54) (-44.15) Overall R2 0.549 0.507 0.646 0.579 0.523 N 10251 2591 1611 2593 3456
* p<0.05, ** p<0.01, *** p<0.001
51
Using a similar method of interpretation as used above for the models of economic
punishment, I calculate vote shares yielded for each presidential type at levels of PresVote that
are average, one standard deviation above average, and one standard deviation below average.
The results are below in Figure 3. Again, these values are not meant to be exact, as standard
deviations in PresVote are different for each presidential type. One trend that is immediately
evident from Figure 3 is the fact that in no case do coattails make the difference between
candidates that are above or below 50 percent of the two-party vote share. Both Reconstructive
and Preemptive leaders find themselves easily above 50 percent even for coattails that are one
standard deviation below average, while Disjunctive and Articulative leaders are far below 50
percent even with coattails that are one standard deviation below average.
Figure 3: Graphical Representation of Presidential Coattails in On-Year Elections
During midterm elections there is evidence of withdrawn coattails for all leadership
types, as well as the full sample, evidenced by the negative coefficient on PresVotePrev. Using
the coefficient for the whole sample in model 1 as a reference, we can compare the point
54.97
31.82 34.05
55.64 56.95
36.23
56.06 58.9
38.4
56.48
0
10
20
30
40
50
60
70
Recon Disjun Artic Preempt
Two-Party Vote Share to Dem
ocrat
Presidential Coattails in On-Year Elections
One SD Below
Average
One SD Above
52
estimates for the other types in Table 8 below. Interestingly, Reconstructive presidents have the
smallest withdrawn coattail, contrary to my predictions. Given these presidents charismatic
nature and the dramatic rhetorical changes they represent, I expected them to have large coattails,
as well as a large corresponding drop-off. However, these findings may indicate that these
presidents are so well received that they draw support to their party, even when they are not
actually running in the election themselves, isolating congressional candidates from some of the
effects of withdrawn coattails.
Table 8: Withdrawn Coattails in Midterm Elections
(1) All
(2) Reconstructive
(3) Disjunctive
(4) Articulative
(5) Preemptive
Point Estimate
Point Estimate
Point Estimate
Point Estimate
Point Estimate
(t-statistic) (t-statistic) (t-statistic) (t-statistic) (t-statistic) President's Share of Vote in Previous Election -0.258*** -0.087* -0.725*** -0.181*** -0.523*** (-14.02) (-2.19) (-7.90) (-5.89) (-16.25)
Incumbent is Republican -11.854*** -14.427*** -9.402*** -9.357***
-13.957***
(-11.64) (-8.35) (-3.59) (-4.15) (-6.11) Incumbent is Democrat 19.231*** 17.107*** 30.512*** 22.596*** 17.577*** (-19.06) (-10.3) (-11.67) (-10.05) (-7.73) Constant 51.041*** 53.581*** 42.624*** 47.429*** 53.760*** (-49.81) (-32.42) (-16.66) (-21.26) (-23.77) Overall R2 0.542 0.474 0.634 0.542 0.553 N 10173 2158 1642 3035 3338
* p<0.05, ** p<0.01, *** p<0.001
Also of note is the fact that Disjunctive leaders have the largest effect from withdrawn
coattails, followed by Preemptive leaders. In fact, the general trend is that those groups of
presidents who had the largest coattails in on-year elections had the smallest coefficients for
withdrawn coattails in midterm elections. There are two possible explanations for this somewhat
counterintuitive trend. The first is that there is far less variance in the presidential vote for
Disjunctive leaders. If for instance, these presidents consistently barely win elections and thus
PresVotePrev is very small, then there may actually be a small effect even with the large
53
coefficient. The second possible explanation is that the coattail variable is picking up an effect
from an excluded factor, such as one of the punishment variables. If these variables are
correlated within the different groups of presidents, then the exclusion of one may bias the
coefficient of the other. The results of the models that test all three theories together during
midterms, presented in Table 11 later in this section, may provide some additional clarity.
Figure 4 below presents the same graphical interpretation, this time for midterm
elections, using the standard deviation and average of PresVotePrev for the entire sample. In
contrast to the depiction for on-year elections, withdrawn coattails do make the difference
between gaining a majority of the two-party vote share or not, but only for Disjunctive
presidents. The other three groups of presidents are either always below or always above 50
percent for the various levels of withdrawn coattails depicted in the graphic.
Figure 4: Withdrawn Coattails in Midterm Elections
In general, the data on coattails are consistent with my hypotheses for the surge
component during on-year elections, but are more puzzling for the decline that occurs in midterm
elections. It is especially perplexing that those types of presidents that had the smallest coattails
52.61
34.56
45.42 47.94
53.03
38.03
46.29 50.45 53.45
41.51
47.15 52.96
0
10
20
30
40
50
60
Recon Disjunc Artic Preempt
Two-Party Vote Share to Dem
ocrat
Withdrawn Coattails in Midterm Elections
One SD Above
Average
One SD Below
54
in on-year elections also had the largest effects from withdrawn coattails during the midterms. I
would hypothesize that some of these unusual effects come from the fact that the variables
designed to measure punishment are omitted from these models. The model may be
misattributing a loss in vote share that actually comes from Systematic or Economic Punishment
to a loss from withdrawn coattails. The results of the model testing all three theories together,
presented in Table 11 later on, provide some insight to this effect.
iv. Combined Punishment Models
I next test the two punishment theories simultaneously. As mentioned in Chapter 3: Data
and Methodology, I cannot test all three models together because of differences in which
president is being considered by the punishment theories verses the coattails theory during on-
year elections. Table 9 presents the results of the two punishment theories together for on-year
elections. Testing the two theories together does not change the expected direction of the effects.
The model (originally specified in Equation 5) is as follows:
(5) Demvoteit = α0 +ϕ1Incdemit + ϕ2Increpit + ϕ3PresPartyt + ϕ4GDPt + νit
GDP should still have a positive coefficient, indicating rewards for above average growth and a
decline in vote share for below average growth. The theory for Systematic Punishment makes no
real predictions for on-year elections, yet the models testing this theory alone found significant
effects for every type of president, but of different magnitudes and in different directions. If the
individual models overestimated the effects of each type of punishment on its own, we would
expect the coefficients to decrease in magnitude for the combined model.
55
Table 9: Economic and Systematic Punishment in On-Year Elections
(1) All
(2) Reconstructive
(3) Disjunctive
(4) Articulative
(5) Preemptive
Point Estimate
Point Estimate
Point Estimate
Point Estimate
Point Estimate
(t-statistic) (t-statistic) (t-statistic) (t-statistic) (t-statistic) President's Party -0.329* 2.633*** -0.004 2.643*** -2.439*** (-2.36) (-3.5) (-0.01) (-10.03) (-9.86) Incumbent is Republican -13.412*** -9.265* -13.108*** -15.001*** -7.010** (-14.24) (-2.56) (-7.37) (-10.42) (-2.95) Incumbent is Democrat 15.999*** 23.294*** 20.708*** 14.637*** 24.233*** (-17.08) (-6.43) (-11.64) (-10.24) (-10.23) Deviation in Growth Rate of GDP 0.423*** -0.088 0.674*** -1.523*** 0.378*** (-14.31) (-0.58) (-11.04) (-6.95) (-6.73) Constant 52.284*** 47.623*** 48.176*** 56.521*** 44.344*** (-55.57) (-13.09) (-27.64) (-41.01) (-18.8) Overall R2 0.546 0.556 0.593 0.532 0.569 N 10251 2151 1238 3470 3392 * p<0.05, ** p<0.01, *** p<0.001
The results presented in Table 9 above are not entirely predictable. The first column (Model 1)
represents the results for the entire sample, and fits the theoretical implications. There is a
positive coefficient on GDP, representing rewards for above average economic performance and
punishment for below average growth. There is also a small negative coefficient on PresParty,
indicating that a small amount of systematic punishment occurs in on-year elections as well. The
coefficients take on a similar pattern for Preemptive presidents, while the other groups present
some interesting differences. For instance, neither Reconstructive nor Articulative presidents
appear to experience systematic punishment in on-year elections; rather, they seem to benefit
from having a president of their party. This is not entirely unexpected, as the two punishment
theories pertain primarily to effects during the midterms, but presents an interesting difference
that exists between different leadership types. GDP seems to be most important for Preemptive
and Disjunctive leaders (who have the largest point estimate), while it has a puzzling negative
effect for Articulative leaders, suggesting that their parties benefit electorally from economic
56
downturn. It has no significant effect for Reconstructive presidents, which fits with the idea that
they are somewhat insulated from concrete expectations of performance.
Table 10: Economic and Systematic Punishment in Midterm Elections
(1) All
(2) Reconstructive
(3) Disjunctive
(4) Articulative
(5) Preemptive
Point Estimate
Point Estimate
Point Estimate
Point Estimate
Point Estimate
(t-statistic) (t-statistic) (t-statistic) (t-statistic) (t-statistic) President's Party -1.880*** -0.814* -2.582*** -0.661* -4.329*** (-12.79) (-2.37) (-6.05) (-2.42) (-15.73) Incumbent is Republican -11.191*** -14.158*** -7.389** -9.022*** -12.945*** (-11.05) (-8.25) (-2.81) (-3.96) (-5.71) Incumbent is Democrat 19.726*** 17.760*** 32.833*** 22.146*** 18.409*** (-19.61) (-10.71) (-12.45) (-9.8) (-8.17) Deviation in Growth Rate of GDP 0.167*** 0.204*** 0.311*** 0.06 0.085 (-7.27) (-4.69) (-3.96) (-1.4) (-1.59) Constant 50.343*** 52.180*** 39.671*** 48.067*** 52.760*** (-49.21) (-31.5) (-15.31) (-21.38) (-23.59) Overall R2 0.543 0.479 0.635 0.536 0.56 N 10173 2158 1642 3035 3338
* p<0.05, ** p<0.01, *** p<0.001
The regression results for midterm elections, presented above in Table 10, are perhaps
more important for this study, as the punishment theories pertain primarily to midterm elections.
Here, we again expect a positive coefficient on GDP, as well as a negative coefficient on
PresParty, indicating the presence of systematic punishment. Here we see that the regression for
the whole sample is almost completely in line with the predictions of the punishment theories.
One interesting observation is the fact that economic performance seems to be more important in
presidential election years than at the midterms. This was true as well in the models that only
tested economic punishment. There are fewer types of presidents for which GDP is significant
during midterms, and when there are significant effects, the point estimates are around half of
what they were in the results for on-year elections.
57
In contrast to the models that considered on-year elections, for midterms all four
presidential types exhibit some effects from systematic punishment, with the strongest effect
coming for Preemptive leaders. These presidents also had the largest point estimates for
regression testing Systematic Punishment alone. Before, I reasoned that this might be the result
of a general dissatisfaction with the middle ground policies these presidents implement, which
tend to make no one happy (Skowronek 1993). Both Reconstructive and Articulative presidents
experienced systematic punishment at lower levels than the full sample, while Disjunctive
leaders saw more severe effects (yet not as strong as for Preemptive leaders).
A final interesting point to note about the combined punishment models is the overlap
with the models testing each punishment individually. There does not seem to be a marked
decrease in the effect size for each type of punishment in the combined models. This indicates
that the individual models did not grossly overestimate each theory on its own and lends
credence to the fact that they are useful for inference. Had the effects been much smaller in the
combined models than for each individually, there would likely have been some omitted variable
bias in the individual models, and we would need to exercise a good deal of caution with those
results. Fortunately, this does not seem to be the case.
v. All Three Theories Together: Midterm Elections Only
Finally, I run a model testing all three theories simultaneously, but only for midterm
elections, due to the complications discussed in Chapter 3. Testing all three theories together
should help to avoid overestimation of each individual theory. Equation 6 below specifies the
model and the results will follow in Table 11.
(6) Demvoteit = α0 +ϕ1Incdemit + ϕ2Increpit + ϕ3PresPartyt + ϕ4GDPt +ϕ4PresVotePrevt + νit
58
The first thing to note is the first column (Model 1), which represents the results for the whole
sample of presidents. All three theories are represented here and have their predicted effects,
even when tested together. The results for each individual group of presidents are generally less
consistent with expectations.
Let us first look at the results for Systematic Punishment. The first row represents the
point estimates for PresParty, the test for Systematic Punishment. While Reconstructive and
Preemptive presidents exhibit the expected negative coefficients, both disjunctive and
Articulative presidents have positive coefficients, indicating the opposite effect from what is
predicted by the theory. In the combined punishment models just presented in Table 10, these
presidents had coefficients with the expected negative effect. This is a case where the addition of
the coattail theory dramatically changes the effects of the other theories, though as I will note
below, I can skeptical of these results.
The results for Economic Punishment are also mixed. While the positive coefficient on
GDP appears as expected for Reconstructive and Articulative leaders, there is no significant
effect for Preemptive presidents and Disjunctive presidents have a puzzling negative effect,
indicating improved vote share for below average growth. This is another case where one of the
punishment theories had its predicted effect in the individual model and that for combined
punishment, but had an opposing effect in this final model. One possible explanation of these
unusual effects will follow.
59
Table 11: All Three Theories in Midterm Elections
(1) All
(2) Reconstructive
(3) Disjunctive
(4) Articulative
(5) Preemptive
Point Estimate
Point Estimate
Point Estimate
Point Estimate
Point Estimate
(t-statistic) (t-statistic) (t-statistic) (t-statistic) (t-statistic) President's Party -1.101*** -3.309* 3.519*** 2.784*** -3.396*** (-4.55) (-2.22) (-3.61) (-5.34) (-7.39)
Incumbent is Republican -11.562*** -13.710*** -11.545*** -7.795***
-13.013***
(-11.38) (-7.91) (-4.34) (-3.45) (-5.74) Incumbent is Democrat 19.505*** 18.154*** 28.779*** 24.154*** 18.308*** (-19.38) (-10.85) (-10.8) (-10.72) (-8.13) Deviation in Growth Rate of GDP 0.153*** 0.285*** -2.894*** 0.118** -0.084 (-6.57) (-4.45) (-6.17) (-2.75) (-0.98) President's Share of Vote in the Previous Election -0.125*** 0.298 -4.384*** -0.458*** -0.215* (-4.06) (-1.72) (-6.93) (-7.73) (-2.54) Constant 50.581*** 51.220*** 49.501*** 45.030*** 53.068*** (-49.39) (-29.31) (-16.93) (-19.91) (-23.71) Overall R2 0.544 0.479 0.645 0.545 0.561 N 10173 2158 1642 3035 3338
* p<0.05, ** p<0.01, *** p<0.001
Before I discuss the limitations of this full model, I turn to the theory of Withdrawn
Coattails. Aside from the statistically insignificant positive coefficient for Reconstructive
leaders, all the presidential leadership types demonstrate the negative effects of withdrawn
coattails. The absence of a significant effect for Reconstructive leaders is important in itself, as
these leaders have large coattails and thus would be predicted to have a larger decline in
midterms. Another peculiarity is the fact that Disjunctive leaders, who did not exhibit any
coattails in the individual test of that theory (Table 8), appear to have the strongest effect here
from withdrawn coattails.
One possible explanation for the unusual effects that appear in this final model is the
effect of multicolinearity. For instance, GDP and PresVotePrev are very highly correlated for
Disjunctive and Preemptive leaders (-.94 and -.79 respectively, relative to -.18 for the whole
60
sample). The correlation of these two variables can bias the results and could explain the strange
coefficients present for all three theories in this last model. It also seems like the smaller sample
size created by segmenting the presidents by leadership type has some effect on the reliability of
the results. There seem to be far fewer irregularities across models for the group containing the
whole sample of presidents (Model 1), with differences only appearing in the magnitude of
coefficients and never in the direction of effects. However, for each group of presidents there is
less consistency between the models testing each theory individually and those testing all three
together. Further study could be undertaken to attempt to determine more exactly the sources of
bias in this final model, yet that analysis is beyond the scope of this project.
61
Chapter 6
Conclusions
Ultimately, the empirical models show that the well-established pattern of midterm
decline does not behave in the same way for all types of presidents. In fact, they show support
for the idea that where a president falls in political time may not only affect his ability to directly
exercise his office, as Skowronek hypothesizes, it may also determine how his party fares in
congressional elections. Given that a president's ability to advance his agenda is often contingent
on his party's congressional strength, his position in political time may also influence his political
success indirectly.
Reconstructive presidents, by the very nature of where they fall in the cycle of political
regimes, may have extraordinary repudiative powers and executive license, not only because of
their personal position as a Reconstructive president, but also because they can muster continued
congressional support for their agenda. If members of their party in congress are isolated from
some of the mechanisms of midterm decline, as these results show, these Reconstructive
presidents have an additional advantage over the presidents who follow.
Disjunctive presidents on the other hand, may be perceived as complete failures for a
reason beyond the fact that their position forces them to advance a platform that has become
antithetical to what the public perceives as good for the country. They may also be hit
particularly hard by the mechanisms that cause midterm decline or be specially insulated from
those mechanisms that traditionally help the president's party in on-year elections, such as the
effect of presidential coattails. These trends diminish the congressional strength of their party
and make it difficult to pass any legislative initiative, even those which might actually help solve
the country's problems.
62
The models testing individual theories show that Reconstructive presidents are somewhat
insulated from economic and systematic punishment, as I hypothesized in the earlier theory
section. It makes sense that as early leaders in a resilient regime, they are given a heavy license
for change without concrete expectations for results. Preemptive and Disjunctive presidents seem
to be hit hardest by both kinds of punishment, which makes sense given that they are always
followed by a president from the other party, a revival of the opposing regime. I found it
particularly interesting that Preemptive presidents experienced such severe systematic
punishment, evidence perhaps of serious dissatisfaction with their middle of the road approach to
politics.
In the coattail models, Reconstructive and Articulative presidents have the largest
coattails, as I predicted in Chapter 2. Disjunctive leaders do not have significant coattails, a
reflection of their lack of personal appeal and perception as an incompetent leader. It is very
difficult for a leader to bring like-minded candidates into office when he can muster little
political support for himself. The initial summary statistics regarding aggregate seat change,
presented in Chapter 3, indicated that the parties of these presidents tend not to gain seats in the
on-year elections, but also experience less of a drastic seat loss during midterms, in sharp
contrast to the pattern evident for other types of presidents.
While the results from the model testing each theory simultaneously are less clear cut or
conclusive, they do indicate that there are certainly differences in how each theory plays out for
the different leadership types. The relative importance of each effect is difficult to test given the
lack of variation for some variables for each segmented group of presidents. This can bias
coefficients and create the kind of puzzling effects seen in Table 11. This is one drawback that
comes with exploring how these theories play out for different types of presidential leaders. In
63
trying to pin down more exact mechanisms of midterm decline for different types of presidents, I
lose some of the variation necessary to produce consistent, unbiased statistical models.
However, I still believe the results presented in this paper are useful and illuminating.
What is clear from the results in Tables 3-11 is that the mechanisms of midterm decline do not
necessarily operate the same for presidents who fall in different segments of political time. For
some of the analyses, there is variation across groups of presidents in both the magnitude and
direction of effects. What this means for the decisions of future leaders is not yet clear, but there
is certainly merit to further exploration, given that presidents may have far less power than they
think to influence the outcomes of congressional elections.
As I noted in the Introduction, the magnitude of midterm seat loss is far from constant
across presidents, and a thorough understanding of the mechanisms that lead to declines of such
varied magnitudes could have enormous implications for legislative strategy. Such an
understanding may influence which policy initiatives the majority party prioritizes in the months
leading up to a midterm election, or cause the minority party to attempt to stall vulnerable
initiatives if they expect to take power in the near future.
The existing literature attempts to parse out some of the variation in midterm seat loss,
through both qualitative and quantitative research that has produced theories such as those of
Economic Punishment, Systematic Punishment, and Presidential Coattails (or Surge and
Decline), all of which are incorporated in this project. However, this thesis adds a final piece to
the puzzle, developed by Skowronek (1993) and studied in the context of Presidential Authority,
through his own work, and even in a case study of midterm elections (Price 2002). It explores
how presidential leadership, as defined by Skowronek, can exert influence on the outcomes of
congressional elections, both in the context of differences in aggregate seat change as well as in
64
the very mechanisms of midterm decline. Throughout the thesis, it becomes clear that while there
is more work to do in determining the exact effects of presidential leadership, there are certainly
important differences in the mechanisms and magnitudes of midterm decline across leadership
types, a trend that has significant implications for legislative strategy, and merits further
consideration and analysis.
65
Bibliography
Abramowitz, A. I. (1985). Economic Conditions, Presidential Popularity, and Voting Behavior in Midterm Congressional Elections. The Journal of Politics, 47(1), 31-43. Bureau of Economic Analysis (2011). National Income and Product Accounts: Current-dollar and "real"
GDP: US Department of Commerce. Born, R. (1990). Surge and Decline, Negative Voting, and the Midterm Loss Phenomenon: A Simultaneous Choice Analysis. American Journal of Political Science, 34(3), 615-645. Brady, D. W., Cogan, J. F., Gaines, B. J., & Rivers, D. (1996). The Perils of Presidential Support: How the Republicans Took the House in the 1994 Midterm Elections. Political Behavior, 18(4), 345-367. Campbell, J. E. (1991). The Presidential Surge and its Midterm Decline in Congressional Elections, 1868-1988. The Journal of Politics, 53(2), 477-487. Brady, David W., Hahrie Han, and Laurel Harbridge (2008). “House General Elections Data, 1876-2008.” Stanford University. King, A. G. a. G. (1990). Estimating Incumbency Advantage Without Bias. American Journal of Political Science, 34, 1132-1164. Levitt, S. D. (1994). An Empirical Test of Competing Explanations for the Midterm Gap in the US House. Economics and Politics, 6(1)(March 94), 25-37. Office of the Clerk of the House of Representatives. (n.d.). Party Divisions of the House of Representatives (1789 to Present). Retrieved November 2010, from House History: http://artandhistory.house.gov/house_history/partyDiv.asp Price, K. S. (2002). The Partisan Legacies of Preemptive Leadership: Assessing the Eisenhower Cohorts in the U.S. House. Political Research Quarterly, 55(3), 609-631. Peters, G. (1999-2011). Presidential Elections Data. (University of Santa Barbara) Retrieved February 2010, from The American Presidency Project: http://www.presidency.ucsb.edu/elections.php Peters, G. (1999-2011). Presidential Job Approval. (University of Santa Barbara) Retrieved November 2010, from The American Presidency Project: http://www.presidency.ucsb.edu/data/popularity.php Skowronek, S. (1993). The Politics Presidents Make: Leadership from John Adams to George Bush. Cambrige, Massachusetts: The Belknap Press of Harvard University Press. Skowronek, S. (2008). Presidential Leadership in Political Time: Reprise and Reappraisal. Lawrence,
66
Kansas: University Press of Kansas. Sutch, Richard (2006) “ Gross domestic product: 1790–2002 [Continuous annual series].” Table Ca9-19 in Historical Statistics of the United States, Earliest Times to the Present: Millennial Edition, edited by Susan B. Carter, Scott Sigmund Gartner, Michael R. Haines, Alan L. Olmstead, Richard Sutch, and Gavin Wright. New York: Cambridge University Press: http://dx.doi.org/10.1017/ISBN-9780511132971.Ca1-2610.1017/ISBN-9780511132971.Ca1-26