It had been an inevitability rivaling death and taxes. The president's party
would lose seats in the U.S. House of Representatives in almost every
midterm election. The president's party had lost House seats in 32 of the 33
midterms from 1866 to 1994. The sole exception was a gain of nine seats for
Democrats in the 1934 midterm election during FDR's first term and in the
throes of the New Deal realignment. Then came the 1998 and the 2002
midterms. With a strong economy and the pending impeachment vote against
President Clinton, Democrats gained four seats in 1998. This past November,
in the aftermath of the 9/11 terrorism and with a president with strong
approval ratings, Republicans also broke from the traditional pattern of
in-party midterm losses and gained five seats in the House and two in the
Senate, regaining their control of the Senate that they had lost when
Vermont Senator Jim Jeffords renounced his Republican affiliation in
2001.1
The five seat gain was the gain
over the number of seats that Republicans had going into the 2002
midterm. The change from the 2000 to 2002 elections represented an
eight seat gain. The difference is the result of special elections
held to fill vacancies between the 2000 and 2002
elections.
The results of the 2002 midterm raise a number of questions. To what extent
was the midterm (and the 1998 midterm) an aberration? Why were Republicans
as the in-party able to gain seats? What explains 2002 and what might the
2002 (and the 1998) results indicate about various theories of
inter-election seat change?
Explaining 2002
While midterm seat gains for any inparty are certainly atypical, from
several other perspectives the results of 2002 are not unusual at all.
The election ran true to form in five respects. First, the small
magnitude of seat changes in both 1998 and 2002 reflect the overall
decline in district competition and are typical of the constrained
electoral change that has marked elections generally since the late
1960s. Second, the 2002 results reflect the nearly equal balance of the
parties nationally and the continued partisan sorting out of districts
in the later stages of the staggered partisan realignment that had
deepened to congressional elections in the early 1990s. Third, some
portion of the Republican 2002 gain reflected the reapportionment and
redistricting that generally favored Republicans. Fourth, the effects of
surge and decline anticipated small in-party gains. The surge in 2000
was a small negative surge for the party winning the White House. Gore,
not Bush, received a narrow popular vote plurality. Following a small
negative on-year surge should come a small negative midterm decline (a
small in-party midterm seat gain). Finally, with the president's
approval ratings in the mid-60% range, signaling an electorate largely
supportive of President Bush's leadership to this point, the midterm as
a referendum on the in-party should have provided a pro-Republican
tilt.
The Politics of Small Change
The increased advantage of incumbency and the resulting decline in
congressional competition has been thoroughly documented now for several
decades (Erikson 1971; Mayhew 1974; Ferejohn 1977).
However, the consequence of these developments for aggregate electoral
change has not been so widely appreciated. Seat changes are not nearly
as large as they used to be. The absolute magnitudes of seat changes for
or against a party have been substantially smaller in recent decades
(Campbell 1996, 158; Campbell and Jurek 2003). With fewer competitive districts
in play, national tides swing fewer districts from one party to the
other and the districts in which local forces (candidates, campaign
spending) are decisive may make more of a difference to the seat swing
bottomline of the election.
Figure 1 plots the absolute amount of net
partisan seat change for elections in the twentieth century, with median
absolute seat changes indicated in boxes for each quarter of the
century. In the first half of the century, the median election shifted
31 or 32 seats from one party to the other. In the century's third
quarter, the median absolute seat change dropped to 20 seats. Since
1976, the typical election shifted a mere seven or eight seats from one
party to the other, about a quarter of what the typical change was in
the first half of the century. Whereas there were nine elections (35%)
with seat changes greater than 50 seats in the first half of the
century, only one election since 1950 has produced a 50-plus seat
change. That was the Republican breakthrough election in 1994. In fact,
in the last nine national elections (from 1986 to 2002), 1994 is the
only election in which a party has gained or lost more than 10
seats. With the decline in competition and the compression of
electoral change, whichever party gained or lost in 2002 was likely to
gain or lose only a few seats. The small number of Republican seat gains
in 2002 is consistent with this general narrowing of seat changes in
recent decades.2
Republican seat gains may
also have been constrained by the electoral system advantage
that Democrats have in winning low turnout districts (Campbell 1996). Although the official
vote counts are not available at this writing, it appears that
Republicans received between 52 and 53% of the national House
vote and 52.8% of House seats. Assuming no electoral system
bias, this would put the swing-ratio at about one, well below
its historical value (and suggesting that the assumption of no
bias is incorrect). By comparison, when Democrats had a similar
sized vote majority in the 1992 election, they won 59.4% of the
seats (258 seats for Democrats in 1992 compared to 229 for
Republicans in 2002).
Sorting Out Party Districts
The slim Republican congressional majorities resulting from the 2002
election also are consistent with the realignment that has produced
polarized parties of nearly equal strength nationally (Ladd 1985, 1995; Bullock 1988; Petrocik
1987; Frymer 1996; Campbell 1996, 1997). The
staggered party realignment that developed from a switch in the parties'
relative positions on racial issues in the early 1960s (Carmines and Stimson 1989) generated a change in
presidential voting in the 1960s and 1970s, in closing the party
identification gap in the 1980s (Bartels 2000;
Campbell 2000; Miller
1991), and in congressional elections from the 1990s until
the present (Abramowitz 1995; Campbell 1997). This realignment produced a
sorting out of districts, a trend toward more partisan-consistent
election results, and fewer split-result districts (Jacobson 2000). This trend continued in 2002. In the 1984
election, 191 districts (44%) were carried by one party's presidential
candidate while the district elected a representative from the opposite
party. The number of split-result districts, using the 1984 presidential
vote, was even a bit greater (199) in 1986. In 1988 the number had
declined to 148 (34%) and declined to between 100 and 110 (23% to 25%)
split-result districts from 1992 to 1998. In 2000, the number of
split-result districts dropped to 85 (20%) and in 2002 it dropped again
to just 62 (14%), the fewest split-result districts since at least the
1940s (Ornstein, Mann, and Malbin 1996).
Table 1 presents the joint distribution of
presidential and congressional voting in the 2000 and 2002 elections. In
the 2000 election, congressional Democrats won 167 (81%) of the 207
districts that Al Gore carried and congressional Republicans won 181
(80%) of the 226 districts that George W. Bush carried. Democrats won 45
districts with Republican presidential pluralities and Republicans won
40 districts with Democratic presidential pluralities. These numbers of
split-result districts were low in comparison to elections since the
1940s, but dropped even lower in 2002. If presidential voting in a
competitive election can be taken as a common metric of partisan turf,
both parties gained more of their natural ground in the 2002 midterm.
The number of districts that Bush carried (or would have carried in the
newly drawn districts) that were won by congressional Democrats dropped
from 45 in 2000 to 35 in 2002.3
I would
like to thank Bill Steiner and the rest of the Republican
National Committee's staff for providing me with the
presidential vote percentages for the newly drawn congressional
districts.
Similarly, the number of Gore districts won
by House Republicans dropped from 40 in 2000 to 27 in 2002. With a small
and declining number of exceptions, the outcomes of congressional and
presidential elections in districts are now aligned.
The prevalence of split-result districts in earlier decades largely
reflected southern politics and the slow development of a viable
Republican Party in the region long after southerners had moved toward
voting for Republicans in presidential elections. It was not unusual for
Republican presidential candidates to carry southern districts by wide
margins and have no Republican congressional candidate at all on the
ticket (Campbell 1997, 202–5). Republican
congressional gains in the South, and the resulting decline in
split-result districts there, caused some to characterize the
realignment as largely a southern phenomenon. Republican gains in 2002,
however, were not centered in the South, but in the Midwest. Table 2 presents where the parties registered
gains and losses by region.
Republicans gained five seats in Midwestern states while Democrats lost
those and an additional seven to reapportionment. The parties fought to
a draw elsewhere. While Republican gains in the Midwest may reflect a
variety of factors, from greater support for President Bush in the “war
on terror” to reapportionment effects to the continued sorting out of
districts by partisanship, it certainly indicates that Republicans are
registering gains beyond the South and that American political battles
may be evolving toward a heartland Republican base pitted against a
bicoastal Democratic base.
Redistricting
The 2002 midterm was also the first after reapportionment and this may
also have affected the results. Whether through redistricting within
states or, more likely, reapportionment across states (as districts were
lost from northeastern Democratic leaning states and gained by southern
Republican leaning states), Republicans gained from the decennial
redrawing of districts. The distribution of the 2000 presidential vote
in the old and new districts reveals that redrawing district lines
produced more Republican leaning districts. Under the old lines, Bush
carried 227 districts to 208 for Gore. If the 2000 election had been
conducted with the 2002 redrawn lines, Bush would have carried an
additional 10 districts. To the extent that the presidential vote
provides for cross-district comparisons, Democrats had a more hospitable
set of lines in 2000 (most probably reflecting the malapportionment that
had developed over the 1990s) and Republicans had relatively more
favorable districts in place for 2002.
A Slight Negative Surge, A Slight Negative Decline
The driving force behind midterm losses historically has been the
presidential surge in the prior on-year election (A.
Campbell 1960; J. Campbell 1991,
1997). The party winning the presidency in
the prior election rides to victory on a “high stimulus” surge of
favorable short-term forces (e.g., issues and candidates) and this
carries a number of candidates below the winning president into office.
The stronger the short-term forces, the bigger the presidential victory
and the more representatives of the president's party are brought into
office. Partisans of the winning president's party tend to be more
enthusiastic and more likely to turn out. Some partisans of the losing
presidential candidate are disaffected and stay home (Campbell 1997, 172–81). The independents swing to the more
popular, winning candidate. As we know from NES, about 70 to 80% of
presidential voters report voting for the same party's congressional
candidate. The arithmetic is inescapable: a party's typical House
candidate gains four to six votes on his or her opponent from every 10
votes brought to the polls to vote for the party's presidential
candidate. For some, this is the margin of difference. However, the
benefits of the presidential surge are short-lived. Two years later,
without the assistance of strong short-term forces systematically
favoring the candidates of the in-party, some number of representatives
of the president's party go down to defeat. As with the effects of the
on-year surge, the effects of the midterm decline are proportional to
the prior short-term forces. Small on-year surges (e.g., 1960) lead to
small midterm declines (e.g., 1962) and big on-year surges (e.g., 1964,
1972) auger big midterm declines (e.g., 1966, 1974).
The 2002 midterm results are much as might be expected by the theory of
surge and decline.4
Republican Senate seat
gains in 2002 also fit the surge and decline theory. With
six-year Senate terms, the relevant presidential surge for the
2002 midterm decline was in the presidential election of 1996
(Campbell 1997, 193). The 1996
presidential election in which Clinton defeated Dole should have
helped a number of Democratic Senate candidates. Running without
that help in 2002, some should have suffered defeat, as was the
case.
The narrow popular presidential vote margin
favoring Al Gore in 2000 suggests that short-term forces favored
Democrats very slightly that year. The slight Democratic advantage in
short-term forces in the 2000 on-year set the stage for slight
Republican gains in the 2002 midterm. This is exactly what happened.
5 The compression of seat changes, the
re-alignment, and the reapportionment after 2000, complicated
making a precise expectation of surge and decline effects. Prior
to the election, and without taking the pro-Republican
reapportionment effects into account, I expected very slight
Democratic gains.
Presidential Approval
As Edward Tufte (1975, 1978) noted many years ago, midterm elections are also
referenda on the administration's performance. When the public is
supportive of the job the president is doing, they are inclined to keep
more representatives of the president's party in office. On the other
hand, if they disapprove for any reason of the way that the president is
handling his job, that disapproval can be registered by returning fewer
representatives of the president's party to office.
The presidential approval rating used by Gallup over the years has proven
to be an effective yardstick of the referendum component of the midterm
verdict. These midterm approval ratings have varied quite a bit. Truman
was very unpopular in 1946 (32% approval) and in 1950 (43%) as was
Reagan in 1982 (42%) and Clinton in 1994 (43%). Other presidents have
had very positive midterm assessments. Following the Cuban missile
crisis, Kennedy had a 67% approval rating. The median presidential
approval rating since 1946 has been 56%.
President Bush's approval rating going into the 2002 midterm stood at
63%, above average and about where Reagan stood in 1986 and only a
couple of points lower than Clinton's rating in 1998. While approval
numbers in this range normally have not led to in-party midterm gains in
the past (though they did in 1998), in conjunction with the lack of a
prior on-year surge and with the general compression of change of any
sort, the generally favorable assessment of the administration provided
Republicans with a context that made it possible for them to gain a few
seats.
Although it is beyond the scope of this paper to determine how much of a
difference the partisan sorting out of districts, redistricting, the
negative surge, and presidential approval made to the election results
of 2002, each of these factors is consistent with the small seat gains
for Republicans. This is not the case for other considerations.
Non-Explanations of 2002
While several factors and theories help to explain the in-party gains in
the 2002 midterm, other potential electoral forces would appear to be
inconsistent. Though the 2002 results are those of only a single
election and though no theory can reasonably be expected to explain all
variation in seat changes, the 2002 results are a piece of evidence that
fits some theories much better than others.
With the long record of divided governments since the 1960s (probably
mostly due to the staggered realignment and the close division of the
parties nationally), several theories proposed that a decisive number of
voters intentionally choose divided government either out of a
preference to have different parties control the different institutions
(Jacobson 1990a) or out of a desire to
achieve more moderate policies by having the parties counterbalance each
other (Alesina and Rosenthal 1989, 1995; Fiorina 1996).
While the large percentage of straight-ticket voters and the strong
positive correlations between presidential and congressional voting had
previously raised doubts about the plausibility of substantial
policy-balancing and intentionally divided government, the 2002 results
would seem to substantiate further doubts. Knowing that the Republicans
controlled the presidency and with the government already divided, the
voters in 2002 elected more rather than fewer partisans of the
president's party thus creating a unified government rather than a
divided government.
The results of the 2002 election also appear inconsistent with the idea
that there is some generalized or acrossthe-board midterm penalty for
the president's party (Erikson 1988) or that
midterm elections are strongly dependent on the state of the economy.
The extent of independent economic effects on congressional elections
has been a disputed issue (Erikson 1990; Jacobson 1990). The 2002 experience would seem
to support the view that the economy is at most a minor consideration in
congressional voting. While there were some signs that the economy going
into the 2002 midterm was not as weak as many believed, it was clearly a
liability for the administration and hardly the basis for midterm seat
gains for the in-party.
Finally, the implications of the 2002 election are unclear for one theory
of electoral change. The exposure thesis suggests that midterm elections
involve a reversion to an equilibrium or an average seat holding for a
party (Oppenheimer, Stimson, and Waterman
1986; Waterman, Oppenheimer, and Stimson
1991). A party having more seats than its equilibrium—a
normal seat analogue to the concept of the normal vote (Converse 1966)—is overexposed and likely to lose
seats in the next election. A party holding fewer than its normal number
is underexposed and likely to gain seats. While the equilibrium number
of seats for Democrats and Republicans in the realigned party system may
still be unknown, Republican gains in 2002 could only be explained by
the exposure thesis if Republicans held fewer seats than their expected
equilibrium. That is, the exposure thesis suggests the Republican
equilibrium is greater than the 221 Republicans elected in 2000 and that
a Republican House majority is now normal politics. Only future
elections will allow us to determine whether this is plausible.
