THE CONTEXT OF THE 2010 MIDTERM

In terms of partisanship, the electorate is nearly evenly divided, certainly more so than the current House division of 257 Democrats to 178 Republicans (59% to 41%) left by the 2008 election. Democrats gained a few points in macropartisanship in 2008, offsetting Republican gains in 2004, but the slight edge for the Democrats left after the 1984 realignment of party identification remains pretty much intact (Campbell Reference Campbellforthcoming; Norpoth Reference Norpoth1987). An average of three Gallup polls conducted between May and mid-June of 2010 show an electorate that is 46% Democratic and 43% Republican (Gallup 2010).

In terms of ideology, self-described conservatives continue to outnumber self-described liberals by a substantial margin (Campbell Reference Campbell, Nivola and Brady2007). In June 2010, 42% of respondents told Gallup that they were conservatives, while 20% claimed to be liberals, and 35% said they were moderates (Saad Reference Saad2010). Not surprisingly, 49% of Americans thought that the Democratic Party was too liberal, and 48% found them to be either about right or even too conservative (Jones Reference Jones2010a). The nearly even division in partisanship and the conservative tilt in ideology suggest that the current equilibrium in the electorate is far more Republican than the status quo in the House.

The results of the last two elections also boost Republican prospects. The midterm decline from the 2008 presidential surge should benefit the Republicans. A number of Democrats will be running without the help they received from President Obama's victory in 2008 (Campbell Reference Campbell1960; Campbell Reference Campbell1997). Democratic gains in 2006 (31 seats) and 2008 (24 seats) have left many House Democrats in the unenviable position of running in districts hospitable to Republicans. Democrats are defending 47 seats in districts that were carried by Bush in 2004 and McCain in 2008. In contrast, Republicans hold only six seats in districts carried by Kerry in 2004 and Obama in 2008. In the language of the exposure thesis, Democrats are overexposed going into the 2010 midterm (Oppenheimer, Stimson, and Waterman Reference Oppenheimer, Stimson and Waterman1986). The number of seats they currently hold far exceeds their base or average holdings in the last 20 years. In the 10 elections since 1990, Democrats won an average of 226 seats, 31 fewer than in 2008.

The political climate as we enter the fall campaign season also favors Republicans. Partisan parity, political polarization, the departure of an unpopular Republican president, and his replacement with a very liberal Democratic president and Congress constitute a powerful political mix that may lead to a Republican resurgence. Having been on the defensive in 2006 and 2008 and then relegated to the sidelines as President Obama and the Democratic-controlled Congress passed liberal policies over the last two years, conservatives are energized for 2010. Polls, primary turnouts, the emergence of the Tea Party movement, and Republican victories in 2009 (including Scott Brown's Senate win in Massachusetts) are unmistakable stirrings of a revitalized right. In June, Gallup reported that 53% of Republicans were more enthusiastic than usual about voting. Only 35% of Democrats were similarly enthusiastic (Jones Reference Jones2010b). And although President Obama is not unpopular at this point (his approval ratings stand in the mid 40s), he also does not have the strong approval ratings that would provide much help to his party in staving off significant midterm losses (Tufte Reference Tufte1978).

SEATS IN TROUBLE

While long-term factors are quite favorable for Republicans in 2010, they indicate only the election's potential and therefore provide only limited guidance in anticipating the results of the election. Greater forecast accuracy requires predictors that take into account the more localized, short-term, and prospective factors that are critical to congressional outcomes as well as the effects of national, long-term, and retrospective considerations. My strategy in devising an accurate and plausible forecasting equation was to build the model around a core predictor that would offer an accurate reading of where the election stood at the time of the forecast, and to then augment the model with contextual variables that would provide guidance on how things were likely to change between the time of the forecast and the election. This is the same strategy that was used in devising the trial-heat model of the presidential vote (Campbell and Wink Reference Campbell and Wink1990).

The model developed and used here for House elections is the “seats in trouble model.” I also think of it as the “exposure-thesis-on-steroids model.”Footnote ¹ The exposure thesis suggests that an overexposed party holds more seats than usual, and that this might leave it with more seats in trouble or in danger of being lost. The exposure-thesis-on-steroids or seats in trouble model is based on estimates of the extent to which one party has more seats actually in trouble.

The core variable of this seat change forecasting model is based on the intensive political handicapping of congressional elections conducted for the last 26 years by Charlie Cook and his colleagues at the Cook Political Report.Footnote ² Beginning in 1984 and in each election since, Cook has made a comprehensive district-by-district assessment of the electoral prospects of each political party at various points before and during the election year. Each seat is scored as “solid” Democratic, “likely” Democratic, “leaning” Democratic, “toss-up,” “leaning” Republican, “likely” Republican, or “solid” Republican. According to Cook, seats identified as likely for a party “are not considered competitive at this point, but have the potential to become politically engaged” (Cook Reference Cook2010). Leaning districts are considered competitive, “but one party has an advantage.” In toss-up districts, “either party has a good chance of winning.” For forecasting purposes in both on-year and midterm elections, I used Cook's latest assessment, made between July and the first day of September in the year of the election being forecast. In most years, the assessment used was made in mid- to late August. Since assessments were not conducted during these months in 1986 and 1990, those years could not be included in constructing the forecasting equation. This leaves 11 usable elections.

Cook's record of accuracy in handicapping individual district elections is impressive. Although the powers of incumbency and district partisanship play a role in predicting outcomes, there is clearly a great deal of value-added in these late summer assessments. In the 11 elections examined, districts rated as solidly in a party's column turned out to be nearly sure bets (99.8% for Democrats and 99.7% for Republicans). The parties were nearly as certain to hold their likely seats (94.0% for Democrats and 95.1% for Republicans) and were very successful in holding their leaning seats (88.2% for Democrats and 85.0% for Republicans)—although leaning districts were not quite so safe for Republicans in the last two election cycles. In 2006, Republicans held only 38% of the seats identified as leaning their way in late August. In 2008, they won only 55% of these leaners.Footnote ³

The aggregate outcomes in toss-up districts were about as anticipated and generally quite different from the outcomes in leaning districts. When the previous party holding the seat could be determined (setting aside a number of seats affected by reapportionment and redistricting), Democrats held about 48% of their toss-up districts and Republicans about 55% of theirs.

From Cook's district data, I constructed an aggregate forecasting measure: seats in trouble. The measure takes the number of seats that a party won in the prior election and deducts the number of seats that Cook viewed as solid, likely, or leaning toward that party. The remainder is the number of seats that are in trouble. This figure is slightly different from the number of toss-up districts, since it also counts districts currently held by a party but anticipated by Cook to be leaning, likely, or solidly in the opposing party's column. This algorithm also addresses the problem of how to count toss-up districts in redistricting years. The predictor variable is the difference between each party's number of seats in trouble. The logic of the indicator is that the more troubled seats a party holds relative to the opposing party, the more seats it should lose in the election.

An alternative measure that included leaning seats in the index as potentially troubled seats was also examined but did not strengthen the equation. This would seem to reflect both the generally high success rates that the parties have had in holding their leaning seats as well as the variance in that rate (the 2006 and 2008 Republican losses). Because of the generally high success rates for parties in their leaning districts and the occasional variance in this rate, the indicator counts as troubled only those districts that are toss-ups or worse.

Table 1 presents the number of troubled seats for each party and the difference between them, as well as the Democratic seat change in these 11 elections. Note that it is possible to have a negative number of seats in trouble for a party if some seats currently held by the opposing party are seen as likely or sure wins for the other party in the next election, or if a party has gained seats in special elections since the previous national election. This was the case for the Democrats in 2006.Footnote ⁴

Table 1

Seats in Trouble for the Political Parties

Note: Half of the seat changes are the result of counting seats held by independents as half for each major party.

The pattern of seats in trouble corresponds quite closely with the extent and direction of seat change. In most of this period, seat losses were small, as were the differences in the number of seats each party held that were in trouble. The three elections in which one of the parties registered significant seat gains were those in which the other party had many more seats in peril. In the Republican realignment year of 1994, the last act of the staggered realignment (Campbell Reference Campbell2006; Paulson Reference Paulson2006), the Democrats could count 47 seats in trouble—and they ended up losing even more. In 2006, with an unpopular president, late August estimates showed 19 Republican seats in trouble. Late-breaking congressional scandals increased this number and eventually led to Republican seat losses that were significant enough to cost them control of the House. In 2008, with 27 seats in trouble (compared to none for the Democrats), the Republicans lost another two dozen seats.

PRIOR SEATS HELD AND PRESIDENTIAL APPROVAL

With the principal “seats in trouble” predictor in place, the second component of the forecasting model was to determine whether any contextual variables improved the accuracy of the forecasting equation. I examined several variables, including the generic vote, but found only two that seemed plausible and added predictive value. The first was the number of seats a party won in the previous election. This takes note of the fact that a party cannot lose seats that it does not have and cannot gain seats that it already holds (Campbell Reference Campbell1997, 131). It also acknowledges the political fact that it becomes increasingly difficult to gain seats as a party's seat holdings increase. A party registers gains first where it is easiest for it to do so, and it becomes progressively more difficult for the party to pick up additional seats in areas that are more inclined to support the opposition party.

The second contextual variable was presidential approval. As the leader of the party, reactions to the president affect the party's fortunes both in on-years (Campbell Reference Campbell1997) and in midterms (Tufte Reference Tufte1978). While presidential approval ratings have long been used to reflect the referendum or retrospective nature of elections, especially midterm elections, not much attention has been given to determining the neutral value of approval—that is, the value of approval necessary for the president to neither help nor hurt the party's congressional fortunes. After examining the empirical evidence, it is clear that presidential approval ratings mean one thing in presidential elections in which two parties are contending for the office and quite another in midterms in which there is no presidential choice to be made. There is little evidence to support the commonly assumed (and rarely justified) 50% mark as being the neutral point in either case. In presidential elections, presidents with 50% approval always win. In midterms, the parties of presidents with 50% approval always lose seats. The on-year politically neutral point seems to be closer to 45%. Some voters who disapprove of the president still vote for him, believing him to be better than the alternative.

The neutral point is quite a bit higher in midterms, and recent midterm successes by both parties provide us with some bearings in its determination. Democratic in-party gains in the 1998 midterm and Republican in-party gains in the 2002 midterm suggest that the neutral point in midterms is around 65%. Between 1870 and 1994, the president's party gained seats in only one of the 32 midterm elections. That year was 1934, the first midterm election of the New Deal realignment. Then, in 1998, with President Clinton's approval ratings in Gallup sitting at 66% at the end of October, Democrats defied the odds and gained five seats. Four years later, after September 11 and with President Bush's approval ratings in Gallup at 63% just prior to the 2002 election, the Republicans also defied the midterm loss rule and gained eight seats. Without a partisan presidential choice on the table, the positivity inclination of most citizens seems to guide their approval ratings toward the high side in midterm elections. A 65% rating in the heat of a presidential year is astounding and a precursor to a landslide. A 65% rating in a midterm sounds great but, politically, is approximately neutral and only a precursor to holding the status quo.

In this model, I calculated a presidential approval index by subtracting the neutral point (45% on-year or 65% midterm) from the Gallup measure of presidential approval near the end of August. The index was oriented by party by taking its negative value when a Republican was president. The index ranges from −21 in 1994, when President Clinton's approval rating stood at 44%, to 27 in 2006, when President Bush's approval rating was 38%. As one might expect, the approval index is highly correlated (r = −.70) with the seats in trouble variable.

THE FORECAST EQUATIONS

Table 2 presents the forecast equations. The predicted election outcome is seat change for the Democratic Party. Outcomes are calculated from the prior election rather than after special election results. The data for the number of seats held by each party are from the U.S. House of Representatives, Office of the Clerk (2010). For comparability, seats won by independent or third-party candidates are divided equally between the two major parties.

Table 2

The Seats in Trouble Forecasting Equations of Seat Change for the Democrats in the U.S. House

Note.

** p < .01, one-tailed.

* p < .05, one-tailed.

N = 11. Standard errors are in parentheses. The equations are estimated using data for 1984, 1988, and the nine national elections from 1992 to 2008. Median absolute errors are calculated from out-of-sample errors.

Equation 1 presents the simple bivariate relationship between Democratic seat change and the relative number of Democratic seats in trouble. This simple association is quite strong. A party should expect to lose slightly more than one seat for every net seat that is in trouble.

Equation 2 adds the initial number of seats held by the Democrats. A party loses about one seat for every one that is in trouble and one seat for every additional five that it holds at the outset. While the equation has a strong fit, a substantial envelope of uncertainty around any forecast remains. The median absolute error, based on out-of-sample estimates, is about 5.5 seats. There are numerous local factors that assessments of seats in trouble may have missed or that may have developed after the last summer forecasts.

In equation 3, the presidential approval index is added in lieu of the lagged seats variable. The overall fit of equation 3 is about the same as that of equation 2. This is reassuring in that two different contextual variables added to the seats in trouble consideration produce equations of approximately equal strength. Equation 3 indicates that a party should expect to lose just under one seat for every seat in trouble, and that a party should expect to lose about six seats for every 10 points that it falls short of the neutral level of presidential approval. Because of the small number of cases available for the estimation, an equation with both the lagged seats and the approval index along with the seats in trouble variable produced coefficients that were not significant at conventional levels for the lagged seats and approval index variables.

THE 2010 FORECAST

What does the seats-in-trouble model predict for 2010? First, as of the Cook Political Report's assessment in late August 2010, Democrats have 42 seats in trouble and Republicans stand at negative two. The seats in trouble variable is thus 44, about as large as it was in the 1994 midterm and consistent with the Republican disposition of the election's fundamentals or context. The lagged number of Democratic seats held is 257. President Obama's approval rating in late August 2010 stood at 44%. With a neutral point at the midterm of 65%, the presidential approval index stands at negative 21.

Based on the seats in trouble indicator and the two contextual variables of equations 2 and 3, the forecast is that Democrats will lose about 51 or 52 seats, leaving them with a total of 205 or 206 seats. The odds appear to be quite favorable for the Republicans regaining the House majority that they lost in 2006.