Explaining Midterm Loss:
The Tandem Effects of Withdrawn Coattails and Balancing
Robert S. Erikson
Department of Political Science
Columbia University
Prepared for the 2010 Meeting of the American Political Science Association,
Washington DC, Sept. 2-5
Abstract
This paper tests coattail and balancing theories of midterm loss in congressional elections. Neither theory can by itself account for the regularity of midterm loss. But together they can. When, following a close presidential election, there are no coattails to withdraw, ideological balancing at midterm generates a loss for the presidential party. When voters can anticipate a presidential landslide allowing them to balance their congressional vote in the presidential year, the winner’s coattails are withdrawn at midterm. Either way, the presidential party loses seats. Testing the variables of each theory while controlling for those of the other, this paper finds strong statistical evidence that both processes are at work to insure midterm loss.
One nearly universal “law” of American politics is that in midterm elections the presidential party almost always loses seats in the House of Representatives. One spectacular run occurred from 1938 through 1994. For fifteen consecutive midterm elections, the “out” party gained seats at midterm. And even when the string was broken in 1998, Clinton’s Democrats lost strength in terms of the national vote. The 2002 election presented one clear exception to the midterm rule as the Republicans gained both seats and votes in the wake of 9/11. The 2006 election presented a return to normal, as George W. Bush’s Republicans lost 30 House seats, based on a 5.5 percent decline in the popular vote.
What accounts for the phenomenon of midterm loss? Familiar explanations include included withdrawn presidential coattails (a.k.a. surge-and-decline), ideological balancing; and a negative referendum on the president. Here, I show how none of these explanations is sufficient in itself, yet they work together to account for a decline in support for the president’s party as a near-certain outcome at midterm.
The challenge is not simply to explain why the presidential party loses seats in most elections but why the phenomenon occurs with near-perfect regularity. One’s first temptation might be to invoke “referendum theory,” or the idea that midterm loss signifies a protest against unpopular presidents (Tufte, 1975). Certainly low presidential approval can lead to a harsh midterm loss. But for referendum theory to account for the regularity of midterm loss requires that presidents must almost always be embattled at the midpoint of their election cycle, and this is decidedly untrue. For the 16 post-WWII midterm years (1946-2006), one can consult the extensive Gallup Poll data bank on presidential approval. In October of 16 midterm years 1946-2006, the average approval of the president is 53 percent, with the president exceeding the 50 percent threshold in 10 of 16 instances. While unusually high presidential popularity can explain the 1998 and 2002 exceptions of midterm gain, a pattern of persistent presidential unpopularity cannot account for the midterm loss in the remaining cases. Even excluding 1998 and 2002, presidents averaged 51 percent approval in the October before the midterm election. To understand the regularity of midterm loss, we must look elsewhere. However, as shown below, referendum theory can account for the severity (including the rare revoking) of the rule of midterm loss.
This brings the discussion to two other prominent theories, left to compete for the explanation of the regularity of midterm loss: “coattail theory” and “balance theory.” Coattail theory accounts for midterm loss in terms of a surge for the winning presidential party in the presidential election year. Balance theory accounts for midterm loss in terms of voters surging to the out party at midterm.
The coattail argument (e.g., J. Campbell, 1991), is that winning presidential candidates sweep an exceptional number of ticket-mates into office on their presidential “coattails,” only to see them swept out again at midterm when the coattails are withdrawn. In the language of “surge and decline” (A. Campbell, 1966), support for the presidential winner’s congressional ticket surges due to “strong short-term partisan forces” which are aided by the participation of persuadable non-“core” peripheral nonpartisan voters. At midterm these peripheral voters stay home, yielding a more party-line normal vote with minimal “short-term partisan forces” at the national level. In other words, the winning president’s ticket-mates are boosted by the same short-term forces that generate the presidential victory while the dissipation of these forces by midterm account for the midterm decline.
How “coattails” work is an intriguing but challenging question. In part it could be that voters lazily vote a straight-ticket with the presidential choice influencing downballot choices. It can also be that presidential and congressional choices respond to identical forces without either actually influencing the other (Ferejohn and Calvert, 1984). For purposes here, this nuance is not important. The interest for this paper is on the simple question of the degree to which the partisan vote for the House of Representatives rises and falls with the presidential vote for whatever reason.
The balancing argument accounts for midterm loss as a surge toward the out-party at midterm. Alesina and Rosenthal’s version of balance theory (see also Fiorina 1995) holds that midterm voters shift toward the non-presidential party in order balance off the ideological tendencies of the president. The theory starts with the fact that the president is more conservative (if a Republican) or more liberal (if a Democrat) than most voters. Thus, ideologically attuned voters will tend to support the opposition party to achieve greater ideological balance between president and Congress. In this way, the midterm vote provides a partial policy corrective to the presidential year verdict. Although balance theory is generally framed in terms of liberal-conservatism, it is possible to broaden the motivational possibilities to incorporate policy motivations generally without the baggage of ideological language. For instance, some voters might try to balance the influence of the two parties from the simple motivation of preventing any one group of rascals to have too much influence over policy. For purposes here, this nuance is not important. Interest is in the degree to which midterm congressional voters are attracted—for whatever reason—to the party that does not hold the presidency.
It is tempting to depict the distinction between the coattail and balancing theories as a clash between competing views of the American voter. Coattail voting is in alignment with the traditional (“Michigan” school) model whereby electoral change is largely the response of inattentive swing voters (e.g., Converse, 1962, Zaller, 1992, 2004). The idea of unenlightened coattail voters temporarily surging to the winning president’s party fits this model. Balance theory is in alignment with models of rational actors who vote based on ideological proximity and think strategically (e.g., Downs, 1957). In extreme one can model midterm balancing voters as solving a complex problem of N-person game theory (e.g., Mebane, 2000). Of course the two types of voter models could apply to different strata of the electorate. And it is also possible that some voters share attributes of the voter as depicted in the separate models. In short, it is possible for the two models to coexist and work together to explain midterm loss.
Evidence can be marshaled for both the coattail theory and balance theory. Consistent with coattail theory, parties win a greater share of the congressional vote in years when they win the presidency rather than lose. (As a postwar average, winning rather than losing the presidency makes a 2.7 point differential in the presidential year national vote for the House of Representatives, as a bonus for presidential victory.) Consistent with balance theory, the parties win more votes at midterm when they do not hold the presidency than when they do. (The differential here is an even large 4.1 point differential, as a reward for not holding the presidency.) Still, these are mere tendencies and not near-universal regularities. By itself, neither theory can fully account for the near-universality of midterm loss. Moreover, at the theoretical level, each has a logical flaw that makes it implausible as a universal explanation.
As a universal explanation for midterm loss, coattail theory has two problems. First, coattails should arise only when the presidential winner achieves a decisive victory. When presidential elections are close, they generate no presidential coattails to withdraw at midterm. Second, the coattail explanation requires that short-term forces return to normal at midterm. Only with a dampening of short-term forces at midterm (e.g., to reflect the normal vote and nothing else) would a coattail-driven surge in the presidential year guarantee a midterm loss. But the facts defy surge-and-decline theory. Historically the variance of the two-party vote at midterm is almost twice as great at midterm than in presidential years (13.7 vs. 7.0 over postwar elections).
Balancing theory also has a problem as a universal explanation for midterm loss. This problem is the possibility of anticipatory balancing. Many presidential elections are not close, with polls informing voters of the likely winner. In theory, this knowledge would allow voters to balance in the presidential year without waiting for midterm, as voters could balance in advance by casting their congressional votes for the party about to lose the presidency. This sort of anticipatory balancing in presidential years would leave no need for further corrective action at midterm that would generate midterm loss.
This leads to the central argument of this paper. Although neither coattails or balancing can account for midterm loss by themselves, they do so by working together. One or the other will apply, depending on the nature of the presidential race. Landslide presidential elections provide coattails to be withdrawn at midterm. Close presidential elections prevent advance balancing, allowing balancing behavior to newly arise at midterm. Either way, the presidential party loses votes from the presidential to the midterm year.
Together, coattails and balancing make a powerful engine that drives midterm loss. Consider the prospects for midterm loss when the presidential outcome is a landslide, complete with coattails, and anticipated in advance by voters who balance the presidential winner by supporting the opposition for Congress. Although balancing in advance negates the need for new balancing at midterm, the midterm withdrawal of coattails still causes a decline in the presidential party’s seats. Consider the opposite case of a close presidential election where the outcome is in doubt (no balancing) and no coattails. Then, despite the lack of coattails, midterm loss occurs as a result of midterm-year balancing.
Note that the argument is more than an appeal to the axiom that two theories should predict better than one. Rather, it is the two theories complement each other so that the circumstances where one theory falters are those where the other theory prevails. Whether the president wins big or small, one expects a subsequent pattern of midterm loss. With a close presidential election, there are no coattails to withdraw; meanwhile the electorate can balance at midterm. With a landslide presidential election, voters can balance in advance; meanwhile coattails lift the congressional vote for the winning party only to fall at midterm.
The greatest midterm loss would occur with a surprise presidential landslide (no advance balancing, withdrawn coattails at midterm). The one combination of circumstances where one would not expect midterm loss would be when the widely anticipated presidential victor wins by a small margin. Then there would be full advance balancing, no coattails to withdraw, and thus little or no midterm loss.
If the two theories complement each other in the manner described, why has this not been evident from empirical studies? The problem is that the evidence for each theory collides with that for the other. When landslide presidential victories are accompanied by lackluster success downballot, the fault might not be a lack of coattails but rather that the coattails are obscured by balancing voters offsetting coattail voters. Similarly, the presence of balancing behavior in presidential years is obscured by the offsetting vote by coattail voters.
This paper tests for the joint effects of coattails and balancing, with the key being the measurement of both the actual presidential vote (representing coattail strength) and expectations of who will win (representing the circumstance for balancing). The paper finds that large presidential victory margins and a strong expectation of victory each influence the congressional vote, with the expected opposite signs. Among coattail voters, landslides generate coattails. Among balancers, the anticipation of presidential victory generates voting for the opposite congressional party.
The crucial test is for the effect of anticipatory balancing. When anticipatory balancing in the presidential year cab be estimated and controlled for, the evidence for coattails is strengthened. Also, evidence for anticipatory balancing in the presidential year helps to bolster ideological balancing as the source of the presidential party’s electoral penalty at midterm. If voters are capable of punishing the winning presidential party in anticipation of their holding office, it is easier to accept that they do so in the midterm year when the presidential party is known with certainty.
The Model
The central task is to statistically model the congressional vote in the 32 post-WWII congressional election 1946-2008, for the purpose of further accounting for the phenomenon of midterm loss.[1] The congressional vote is measured as the Democratic percent of the national two-party vote for the House of Representatives, minus 50 percent so that a 50-50 vote becomes the zero baseline. The congressional vote is modeled separately for presidential and midterm years.[2]
The starting point is the classic “Michigan” model identified with the American Voter authors. In the presidential year, the national vote for Congress is modeled as a function of the normal vote, plus short-term forces. The normal vote is represented by the national division of party identification in October of the election year. October party identification is measured as the percent Democratic minus percent Republican, pooling all available October polls.[3] The short-term forces are represented by the presidential vote. The presidential vote is measured as the Democratic percent of the national two party vote minus 50 percent.
For the presidential year model, two additional variables are included to represent possible balancing behavior. One is the expectation of the presidential vote winner in terms of knowledgeable voters’ perception of the probability of a Democratic presidential victory. This crucial variable is estimated from the gambling odds from betting markets as reported on election eve. Details are presented below.