It is often argued that voters rarely monitor the behavior ...



The Presidential Shadow On Midterm House Elections:

Presidential Support, Presidential Agendas, and Seat Loss

Brian J. Gaines

bjgaines@uiuc.edu

and

Timothy P. Nokken

tnokken@uiuc.edu

Department of Political Science

and

Merriam Laboratory for Analytic Political Research

University of Illinois at Urbana-Champaign

361 Lincoln Hall

702 S. Wright Street

Urbana, IL 61801-3696

November 1998†

Abstract

With only two exceptions, each midterm election since 1860 has seen the president’s party lose seats in the House of Representatives. Research on this midterm loss phenomenon is usually conducted at the aggregate level, so that the behavior of individual members of Congress is conspicuously absent from the factors hypothesized to drive the election results. Building on the work of Brady et al. (1996), we examine the midterm loss at the district level in elections from 1958 to 1994. We conjecture that legislators sometimes face a difficult choice, between backing a comparatively extreme president, at the risk of alienating their constituents, or else “playing to the district,” at the risk of harming and annoying their president. An implication is that a record of presidential support is sometimes perilous. We find that a Representative’s level of support for the president’s agenda does, under some circumstances, have a significant, positive effect on the probability of losing in a midterm. We are not able, however, to identify a priori in which elections this effect is most pronounced.

One of the most robust empirical regularities in American elections is the midterm loss: with only two exceptions (the old case of 1934 and the brand new case of 1998), the presidential party has lost House seats at every midterm since 1860. A number of reasons have been proposed for this pattern. Notably, few explain midterm losses in terms of the behavior of members of Congress (MCs). Instead, explanations for the seemingly inevitable presidential party losses have converged on a number of factors beyond the direct control of individual members. In part, this is a byproduct of the favored level of analysis: the typical paper uses national elections as observations, models aggregate seat loss, and so focuses on a set of aggregate explanatory variables such as turnout, economic performance indicators, and so on. Member activities do not easily fit into such a framework. In this paper, we drop down a level and seek to determine whether the probability of reelection for US Representatives is significantly affected by their roll-call voting behavior. Specifically, we investigate the question of whether presidents sometimes pursue agendas that leave their partisan counterparts in the House with a Hobson’s choice: support the president and risk alienating constituents, or put the district ahead of the party, and risk presidential wrath.

Aggregate midterm losses of House seats (or, sometimes, vote share) are usually modeled as a function of changes in the larger political environment. Hence favorite predictors include: presidential popularity (Kernell, 1977; Marra and Ostrom, 1989); an alleged tendency for public dissatisfaction with (early) presidential performance (Erikson, 1988); prevailing economic conditions (Tufte, 1975; 1978); changes in levels and composition of voter turnout from presidential to midterm elections (A. Campbell, 1966; J. Campbell, 1993); voter evaluations of the president’s party (Abramowitz, Cover, and Norpoth, 1986); and attempts by voters to “balance” policy outcomes by dividing control of the White House and Congress (Alesina and Rosenthal, 1995). Virtually all of these interpretations seem to imply that midterm losses are largely independent of members’ actions in office. Thus, somewhat perversely, it has become common to neglect the electoral consequences of individual-level member roll-call voting behavior when explaining congressional election results.

There are, of course, some analyses of how individual MCs’ voting records can contribute to their defeat. For example, a large literature, implicitly grounded in the delegate notion of representation, has analyzed “shirking,” by which it is usually meant casting roll call votes that do not correspond well to the preferences of constituents (e.g. Lott (1987), Dougan and Munger (1989), Kalt and Zupan (1990), Lott and Davis (1992), Lott and Bronars (1993)). These works are mostly concerned with the consequences of shirking, and so the sources of MC miscreance are not much explored, except insofar as it follows a decision to retire. In this paper, by contrast, we consider a simple dilemma that many MCs can face. Representatives from the president’s party can get caught between a comparatively extreme president and comparatively moderate constituency opinion. Our core hypothesis is that at least some of the time, presidential support should be perilous (yet tempting) for MCs, so that it ought to show a significant, direct effect on the probability of losing office. We do not attempt anything like a complete model of presidential strategy, congressional responsiveness to presidential cues, the roll-call generating process, agenda control, and so on. For now, we focus on only a small piece of the theory by developing a simple model of re-election probabilities that incorporates as an explanatory variable an index of the presidential support in roll-call voting behavior over the immediately prior period.

Ours is not an altogether novel thesis. Brady, Cogan, Gaines, and Rivers (1996) have shown that roll-call voting behavior did, indeed, have a significant effect on the electoral fate of incumbent House Democrats in the 1994 elections. Higher levels of presidential support by Democratic Representatives over 1993 and 1994--especially by those members in whose districts Clinton ran poorly in 1992--led to a greater likelihood of losing in 1994. Hereafter, then, we explore how general is that result. How and where does presidential support hurt? Is there evidence for the peril of standing by a (relatively extreme) president in other modern elections? If so, does the effect occur when expected? That is, is there a reliable way to determine, ex ante, which presidents (if any) seem to have placed their House colleagues in jeopardy by pressing too extreme an agenda?

The paper proceeds as follows. Section two briefly reviews prior work on the midterm loss. Section three discusses what can happen when the preferences of the president and his partisans in Congress diverge. We expect presidential support by MCs to affect their re-election chances only when presidents are in some sense out-of-step, and, in those cases, to work in a multiplicative rather than additive fashion. That is, for members with districts having relatively left- (right-) leaning electorates, yielding to pressure to support a president who veers left (right) is no liability. For moderates, by contrast, the dilemma is real. We consider possible means of identifying the necessary conditions for this dilemma to show. In section four, we replicate and extend a model of individual seat loss developed by Brady et al. (1996). Section five concludes.

2. Explanations of Midterm Loss

Modern midterm elections nearly always feature presidential-party seat loss, but the magnitude of these losses varies considerably. Large losses have been most common when the president’s popularity was low, as was the case with Clinton in 1994 and Nixon in 1974. Indeed, (national) presidential popularity is a consistently significant predictor of the size of the midterm loss (e.g. Lewis-Beck and Rice 1992: ch.4). An unpopular president, however, is not necessarily a recipe for an electoral bloodbath, as Reagan showed in 1982 by losing a fairly modest 26 seats despite widespread disapproval of his job performance. In part, the difference is a simple matter of numbers: there were fewer Republicans in non-safe seats in 1982 than there were Democrats at risk in 1994. Hence, some variety of “exposure” variable is also a staple of the aggregate models. More importantly, though, it is not self-evident that voters will automatically take out displeasure with a sitting president on his partisan counterparts in House races. Some exploration of when, how, and why such linkage occurs is in order.

Consider three major classes of explanation for midterm losses. The key difference between “on-” and “off-” year elections could be turnout patterns, i.e. systematic differences in the composition of the midterm and presidential-year electorates. Presidential popularity makes only a slight contribution to such accounts, insofar as it might be one factor causing mid-term ennui. Congressional behavior, meanwhile, is beside the point. A very different take on the election patterns is that the most important difference resides in the set of issues and evaluation criteria that tend to play out in each kind of election. Specifically, there might be a natural cycle to presidential approval that makes unwitting victims of legislators associated with the president at midterms. The president lies at the center of these accounts, whether they are framed in terms of a political business cycle, a tendency for modern presidential elections to breed exaggerated expectations, or whatever. Members of Congress, by contrast, are mere ciphers in these explanations. What matters about MCs is only their partisan label, and not the degree to which they are somehow linked (or linkable) to the president.

Finally, various “balancing” models propose that midterms allow voters an opportunity to play the Legislative branch against the Executive branch, so that the midterm loss should be understood as originating in the fact that some portion of the electorate has policy preferences more moderate than both the right-wing ideals of the Republicans (president and/or House members) and the left-wing goals of the Democrats (in one or both branches). Such models could involve more careful consideration of the actual profiles of MCs, but, thus far, balancing theories have advanced more on the formal and theoretical front than on the empirical side. Where authors have tested these hypotheses at the district level, their operationalizations have been unsatisfactory (e.g. Burden and Kimball 1998, cf. Cho and Gaines 1999).

Despite a wide array of approaches, then, research on midterm elections usually implies that the outcomes are influenced by conditions beyond the direct control of individual MCs. There is, nonetheless, a surface plausibility to the conjecture that legislators should have a direct role in this phenomenon too. We do not propose to ignore the president, but, rather, to explore which Representatives seem most likely to be tied to the president in voters’ minds, making them especially likely to be the victims of the almost inevitable midterm purge.

3. Presidential Support Scores and Midterm Election Outcomes

Accordingly, we wish to consider whether roll-call voting by individual Representatives can be shown to affect their re-election prospects. Such analysis need not be limited to midterm elections: the conjecture that some Representatives allow themselves to be pulled away from median constituency opinion by presidential pressure, and then suffer accordingly, is not midterm-specific. Nonetheless, we focus exclusively on midterms here, chiefly because these elections do not allow voters any direct say on presidential performance. Midterms thus provide a fairly clean test of whether voters appear to punish Representatives for ideological drifting. By limiting attention to midterms, one can duck a set of complications stemming from untangling the various possible sources of coattail voting. The theory is also not specifically oriented to the House, but for simplicity we consider only Representatives in the analysis that follows, leaving Senators for another day. Finally, for methodological reasons to which we shall return, we opt to study the relatively crude win-or-lose variable hereafter, forgoing any modeling of actual vote shares.

Brady et al. (1996) show that increased support for the president significantly influenced the likelihood of defeat for incumbent House Democrats in the 1994 midterm elections. They contend that Clinton’s first two years of office featured a sufficiently leftist policy agenda that, by 1994, a record of having supported Clinton had become a serious liability for moderate Democrats. Of course, if this was true, it was true only because Republican challengers took pains to call this association to voters’ attention. And, presumably, there were several observable instances of overly-Clinton-friendly Representatives losing precisely because the president had succeeded in building House coalitions that included some (reluctant) moderates. The general version of this dynamic, then, is that when a president is an outlier within his party(an especially liberal Democrat or an unusually conservative Republican(and that president is aggressive in his pursuit of congressional support, the president’s partisans in Congress end up cross-pressured. If one can identify those whose districts are least favorable to the president, these are the members for whom support should be costly. In turn, one would like to identify which presidents have indeed shown signs of having overplayed their hands in the years preceding the midterms.

Brady et al. rely on journalistic reporting and the reader’s tacit concurrence that Clinton’s leftness in 1993 and 1994 was self-evident. Clearly, it is a necessary condition for the hypothesized dynamic that the president be more extreme than his party’s House delegation. Identifying which presidents fit this bill in a systematic manner requires some metric for presidential behavior. It is important too that one measure not mere presidential preferences in the abstract, but the presidential agenda to which legislators can be plausibly linked, more or less strongly, according to their actual levels of support. The most obvious way to estimate locations for presidents and legislators in a common ideological space is to treat legislators’ roll call votes and presidents’ announced positions on these roll call votes as equivalent preference-revelation mechanisms. Then, for example, Poole and Rosenthal’s various Nominate® procedures for scaling observed binary votes into voter ideal points can be applied to presidents (McCarty and Poole, 1995). Alternatively, presidents can be scored according to their announced positions using interest-group indices (e.g. Zupan 1992). The ADA, for example, chooses 20 votes in each chamber, and then scores MCs “right” or “wrong” on each vote. Their liberalism score is simply the percentage of all votes on which legislators voted the ADA-preferred way, so presidents can be given ADA scores on the basis of their announced preferences on this identical set of votes.

Table 1 shows how recent presidents have scored, and how they compared to their House delegations. Panel A., which uses W-Nominate scores, ranks the presidents by their distance from their parties’ House delegations. Figures 1 and 2 illustrate, using one case in which a president appears “extreme,” (Johnson in 1966), and one in which he does not appears to be an outlier relative to his party (contra Brady et al., Clinton in 1994). A difficulty in detecting which presidents were genuine outliers relative to their House delegations, however, is that our scores for legislators presumably already contain presidential influence. Some of the reason Clinton appears much closer to House Democrats by 1994, as compared with Johnson and the House Democrats in 1966, could be because Clinton was more successful in pressuring Representatives to vote with him.[1] Indeed, although Johnson (as of 1966) lies left of Clinton (as of 1994) in Nominate space, much of the difference in their separation-from-House-median scores is due to Clinton’s House Democrats lying well to the left of Johnson’s. In general, then, the observed gap between a president and his House colleagues is not precisely the measure of presidential outlyingness we are after.

[Table 1 – Nominate and ADA Scores, Presidents and Their House Colleagues]

On the other hand, to the extent that a president’s ideological rating is moderated because he was strategic with respect to what could pass Congress, we no longer expect to see signs of the presidential-support dilemma. Hence, the most important statistic for identifying presidents whose programs put their colleagues at peril should be the president’s rating, rather than his separation from his House copartisans. Focusing only on the presidents’ scores, whether using Nominate scores in panel A or ADA scores in panel B, one would reject the maintained hypothesis of Brady et al., that Clinton had been extreme in his first two terms.

[Figure 1 – Johnson and House Democrats, 1965-66, W-NOMINATE Scores]

[Figure 2 – Clinton and House Democrats, 1993-94, W-NOMINATE Scores]

The next point about Table 1 is that there is some disagreement between the different scoring techniques. Bush is the single most conservative president in the set by ADA scores, but is only moderately conservative (a mirror image of fellow one-termer Carter) in W-Nominate metric. Johnson is Reagan’s opposite under ADA scheme (a) and W-Nominate, but scores to the right of Clinton if one takes advantage of the extra information about presidents contained in their announced preferences over Senate votes when calculating ADA scores. We are not the first to notice that rival methods of computing ideological scores from roll call votes perform quite differently (see Adams and Fastnow 1998). If we attempt to take account of all sets of scores, a general conclusion might be that Reagan and Johnson are the most extreme in the set. That corresponds to conventional wisdom, and the quantification, alas, has contributed little. Clinton, notwithstanding gays-in-the-military, nationalizing health care, abandoning promised tax cuts, and so on, does not appear particularly extreme in 1994, except under the second method of ADA scoring, in which he gets the most liberal rating of any president in the table, except for Kennedy.

Our main conclusion from this initial stab at analyzing presidential agendas ideologically is pessimistic. Quantification is desirable and should be possible, but at present we have relatively little confidence that we are able to improve on prior subjective beliefs grounded in informal, hard-to-quantify impressions. Our hunch at the outset was that prime examples of presidents over-reaching and placing their partisan colleagues in the House in jeopardy are Clinton in 1994, Reagan in 1982, and Johnson in 1966. Support of Nixon in the years prior to 1974 might have been costly for Republicans, but for different reasons: what (might have) mattered was the incumbent’s association with a scandal-ridden figure rather than with an overly conservative agenda. We proceed next to estimations.

4. A Simple District-Level Model of Midterm Election Outcomes

In this section, we estimate probit models of the probability of victory for those members of the president’s party who seek reelection.[2] Like Brady et al., we include a relatively small set of control variables: turnout, the president’s share of the two-party vote in the preceding election, the share of the vote won by significant non-major-party candidates (Perot in 1994, Anderson in 1980, Wallace in 1968) and an indicator variable for a challenger who has previously held any elective office.

Voter turnout, of course, holds a prominent place in the literature on midterm election outcomes. Campbell’s surge-and-decline thesis predicts that the lower levels of turnout observed in midterms contributes to the defeat of members of the president’s party. At least two kinds of turnout effect seem plausible. First, lower turnout might lead to an increased probability of defeat for incumbent members of the president’s party. No longer mobilized by the presidential election, marginal partisans might stay home. In other words, there are no coattails to help Representatives in marginal districts keep their offices. Second, one might expect higher turnout to assist the out party in midterms on different grounds. If loss in a presidential election can be understood as turnout having been above average for the winner’s party and below for the losers, regression to the mean should assist the out party in the next election.[3] Hereafter, we measure turnout as the total number of votes cast in the House election at the midterm divided by the total voting age population.

The president’s share of the two-party vote from the previous presidential election is essentially a proxy for district “type.” Specifically, districts where the president ran strong ought generally to be more safe for that party, so that this variable can be considered a normal vote estimate. The variable can also be construed as a measure of the appeal of the particular president, so that it can be used to separate those districts in which a record of support is a liability from those in which such a record might be worn as a badge of honor. Our expectation is that the coefficient on this variable should be positive and significant. Where Democratic (Republican) presidents run well, their congressional candidates should also run well, ceteris paribus.

It is often argued that the emergence of significant third-party challenges are indicative of voter dissatisfaction (Sundquist, 1983). Exactly what effect Ross Perot’s 1992 candidacy had on that election remains a subject of some controversy, chiefly between those who maintain that Perot harmed Bush most and those who propose that he drew away roughly equal numbers of Democratic and Republican votes. By some tellings, the Perot factor played a part in the dramatic 1994 result as well. If, for example, Perot mobilized previously inactive citizens who then turned out again in 1994, he may have injected into the process a set of disgruntled moderate voters who turned on Clinton via House votes in 1994. Or, again, Perot’s strength may not only have masked Republican tendencies, but assisted Democrats in getting elected by discouraging voters who might otherwise have cast straight Republican ballots from doing so. In that case, one could expect a negative correlation between Perot’s showing in 1992 and the odds of Democrats returned in 1992 winning again in the 1994 race once the distraction of a non-major-party force was removed. Brady et al. in fact find that the “Perot factor” was not a significant contributor to Democratic defeats.

Anderson, like Perot, aimed to capitalize on voter dissatisfaction with the two parties. Arguably, he ran to the left of both Reagan and Carter. Unlike Perot, he did not pose much of an electoral threat. Not even Anderson’s biggest fan (Gary Trudeau?) predicted that the 1980 presidential race would end in the House. And, although as of October 1980 the election was perceived to be much closer than it ultimately turned out to be, few thought Anderson would draw votes in a manner that would systematically advantage one party or the other. Hence, it is hard to construct a rationale for why his candidacy should have had after-shocks in 1982. Finally, Wallace’s success was largely confined to the South. Since most southern Representatives were Democrats, we would not expect Wallace’s candidacy to have had a significant effect on the reelection bids of many of Nixon’s partisans in 1970. In sum, without endorsing a particular scenario by which the third-candidate vote matters, we will include a control variable for the Perot, Anderson and Wallace votes in the relevant models to assess the effect their candidacies had on midterm election outcomes. In each case, we use their share of the district-level three-way presidential vote, and we hypothesize no significant effects.

Experienced challengers generally outperform their colleagues who have never won any sort of office. A large literature has explored how best to measure candidate and challenger quality, and most analysts still find that Jacobson’s dummy for previously-elected works about as well as far more complicated measures. Hence, we include an indicator coded one for those challengers who have held previous office, zero otherwise. The coefficient on this variable should be statistically significant and negative in each of the midterms analyzed.

The factor most often omitted from analysis of midterms is some variable capturing member behavior in reelection bids. This analysis focuses upon a relatively narrow aspect of member behavior, support for the president’s announced positions on roll call votes. Members of course have at their disposal a vast array of resources to help them stay in power. That said, midterm elections are a special case where the electoral welfare of the president’s party in Congress is explicitly linked to the performance of the president. A common rhetorical tactic on the midterm campaign trail is to link members of the president’s party to the president, especially if he is unpopular. In such instances, support for the president’s agenda could prove to be an important issue in a congressional campaign. We expect the coefficient on Congressional Quarterly’s presidential support scores be negatively related to a member’s probability of winning in years in which the president was extreme and in districts in which the president is comparatively weak. That passage in italics is important. We have already punted on quantifying presidential extremism, and so we proceed to analyze the midterms year by year, with only fuzzy expectations about when we expect to see signs of presidential support hurting incumbents. On the other hand, it is important to re-emphasize that we do not expect presidential support to be uniformly damaging, even in years in which the extreme-president condition is met. Our hypothesis is that presidential support interacts with district safety, so that increasing levels of support are associated with greater likelihood of defeat in relatively moderate districts, but not in relatively extreme districts. This makes the specification of the model tricky, since we expect roughly the following combinations for member’s presidential support, district presidential vote, and member’s risk of defeat: (low, low, medium); (high, low, high); (low, high, low); (high, high, low). For simplicity, Brady et al. relied strictly on the non-linear properties of the probit estimator to handle this property. We copy their basic specification, but also explore an explicit interactive term (see Nagler 1991 on specifying interactive terms in non-linear models).

In the models that follow, the observed dependent variable is always victory or defeat (the latent variable is the probability of victory) for members of the president’s party in midterm elections. We estimate probit models for re-election of presidential-party incumbents in all midterm elections from 1958 to 1994.[4] There are various decisions to be made in exact operationalization, for example whether to use presidential support scores for the year of the election, the two years prior, only the first of the two years, etc. Our main concern lies in the non-linear and multiplicative effects we expect from presidential support. For each year we estimate two models, of the following form:

Pr(Y=0) = (((0+(1X1+(2X2+(3X3+(4X4+(5X5)

Pr(Y=0) = (((0+(1X1+(2X2+(3X3+(4X4+(6X5D1+(7X5D2)

where Y is an indicator variable for re-election; X1 represents turnout in the midterm election; X2 is an indicator for an experienced challenger; X3 is the percentage of the total vote won by any third candidate in the prior presidential election; X4 is the share of the two-party vote won by the president in the prior presidential election; X5 is the incumbent’s level of presidential support in roll-call voting; D1 and D2 are dummies for districts having values of X4 (presidential vote) below or at least above the median value of X4 for the presidential incumbents, respectively; ((() represents the standardized cumulative normal distribution function; and, observations are presidential-party incumbents. The second model, which estimates two coefficients on X5, one for “safe seats” ((7) and one for “at-risk” seats ((6), is a simple, explicit attempt to get at the multiplicative logic already noted more directly than the first specification.

We now extend the analysis and estimate the model for each midterm election from 1958 to 1994. We estimate separate models for each election because it is not clear that the factors have the same influence at each point in time. (Ultimately, we might test the poolability of the data explicitly.) Of particular interest is whether support for the president increases the likelihood of defeat in midterm elections. The results of the estimates of the first model are displayed in Table 2. Contrary to the turnout-based explanations of midterm loss, voter turnout is, with the exception of the 1962 and 1966 elections, statistically insignificant. As expected, in districts where the president ran well, House members generally perform better at the midterm. The president’s vote share in the previous election is significant and positive in eight of the ten elections, statistically insignificant only in 1962 and 1974. Facing an experience challenger significantly reduces a member’s reelection prospects for incumbent House members from the President’s party in half of the midterm elections we analyze (1958, 1966, 1974, 1978, and 1994)—surprisingly few given our expectations about the importance of challenger quality. The third party challenges of Wallace, Anderson, and Perot played no significant part in the defeat of incumbent House members of the presidents party, which is not an unexpected result.

[Table 2 -- Maximum Likelihood Estimates, Probability of Winning, 1958-1994]

We expect presidential support to be most salient in 1994 and in those cases where presidents were ideologically extreme compared to their fellow partisans in the House. Supporting the president would seem to have been most dangerous during the Johnson (1966) and Reagan (1982 and 1986) Administrations. We find, however, that presidential support played a significant role in midterm outcomes in only two elections, 1994 and 1970. Increased presidential support among Democrats in the 103rd House contributed to their defeat in the 1994 midterm elections. Contrary to the expectations, the effect in 1970 runs the other way. Support of Nixon actually helped the electoral prospects of incumbent Republicans. That results is initially surprising, though it might be interpreted as the flip side of our hypothesis: a sufficiently moderate president can insulate his partisan colleagues in competitive seats.

That interpretation calls attention again to the importance of the interactive specification. If the negative (or positive?) consequences of increased presidential support should primarily affect those members of the president’s party from districts where the president ran poorly, our interest lies in the interaction of a member’s presidential support score with dummy variables indicating whether the president ran poorly in the given district. The results for our second model are displayed in Table 3. We include the same set of control variables as in the previous model. The effect of these variables are less pronounced than in the previous model. Presidential vote share significantly contributed to incumbent members’ victory in only two elections: 1970 and 1994. Experienced challengers, once again, reduced the probability of reelection in half of the midterm elections we analyze, the effect reaching statistical significance in 1958, 1966, 1974, 1978, and 1994. As in the previous model, we find no evidence that significant third party presidential challenges contribute to midterm losses. These estimates thus provide further disconfirming evidence for turn-out based explanations of midterm losses.

Identifying members in “at-risk” seats improves somewhat our estimation of the effect of individual-level behavior on reelection prospects. Consistent with the expectations that cross-pressured members in marginal seats take the greatest risk in supporting the president’s agenda, the interactive term was statistically significant and correctly signed in 1966 and 1994. LBJ’s policies were, by most accounts, far to the left of the preferences of a large number of his fellow Democrats in the House, and the ones from not-very-Democratic seats suffered when they supported him. Democrats from marginal districts had swept into office in 1964 with the help of Johnosn’s landslide victory, but they saw their prospects for reelection diminish with increased levels of support for Johnson’s Great Society legislation. Members from safe seats, however, experienced no penalty for their support of the president, consistent with our expectations.

Similarly, in 1994, for Democrats from marginal districts, higher levels of presidential support served to increase the probability of defeat. In 1994, in fact, increased presidential support by Democratic House incumbents was detrimental even in districts where the president ran well in 1992.[5] Support for the Clinton agenda appears to have been a risky undertaking in 1994.

Also of note are the statistically significant results from the 1970 and 1990 midterms where, contrary to our original expectation, increased presidential support actually increased the likelihood of reelection for Republican incumbents. The effect is especially strong for those members from districts where Nixon ran poorly in the 1968 elections. We also detect an electoral boost for members from relatively safe districts in 1970 and 1990. It appears that presidential support may, in some instances, actually benefit members of the president’s party in midterm elections.

[Table 3 -- Maximum Likelihood Estimates, Probability of Winning, 1958-1994]

Results from both sets of probit models reveal some signs that individual-level behavior can contribute to midterm loss. The results are not, however, an unambiguous vindication of the theory. In some elections in which we expected to observe significant negative effects for support of ideologically extreme presidents (e.g. 1982), no such effects are discernible. Significantly negative consequences from increased levels of support for LBJ were observable, once one restricts attention to members from marginal seats. The midterm elections of 1970 provide somewhat surprising results, since the electoral fortunes of Republican House members were apparently buoyed by increased levels of support for Nixon. Since Nixon’s ideological profile in 1969 and 1970 was plainly moderate, this is not inconsistent with our conjecture, even though it was not an effect we set out to find.

Conclusion

Explanations of midterm losses in the House focus almost exclusively on the aggregate patterns. A variety of models of total seat loss perform very well as forecasting devices. They are, however, somewhat less helpful in understanding the micro-level of the midterm loss. In particular, they offer no guidance as to who will lose, or why. We have taken a different approach, by exploring a dynamic whereby some individual Representatives place themselves at risk at midterms by following their leader. There is a surface plausibility to the conjecture that the likelihood of defeat might, under certain conditions, be influenced by past roll-call voting behavior. Our empirical results are somewhat mixed: we observe the dynamic as expected in 1994, though the negative influence of presidential support extended even into safe seats, somewhat to our surprise. In 1966, another year in which most reasonable observers would concur that the president was well left of his House colleagues, the result is a better match to our theory. On the other hand, we detect no such effects for 1982 or 1986, against expectations.

A prudent conclusion is that we have weak confirmation of the theory, but that further analysis is warranted. Several subsidiary analyses, for example, can bolster our confidence that the estimation is capturing the presidential-support dilemma. In 1994, models in which voting behavior on a few, key issues (gays-in-the-military, the budget, health-care reform) was substituted for the overall presidential-support score performed about equally well. The possibility remains that it is not really objective support for the president’s program over the whole term that matters electorally, but, rather, whether the Representative voted “with” or “against” his median constituent in a few highly salient cases. That is a rival hypothesis crying out for more careful examination. Further, it would improve our confidence in the theory if we detect no presidential-support effects in open seats or seats of non-presidential incumbents.

Generally, a number of alternative specifications may provide us with a better picture of how individual member behavior influences electoral outcomes. In particular, modeling defeat-or-victory sets a very high standard for the level of peril in evidence from pro-presidential voting. It would seem advisable also to model vote shares, in search of signs that some members pay for their support, but hang on to win the elections (narrowly) all the same. The chief barrier to that project is simply technical: a sensible model should incorporate normal vote, incumbency effects, inter-election (uniform) swing, and district-level behavior by some incumbents only at only some midterms, and we have yet to devise an acceptable panel specification.

Though far from conclusive, our results are, we believe, strong enough to confirm that the precise links between roll-call voting, presidential cues, and electoral consequences deserve further attention. In particular, the midterm loss is well understood as an aggregate phenomenon, but remains relatively mysterious on the ground. That should bother not only those running scared (the Representatives), but also those running regressions.

References

Abramowitz, Alan I., Albert D. Cover, and Helmut Norpoth. 1986. “The President’s Party in Midterm Elections: Going from Bad to Worse.” American Journal of Political Science 30, 3 (August): 562-576.

Adams, Greg D. and Christina Fastnow. 1998. “On the Differences Between Legislative Ideology Measures.” Paper presented at annual meetings of Midwest Political Science Association, Chicago, IL.

Alesina, Alberto and Howard Rosenthal. 1995. Partisan Politics, Divided Government, and the Economy. Cambridge: Cambridge University Press.

Brady, David W., John F. Cogan, Brian J. Gaines, and Douglas Rivers. 1996. “The Perils of Presidential Support: How the Republicans Took the House in the 1994 Midterm Elections.” Political Behavior 18, 4 (December): 345-367.

Burden, Barry C. and David C. Kimball. 1998. “A New Approach to the Study of Ticket Splitting.” American Political Science Review 92, 3 (September): 533-544.

Cain, Bruce, John Ferejohn, and Morris Fiorina. 1987. The Personal Vote. Cambridge, MA: Harvard University Press.

Campbell, Angus. 1960. “Surge and Decline: A Study of Electoral Change.” Public Opinion Quarterly 24, 3 (Fall): 397-418.

Campbell, James E. 1993. The Presidential Pulse of Congressional Elections. Lexington: University of Kentucky Press.

Cho, Wendy K. Tam and Brian J. Gaines. 1999. “Reassessing the Study of Split-Ticket Voting” Typescript, University of Illinois.

Dougan, William R. and Michael Munger. 1989. “The Rationality of Ideology.” Journal of Law and Economics 32, 1 (April): 119-142.

Erikson, Robert S. 1988. “The Puzzle of Midterm Loss.” The Journal of Politics 50, 4 (November): 1011-1029.

Fenno, Richard F., Jr. 1973. Congressmen in Committees. Boston: Little, Brown.

(((((. 1978. Home Style: House Members in their Districts. Boston: Little, Brown.

Fiorina, Morris P. 1981. Retrospective Voting in American National Elections. New Haven: Yale University Press.

Hinckley, Barbara. 1967. “Interpreting House Midterm Elections: Toward a Measurement of the In-Party’s ‘Expected’ Loss of Seats.” American Political Science Review 61, 3 (September): 694-700.

Jacobson, Gary C. 1997. The Politics of Congressional Elections, 4th ed. New York: Longman.

Jacobson, Gary C. and Samuel Kernell. 1983. Strategy and Choice in Congressional Elections, 2nd ed. New Haven: Yale University Press.

Kalt, Joseph P. and Mark A. Zupan. 1990. “The Apparent Ideological Behavior of Legislators: Testing for Principal-Agent Slack in Political Institutions.” Journal of Law and Economics 33, 1 : 103-131

Kernell, Samuel. 1977. “Presidential Popularity and Negative Voting: An Alternative Explanation of the Midterm Congressional Decline of the President’s Party.” American Political Science Review 71, 1 (March): 44-66.

Kuklinski, James H. and Darrell M. West. 1981. “Economic Expectations and Mass Voting in United States House and Senate Elections.” American Political Science Review 75, 2 (June): 436-447.

Lewis-Beck, Michael S. and Tom W. Rice. 1992. Forecasting Elections. Washington, DC: CQ Press.

Lott, John R., Jr. 1987. “Political Cheating.” Public Choice 52, 2: 169-186.

Lott, John R., Jr. and Stephen G. Bronars. 1993. “Time Series Evidence on Shirking in the U.S. House of Representatives.” Public Choice 76, 1-2 (June): 125-149.

Lott, John R., Jr. and M. L. Davis. 1992. “A Critical Review and Extension of the Political Shirking Literature.” Public Choice 74, 4 (December): 461-484.

Marra, Robin F. and Charles W. Ostrom. 1989. “Explaining Seat Change in the U.S. House of Representatives, 1950-1986.” American Journal of Political Science 33, 3 (August): 541-569.

Mayhew, David R. 1974. Congress: The Electoral Connection. New Haven: Yale University Press.

McCarty, Nolan M. and Keith T. Poole. 1995. “Veto Power and Legislation: An Empirical Analysis of Executive and Legislative Bargaining from 1961 to 1986.” The Journal of Law, Economics, and Organization 11, 2 (October): 282-312.

Nagler, Jonathan. 1991. “The Effect of Registration Laws and Education on U.S. Voter Turnout.” American Political Science Review 85, 4 (December): 1393-1405.

Poole, Keith T. 1997. “Changing Minds? Not in Congress!” GSIA WP#1997-22 Graduate School of Industrial Administration, Carnegie-Mellon University.

Poole, Keith T. and Howard Rosenthal. 1997. Congress: A Political-Economic History of Roll Call Voting. New York: Oxford University Press.

Sundquist, James. 1983. Dynamics of the Party System: Alignment and Realignment in the United States. Washington, DC: Brookings Institution

Tufte, Edward R. 1975. “Determinants of the Outcomes of Midterm Congressional Elections.” American Political Science Review 69, 3 (September): 812-826.

(((((. 1978. Political Control of the Economy. Princeton: Princeton University Press.

Zeidenstein, Harvey. 1972. “Measuring Congressional Seat Losses in Mid-Term Elections.” Journal of Politics 34, 1 (February): 272-276.

Zupan, Mark. 1992. “Measuring the Ideological Preferences of U.S. Presidents: A Proposed (Extremely Simple) Method.” Public Choice 73, 3 (April): 351-361.

Table 1. Presidents Compared To Their Parties’ House Delegations

A. W-Nominate Scores, President and Presidential Party House Median

|President |Midterm Election |President’s |Presidential Party |President- |Rank |

| |Year |Score |House Delegation |House Delegation | |

| | | |Median |Separation | |

|Eisenhower |1958 |0.51 | |0.24 | |0.27 | |7 |

|Kennedy |1962 |- 0.85 | |- 0.42 | |0.43 | |1 |

|Johnson |1966 |- 0.72 | |- 0.31 | | 0.41 | |2 |

|Nixon |1970 |0.26 | | 0.35 | |- 0.09 | |10 |

|Nixon |1974 |0.55 | | 0.34 | | 0.21 | |8 |

|Carter |1978 |- 0.53 | |- 0.24 | | 0.29 | |5 |

|Reagan |1982 |0.72 | | 0.33 | | 0.39 | |3 |

|Reagan |1986 |0.77 | | 0.41 | | 0.36 | |4 |

|Bush |1990 |0.52 | | 0.23 | | 0.29 | |5 |

|Clinton |1994 |- 0.43 | |- 0.44 | |- 0.01 | |11 |

|Clinton |1998 |- 0.67 | |- 0.55 | | 0.12 | |9 |

B. ADA Score, President and Presidential Party House Median

|President |Midterm Election |President’s |Presidential Party |Presidential |

| |Year |Score* |House Delegation |Outlyingness |

| | | |Median |Rank |

| | |(a) (b) | |(a) (b) |

|Eisenhower |1958 |87 |66 | |-- |6 |10 | |

|Kennedy |1962 |100 |94 | |-- |1 |4 | |

|Johnson |1966 |95 |72 | |74 |3 |9 | |

|Nixon |1970 |33 |24 | |14 |10 |6 | |

|Nixon |1974 |13 |18 | |17 |6 |5 | |

|Carter |1978 |70 |74 | |53 |8 |8 | |

|Reagan |1982 |5 |5 | |13 |4 |3 | |

|Reagan |1986 |5 |3 | |10 |4 |2 | |

|Bush |1990 |0 |0 | |11 |1 |1 | |

|Clinton |1994 |69 |76 | |75 |9 |6 | |

* Presidential “ADA scores” were computed using announced presidential positions on ADA-selected roll call votes in the two years preceding the midterm. The scores are computed as (100r/(r+w)) where r is the number of “right” votes (in ADA terms) and w is the number of “wrong” votes, and we rounded to whole numbers. There is no expectation that presidents take sides on all votes held in Congress, so we divide by (r+w) rather than the total number of ADA-picked votes, thus not penalizing presidents for “abstention.” (By contrast, the ADA now treats abstention by legislators as equivalent to voting the “wrong” way.) The difference between (a) and (b) is that (a) uses only the votes chosen by the ADA in the House, while (b) scores presidents on their House and Senate announced positions. Our scores do not match those of Zupan, both because they span multiple years and because of apparent coding errors in his analysis.

Table 2. Probability of Presidential-Party Incumbent Victories At Midterms

(Maximum Likelihood Probit Coefficient Estimates)

|Variable |1958 |1962 |1966 |1970 |1974 |1978 |1982 |1986 |1990 |1994 |

|Constant |-3.10 |-0.71 |1.58 |-3.74 |2.00 |0.46 |-4.98 |-2.11 |1.62 |0.20 |

| |(1.73) |(1.42) |(0.60) |(1.94) |(1.76) |(1.37) |(2.06) |(2.73) |(2.03) |(1.54) |

| | | | | | | | | | | |

|Turnout |0.44 |-2.26 |-1.19 |0.36 |-1.73 |-2.91 |-3.30 |-4.45 |-3.07 |-0.69 |

| |(0.44) |(1.29) |(0.41) |(2.51) |(1.39) |(2.17) |(2.06) |(2.95) |(2.12) |(1.76) |

|Presidential Vote | | | | | | | | | | |

| |6.43 |5.03 |1.91 |4.69 |0.06 |4.16 |11.65 |15.08 |-0.05 |7.38 |

| |(2.47) |(2.72) |(0.97) |(2.29) |(1.96) |(2.11) |(3.19) |(6.16) |(3.45) |(1.80) |

|Experienced | | | | | | | | | | |

|Challenger |-0.78 |0.08 |-0.75 |-0.38 |-0.53 |-0.73 |-0.07 |-0.83 |-0.46 |-0.77 |

| |(0.24) |(0.38) |(0.21) |(0.35) |(0.23) |(0.29) |(0.35) |(0.54) |(0.38) |(0.26) |

| | | | | | | | | | | |

|Voting Record |-0.00 |0.01 |-0.01 |0.04 |-0.00 |-0.00 |0.01 |-0.05 |0.02 |-0.03 |

| |(0.01) |(0.01) |(0.01) |(0.02) |(0.01) |(0.01) |(0.01) |(0.03) |(0.02) |(0.02) |

| | | | | | | | | | | |

|Wallace |- |- |- |2.10 |- |- |- |- |- |- |

| |- |- |- |(2.07) |- |- |- |- |- |- |

|Anderson |- |- |- |- |- |- |3.73 |- |- |- |

| |- |- |- |- |- |- |(5.07) |- |- |- |

|Perot |- |- |- |- |- |- |- |- |- |-2.80 |

| |- |- |- |- |- |- |- |- |- |(2.69) |

|Log Likelihood | | | | | | | | | | |

| |-72.35 |-29.30 |-94.17 |-34.41 |-82.05 |-44.65 |-39.23 |-15.17 |-31.14 |-69.69 |

| | | | | | | | | | | |

|Pseudo R2 |0.14 |0.17 |0.12 |0.15 |0.05 |0.16 |0.27 |0.32 |0.09 |0.25 |

| | | | | | | | | | | |

|N |165 |175 |256 |168 |163 |237 |127 |158 |150 |223 |

Standard Errors in Parentheses

Table 3. Probability of Presidential-Party Incumbent Victories At Midterms

(Maximum Likelihood Probit Coefficient Estimates)

|Variable |1958 |1962 |1966 |1970 |1974 |1978 |1982 |1986 |1990 |1994 |

|Constant |-2.33 |1.86 |2.42 |-6.47 |3.28 |2.02 |-2.08 |-0.24 |5.42 |1.20 |

| |(2.49) |(2.11) |(0.69) |(2.58) |(2.23) |(1.87) |(2.85) |(3.54) |(3.12) |(1.89) |

|Turnout |0.45 |-2.26 |-1.20 |0.40 |-1.68 |-3.57 |-3.08 |-4.48 |-3.14 |-0.48 |

| |(0.44) |(1.43) |(0.42) |(2.66) |(1.40) |(2.31) |(2.04) |(3.12) |(2.21) |(1.79) |

|Presidential Vote |5.18 |-1.26 |0.28 |9.42 |-1.98 |1.06 |6.84 |11.53 |-6.54 |5.76 |

| |(3.83) |(4.59) |(1.15) |(3.55) |(2.92) |(3.21) |(4.53) |(7.26) |(5.15) |(2.46) |

| | | | | | | | | | | |

|Experienced |-0.77 |-0.01 |-0.69 |-0.36 |-0.52 |-0.78 |-0.18 |-0.84 |-0.48 |-0.77 |

|Challenger |(0.24) |(0.40) |(0.22) |(0.36) |(0.23) |(0.30) |(0.36) |(0.55) |(0.40) |(0.26) |

| | | | | | | | | | | |

|Hi Pres Vt ( Votng |0.00 |- |-0.001 |0.03 |-0.00 |0.01 |0.02 |- |0.03 |-0.03 |

|Record |(0.01) |- |(0.01) |(0.02) |(0.01) |(0.01) |(0.02) |- |(0.02) |(0.02) |

|Lo Pres Vt ( Voting|-0.002 |0.01 |-0.01 |0.05 |-0.01 |0.00 |0.00 |-0.04 |0.01 |-0.03 |

|Record |(0.01) |(0.01) |(0.007) |(0.02) |(0.01) |(0.01) |(0.01) |(0.03) |(0.02) |(0.02) |

|Wallace |- |- |- |2.88 |- |- |- |- |- |- |

| |- |- |- |(2.07) |- |- |- |- |- |- |

|Anderson |- |- |- |- |- |- |2.33 |- |- | |

| |- |- |- |- |- |- |(5.25) |- |- |- |

|Perot |- |- |- |- |- |- |- |- |- |-3.17 |

| |- |- |- |- |- |- |- |- |- |(2.75) |

|Log Likelihood | | | | | | | | | | |

| |-72.26 |-27.07 |-89.29 |-32.68 |-81.60 |-43.85 |-38.12 |-14.35 |-28.69 |-69.22 |

|Pseudo R2 |0.14 |0.06 |0.17 |0.20 |0.05 |0.18 |0.29 |0.24 |0.16 |0.26 |

|N |165 |85 |256 |168 |163 |237 |127 |82 |150 |223 |

Standard Errors in Parentheses

In 1962 and 1986, there was no variation on the dependent variable in districts where the president’s share of the two-party vote exceeded the median value: i.e. all Representatives in those districts who sought re-election won. Hence, for those years, we report the effect of presidential support on the probability of victory only for members from districts where the president’s share of the two-party vote was below the median value.

Figure 1

First-Dimension W-Nominate Scores Johnson and House Democrats, 1965-66

Johnson’s score is represented by the vertical line.

Figure 2

First-Dimension W-Nominate Scores Clinton and House Democrats, 1993-94

Clinton’s score is represented by the vertical line.

-----------------------

† An earlier draft of this paper was delivered at the annual meetings of the Southern Political Science Association, Atlanta, Georgia, October 28-31, 1998. Thanks to participants in the Merriam Lab seminar and to James Cox and John Hickman for helpful comments. All of the (doubtless numerous) errors that remain can be attributed to the authors, jointly or randomly between authors, as the attributer prefers.

[1] Some of the difference, of course, lies in the slow transformation of the South between the 1960s and 1990s, as a number of conservative Democrats were replaced by Republicans.

[2] We do not attempt to analyze retirement decisions. Nor do we examine the relatively rare cases of primary-election defeat, although these might conceivably exhibit similar features.

[3] Kernell (1977) argues that voters with negative evaluations of the president are more likely to turnout to vote, suggesting they are more likely to oppose those MCs who are seen as supporters of the president.

[4] We lose a few cases because of missing data. Presidential support, for example, is usually missing for the speaker. Turnout is missing for some cases of uncontested races where state law does not require tabulating votes of the winner.

[5] Note that this coefficient (Low Pres Vote*Voting Record) is significant at the 0.10 level.

-----------------------

[pic]

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download