The 2002 House Seat Swing in Historical Perspective

Bernard I. Tamas

Williams College

With the Republican Party holding a slim majority in the House of Representatives, even a small seat swing in 2002 can change the dynamics of power in Congress.  Party control over this institution may change for only the second time since 1954.  It is certainly conceivable that the Democrats would win the 6 seats that they need to regain majority rule.  As students of congressional elections like to state, the party in the White House has lost seats in the House in all but two elections over the past 100 years: 1934 and 1998.  Moreover, since the Second World War, only three seat swings in midterm elections have been smaller than 6 seats: 1962, 1986 and 1998. 

And yet, even with a sluggish economy and numerous corporate scandals, a return to a Democratic majority in the House is not a sure thing.   By most accounts, this election appears too close to call.  The Gallup corporation has found House Democratic and Republican candidates in a dead heat.[1]  The MSNBC/Zogby International 10-State Tracking Poll showed that while President Bush continues to hold significant popularity, possibly partially because of a rally around the flag effect related to the prospects of war with Iraq, the calls for war are also making the public nervous in a way that might fuel anti-incumbent sentiments towards the Republicans.[2]  Conversely, while most people think that the economy is in a recession, any Democratic message about or blame of the Republicans for the poor economy appears to be drowned out by the constant discussion of war.[3]

The historical evidence paints a more optimistic picture for Democrats and negative prediction for Republicans, though with a high level of uncertainty.  While there are no signs of a large seat swing coming, the best guess is that the Democrats will regain control over the House of Representatives with a slim majority.

It’s Still Partly the Economy, Stupid:

Politically speaking, the September 11 terrorist attacks have likely diverted a significant Republican defeat in the 2002 congressional elections.  Just before these attacks, President Bush’s approval level was a weak 51%, according to Gallup’s survey measure, “Do you approve or disapprove of the way that President Bush is handling his job as president?”  His approval rating also appeared to be dropping steadily, and with the continued weakness of the economy and other problems one could certainly predict that those numbers would have continued to decline had the events of September 11^th not occurred.  But there was an attack, and his approval rating quickly skyrocketed to 90%.  It has since been in gradual decline but is still at a respectable 67%, according to the most recent measurement.[4]

However, presidential popularity is hardly the only influence on House seat swings.  Many factors can affect how many seats each party will gain or lose.  In probably the most important article about midterm seat swings, published a quarter century ago, Edward Tufte argued that two main factors influence the size of midterm seat swings: dissatisfaction with the president and anger about the economy.[5]  Tufte’s real point was to argue that midterm congressional elections were not simply a correction from the previous presidential election (the popular argument made with the surge-and-decline theory.)  Nonetheless, a modified and updated version of his equation makes it possible to take a first predictive stab at the upcoming seat swing.

Graphs 1.1 and 1.2 are Monte Carlo simulations derived from a simple OLS regression based on Tufte’s 1975 equation.[6]  The dependent variable is midterm seat loss by the president’s party from 1950 to 1998.  The independent variables are presidential popularity as measured by the Gallup corporation and change in per capita income over the previous year.  Both variables showed a statistically significant impact on House seat losses.  (Two other variables were also loaded into the equation in an attempt to reduce the size of the confidence intervals: Following a technique of Gary Jacobson,[7] “exposure” was estimated by measuring if the party had a greater percent of seats than their average over the previous three elections; the assumption is that if they have an inflated number of seats, many of those seats would be lost in the next election.  Similarly, in order to offset the impact of greater incumbency advantage, Gelman and King’s estimate of incumbency advantage was also made an independent variable.[8]  However, these two variables had no significant impact on the results.)

The graphs indicate how many seats the model estimates would be lost by the president’s party given varying levels of presidential approval and change in per capita income.  In the top graph, real change in per capita income is held constant at $53, the amount of change from August 2001 to August 2002 in 1996 dollars, and presidential approval is allowed to vary from 40% to 70%.  Conversely, in the bottom graph, presidential popularity is held constant at 67%, Bush’s most recent approval rating, and real change in per capita income (in 1996 dollars) over the previous year is allowed to vary from a decrease of $200 to an increase of $400.  The graphs indicate that both economic conditions and approval of the president influence the number of seats each party will gain or lose in the House of Representatives.

Moreover, if the assumptions behind these graphs are correct, history leans towards a Democratic takeover of the House of Representatives in 2002, though with a slim majority.  The model predicts the Democrats gaining 12 or 13 seats.  The critical question is therefore the degree to which these assumptions are correct.   One issue is the measurement of change in per capita income.  The change from August 2001 to August 2002 seems to be the current measure most consistent with the historical data; however, any change in months would produce a different prediction of Democratic gain (though, interestingly, the prediction is always a Democratic takeover.)  Similarly, the confidence intervals for this equation are wide, around 15 seats in each direction.  So, while the model predicts that the Democrats will gain 12 or 13 seats, there is considerable uncertainty about how close to 12 the actual seat swing will be. 

Patterns of Seat Swings:

But other historical information gives reasons to believe that 12 seats is more likely to be an overestimate than underestimate.  There are two other factors that must be taken into consideration.

First, over the past half century, the magnitude of seat swings has declined considerably.  In the 1950s, the average seat gain or loss by the Democratic or Republican Party was 24 seats.  By the 1980s, it was just under 17 seats, and in the 1990s it was 18.4 seats, a decline in average magnitude over this period by around one-quarter. 

Second, even these numbers are deceptive.  Over the past half century, seat swings have come in spikes.  In certain elections, one of the parties gains a large number of seats, defeating incumbents of varying seniority and producing turnovers in many of the open-seat races.  Then, there tends to be a correction, when many of the representatives that had just won their seats during this large swing are defeated.  So, in 1958, during a recession and just after the Russians launched Sputnik, the Democrats gained 49 seats.  Two years later, as Kennedy was winning the White House, the Republicans gained back 21 seats; most of that gain was made by defeating Democrats who were swept in during the 1958 election.  Similarly, during the ideologically divisive 1964 campaign, the Democrats gained 37 seats and then lost many of those seats in 1966.  In 1974, during the oil crisis and following Ford’s pardon of Nixon, the Democrats gained 48 seats; while they lost few of those seats in 1976, those new Democrats became the key victims during the Carter 1978 midterm election.  Then, after the Republican sweep of 1980, many of these new Republican representatives where defeated during the 1982 seat swing.  This pattern repeated itself in a significantly weaker manner in the 1990s.  After the Republican sweep in 1994, that party lost 9 of those seats in 1996.

The reasons for this pattern are open for debate.  Oppenheimer, Waterman and Stimson deal with it most directly with their exposure thesis.[9] They see it as produced both by an underlying equilibrium pulling the vote back towards a baseline and by vulnerable first-term incumbents attracting strong challengers.  For our purposes here, this means that the larger seat swings occur either during key elections, like 1994, when there is a spike, or during the election immediately following that spike, when there is a regression towards the previous level.  In other elections, the seat swing tends to be significantly smaller. 

This pattern has an important implication on the predictive model presented above: The model tends to underestimate the seat swing of these spike or regression years but overestimate the other years.  From a research perspective, this creates a question of how House seat swings could be better estimated, especially since exposure measured by the number of seats over the norm does not improve the estimation once the 1990s are added to the data.  But, for our current purposes, it means that the 12 or 13 seats predicted simply with presidential popularity and economic indicators is likely an overestimation. 

Conclusion:

But even with a likely overestimation, the evidence as a whole still leads to a tentative conclusion that the Democrats will regain control over the House.  Two pieces of historical evidence may be added to what has already been discussed.  (1) The odd 1998 Clinton-impeachment midterm aside, when the party in the White House actually gained seats, there has been only two midterm elections when the president’s party lost less than 6 seats: The Democrats lost 4 seats in 1962 and the Republicans lost 5 seats in 1986.  Both elections occurred during economic expansions and with a popular president in office.  (2) Even with the decline in magnitude of seat swings, and even if one removes 1994, the average seat swing in the 1990s was 10 seats.  While the 1992 numbers are inflated because of redistricting and the Democrats gained more seats in 1996 because of vulnerable first-term Republicans from the Class of 1994 running for reelection, the smaller seat swings in recent elections may have been influenced by a strong economy through the 2000 election. 

In other words, while one must take into consideration that many factors influence House elections and can shift the seat gains in either direction, one would also have to give many allowances before reaching the prediction that the Republicans would lose less than a total of 6 seats, provided that nothing changes drastically in the second half of October.  While the modified and updated Tufte model is probably overestimating the size of the seat swing, it is also clear that high presidential approval is not likely to erase the impact of economic conditions on aggregate seat losses.  Similarly, while seat swings have gotten smaller over the last half century, they have not disappeared.  Seat swings over six are far more common than those of only a handful of seats; the small gains in the House by the Democrats in 2000 seem more the exception than the rule.  There is much predictive uncertainty, but the history of seat swings during midterm elections over the past half century lean towards the Democrats regaining control over the House of Representatives in 2002 with a razor thin majority.

Graphs 1.1 and 1.2.  Aggregate House Seat Losses by the President’s Party in Midterm Elections by Presidential Approval and Real Change in Per Capita over the Previous Year.