The 1998 Midterm That Went the Wrong Way
Republicans went into November 1998 hoping for as many as forty House seats. The president’s party gained five, and the late polling had already stopped pointing the other way.
Published 06/30/2026
Democrats gained a net of five seats in the United States House of Representatives on 11/03/1998, which made the 1998 midterm the first midterm since 1934 in which the party holding the White House added House seats. The gain came while the House Judiciary Committee was running an impeachment inquiry into President Bill Clinton, against an expectation of Republican gains that was close to unanimous in Washington. Within days the Speaker of the House had given up the job.
The standard account calls 1998 a backlash against impeachment. The evidence points one step earlier, at the expectation itself. The forecast of large Republican gains was built on a historical base rate applied without testing whether its conditions held, and four of those conditions had already been measured and published before anyone voted. The surveys measured that race about as well as surveys measure anything. The errors sat on top of them.
That claim is falsifiable in a specific way. If the late public polling had shown a Republican lead of the size the forecasts implied, or if Bill Clinton’s approval had been weak, or if voters had told interviewers the impeachment fight was what they were voting about, the consensus would have been a reasonable read of available information. None of those three things was true in the last two weeks of October 1998.
What happened in the 1998 midterm
Republicans held the US House in 1998 with 223 seats against 211 Democrats and one independent, down from 226 Republicans and 207 Democrats at the opening of the previous Congress. Those totals come from the Office of the Historian of the US House. Republicans also won the national House popular vote by 1.1 points, 48.4 percent to 47.3 percent, and still lost ground in seats. The certified vote totals are 32,237,964 Republican and 31,490,298 Democratic.
House seats before and after the 1998 election
Party divisions at the opening of the 105th and 106th Congresses, as recorded by the Office of the Historian of the US House.
The Senate finished exactly where it started, at 55 Republicans and 45 Democrats, with three seats changing hands in each direction. Democrats took New York, where Chuck Schumer beat Alfonse D’Amato by 10.5 points, North Carolina, where John Edwards unseated Lauch Faircloth, and the open Indiana seat, won by Evan Bayh. Republicans took the open Kentucky and Ohio seats and defeated Carol Moseley Braun in Illinois. Each of those six races is recorded in the 1998 Senate returns. The Senate Historical Office records 55 Republicans and 45 Democrats in both the 105th and 106th Congresses.
Why the 1998 forecasts pointed the wrong way
The expectation of Republican gains in the 1998 midterm rested on a historical regularity and on a theory about scandal, and by late October the available polling supported neither. Republicans went into the election with hopes of picking up as many as 40 congressional seats. Christianity Today put the figure at as many as 40 seats in its December 1998 account. Since the Second World War the party holding the White House had lost an average of 27 House seats in comparable midterms. CBS News reported that average on 11/04/1998. Both of those are real facts. Neither is a measurement of the 1998 electorate.
The measurements were published and they were close to the result. Gallup, writing on 10/16/1998, had the generic congressional ballot at 47 percent Democratic to 46 percent Republican among likely voters and called the outlook unclear, adding that Republicans would need a majority of the national two-party vote to hold or expand their House seats. That analysis is on the record at Gallup. Pew Research Center’s final pre-election survey, fielded 10/28 to 10/31/1998 among 1,714 registered voters including 743 likely voters, found Democrats at 46 percent and Republicans at 44, after a Republican lead of 48 to 43 in mid-October. Pew published those numbers on 11/01/1998
Late generic ballot polling against the certified House vote
Two public polls fielded in the final month of the 1998 campaign, next to the national two-party House result.
Two inferences did the damage. The first was arithmetic dressed as prophecy. An average across a dozen midterms held under different conditions is a prior, and a prior is supposed to be updated when a measurement arrives. The 1998 measurements arrived, showed a tie, and were discounted because the pattern said otherwise.
The second inference was about behavior. The theory held that the Lewinsky investigation would raise Republican turnout and depress Democratic turnout. That is a testable claim about composition, and the same Pew survey carried evidence against it: 54 percent of voters took an unfavorable view of members of Congress who had voted for the impeachment inquiry, up from 49 percent in mid-October, and only 35 percent called the congressional impeachment debate very important to the nation, against 88 percent for education. Those figures are in the same Pew release
Four conditions the 1998 forecast needed
Each of these was measured and published before election day, in Pew Research Center’s final pre-election survey of 1,714 registered voters fielded 10/28 to 10/31/1998.
Those four readings are the whole case. A base rate earns its authority from the conditions that generated it: midterms in which a president’s party is punished are usually midterms in which the president is unpopular. In the autumn of 1998 the president’s approval stood at 65 percent and the inquiry into him was less popular than he was. The instrument that would have caught the error existed, was in the field, and reported on 11/01/1998, two days before the vote.
Three pictures of the same election
Net change in the president’s party’s House seats: what Republicans hoped for, what the postwar pattern implied, and what the count produced.
| Measure | Before the 1998 election | Result |
|---|---|---|
| Republican House seat gain | Hoped for as many as 40 | Lost 4, finishing at 223 |
| President’s party, House seats | Postwar average loss of 27 | Gained 5, finishing at 211 |
| Generic congressional ballot | Gallup 47 D to 46 R; Pew 46 D to 44 R | House vote 48.4 R to 47.3 D |
| US Senate | 55 R to 45 D | 55 R to 45 D, three seats each way |
| Bill Clinton job approval | 65 percent in Pew’s final poll | Impeached 12/19/1998, acquitted 02/12/1999 |
What voter turnout did in 1998
Turnout in the 1998 midterm was 73,117,022 votes, or 36.4 percent of the voting-age population, according to the US Election Assistance Commission. The Commission’s 1998 registration and turnout table carries both figures. That is a low number, and the conventional reading at the time was that low turnout helps Republicans. Gallup had said as much in October. The aggregate went the way the conventional reading expected. The composition did not.
A get-out-the-vote campaign organized by the Black Leadership Forum targeted 10 states in 1998, running door-to-door canvassing, rallies and radio spots. The Joint Center for Political and Economic Studies documented the program and the state-level shares in FOCUS, December 1998. Black voters made up 10 percent of national turnout and voted 88 percent Democratic. In Georgia they were 29 percent of the voters at the polls, above their 27 percent share of the state population, and Roy Barnes won the governorship with 37 percent of the white vote and 90 percent of the Black vote while Guy Millner took 58 percent of the white vote. In South Carolina, where Black voters were 25 percent of turnout, Jim Hodges beat the incumbent governor David Beasley with 40 percent of the white vote and 92 percent of the Black vote. In the Senate races there, John Edwards carried North Carolina with the support of nine out of ten Black voters and Ernest Hollings held South Carolina.
Black share of election-day turnout in 1998
Voter News Service exit poll shares reported by the Joint Center for Political and Economic Studies, in two Southern states and nationally.
An aggregate turnout rate is a different quantity from a composition, and only one of them decides seats. Two electorates of identical size can produce opposite results if the shares inside them differ. The 1998 forecasts used a national rate as a stand-in for composition in a set of states where composition was being deliberately changed, which is the part of the error that a campaign can actually do something about.
What the 1998 result cost the party that misread it
Newt Gingrich gave up the speakership within days of the 1998 election. Christianity Today reported at the time that the poor showing and Republican grumbling led him to resign the post. Five seats was the smaller cost. The larger one was a closing month of money, travel and message built around a picture of the electorate that no published measurement supported, with nothing else ready when it failed. Christian Coalition exit polling put religious conservatives’ support for Republicans at 54 percent in 1998, down from 66 percent in 1994, and the group’s president said afterward that the party had gone to the voters without an agenda. Christianity Today reported both figures
The impeachment itself proceeded. The House adopted two articles on 12/19/1998, the first by 228 to 206, The roll call is archived at Voteview and the Senate acquitted President Clinton on 02/12/1999. The Senate’s proceedings that day are in the Congressional Record. The party that had read the electorate wrongly in the autumn still had to govern through the winter with a majority of five.
What a campaign can take from the 1998 midterm
Three things in 1998 are portable to any race on the ballot. The first is that a base rate is a starting point and loses to a measurement of the race in front of you, every time, as long as the measurement is honest about its own error. The second is that a turnout assumption should be written down as composition, by geography and by group, in numbers specific enough that somebody can later point at the line that was wrong. A single projected turnout rate cannot be audited, and an assumption that cannot be audited will quietly carry a strategy.
The third is about moving targets. Where an organized mobilization program is running on the other side, the electorate you measured in September is being rebuilt in October. That is a reason to measure late and to measure composition rather than only the horse race, because the horse race moves last.
Frequently asked questions
Who won the 1998 midterm election?
Republicans won the 1998 midterm in the sense that they kept control of both chambers of Congress, finishing with 223 US House seats and 55 US Senate seats. Democrats gained a net of five House seats, which was the first midterm gain for the party holding the White House since 1934, and the Senate was unchanged at 55 Republicans to 45 Democrats.
Why did Democrats gain House seats in 1998 during the Clinton impeachment inquiry?
Public opinion in 1998 ran against the impeachment inquiry rather than against President Bill Clinton. Pew Research Center's final pre-election survey put Clinton's job approval at 65 percent and found 54 percent of voters viewing members of Congress who backed the inquiry unfavorably. At the same time, a get-out-the-vote program targeting 10 states raised the Black share of turnout well above its national level in several Southern contests.
What was voter turnout in the 1998 midterm election?
Turnout in the 1998 US midterm was 73,117,022 votes, which the US Election Assistance Commission reports as 36.4 percent of the voting-age population. The national rate was low, and the composition of the electorate in several Southern states shifted at the same time, which is why the low aggregate did not produce the Republican gains that were expected.
Did the polls get the 1998 midterm wrong?
The late public polls in 1998 were close to the result. Gallup had the generic congressional ballot at 47 percent Democratic to 46 percent Republican in mid-October, Pew Research Center had it at 46 to 44 Democratic in the final week, and the certified national House vote was 48.4 percent Republican to 47.3 percent Democratic. The forecasts of large Republican gains came from historical patterns and turnout theories rather than from that polling.
More from the blog
We help candidates win races they normally shouldn’t. A turnout assumption is a decision, and in 1998 it was the decision that settled the outcome. Our Modeling work writes the assumed composition down by precinct and by group before a field plan is built on it, and our Insights work measures late enough to catch a mobilization program that is still moving. Tell us which race you are working and we will come back with a written scope.