Who needs polls?

Public opinion polls have been back and forth over the summer and into the fall in regard to voter preferences in this year’s midterm elections. Early in the summer, Republicans led Democrats in the national generic vote for Congress, but by September the same polls gave Democrats a slight edge, according to a running average of polls collected by Real Clear Politics. The polls then shifted again in the first week of October as Republicans accumulated a lead in the generic ballot, leading some to forecast a “red wave” in November’s congressional elections. As of October 22, roughly two weeks in advance of the election, Republicans led Democrats 48 to 45 percent in the Real Clear Politics average of thirteen separate polls taken over the previous two weeks. Yet that three-point advantage was still within the margin of error of most of these polls, and not yet large enough to forecast a Republican sweep.

In looking at these polls, it is not altogether clear if they are picking up a newly developing trend in favor of Republicans, or if the pollsters merely revamped their polling techniques to measure a Republican advantage that was there all along. Many observers have a sneaking suspicion that some pollsters rig their surveys to favor one or the other political party in the hope of shaping the eventual election outcome. That might be accomplished by publishing results that discourage some candidates while encouraging others such that candidates with the lead can appeal to donors for funds and newspapers can publish articles with the narrative that one party is destined to come out on top.

There has been a clear pattern in recent elections for some polls to overstate the strength of Democratic candidates in pre-election forecasts. For example, most polls forecasted an eight- or ten-point victory for Biden over Trump in the 2020 presidential election, though Biden’s margin in the actual election was just four points. One prominent pre-election poll had Biden winning in Wisconsin by seventeen points when the actual margin was less than one point—a result that is unlikely to have happened by chance. In the 2020 election for the Senate in Maine, pre-election polls gave the Democratic candidate a lead of between five and ten points when in fact Republican Susan Collins won the election comfortably by nine points. In that case, the pre-election polls were off by as much as fifteen or twenty points. Those examples might be multiplied many times over in recent election cycles. In view of those systematic errors, many voters, especially Republican voters, are understandably skeptical these days about the results published by major polling organizations.

Are these pollsters trying to manipulate election outcomes? Are they telling subscribers and clients what they want to hear? Or are their polling techniques flawed so that they consistently err in one direction? Are voters lying to the pollsters or trying to mislead them about their real preferences? It is difficult to answer these questions with any certainty. The one thing we know is that the polls are often wrong.

Fortunately, there are alternatives to the polls when it comes to forecasting overall election results.

Alternatives to the polls

In recent decades, several election analysts and statisticians have tried to quantify the key economic factors, such as inflation, unemployment, and economic growth, that shape campaigns and motivate swing voters to support one of the two major parties. Those metrics are easy to measure and similarly easy to use in forecasting election results.

Ray C. Fair, Professor of Economics at Yale, has developed simple models for forecasting presidential and congressional elections based upon such metrics, and has tested them against election results going back to 1918. In his view, the performance of the economy is the main factor driving election outcomes; the rest of it—personalities, polls, press and television coverage, etc., is mostly “noise” obscuring the main causes. His equations are generally more accurate than the polls in forecasting election results and are easier and far less expensive to use.

The main limitation of this method is that it only forecasts the national results and does not reach down into the states or individual districts to predict which ones will flip from one party to the other. In addition, it cannot be easily adapted to the elections for the U.S. Senate. Nor does it account for non-economic issues such war, corruption, or crime.

Professor Fair’s midterm election model (he has said as reported by The Guardian) is “based on estimations back to 1918, 100-plus years of data. In that period what seems to matter, election after election, is inflation, output, growth, and the penalty you get for being the incumbent party in the White House.”

His model forecasts the Democratic share of the national two-party vote based upon two economic metrics, the rate of inflation and the growth in real per capita income over the seven quarters prior to the election (calculated as the number of quarters in which real per capita income grew by more than 3.2 percent). He also incorporates a penalty of around 3 percent of the vote for the party holding the White House due to the long-term pattern by which that party loses seats in the House of Representatives in mid-term elections.

Forecasts for the House of Representatives

According to data compiled by the Federal Reserve Bank of St. Louis and the Bureau of Economic Analysis, inflation grew by 6.1 percent in 2021 and 8.8 percent in the first half of 2021. Recent estimates suggest that inflation advanced by 8.3 percent in the third quarter of 2022, which leaves us with an average quarterly increase in inflation of 7.2 percent over the seven quarters of the Biden administration. In addition, there have been three quarters—all in 2021—in which real per capita GDP grew by more than 3.2 percent. Economic growth has been in negative territory for most of 2022. Both the inflation and growth numbers are unimpressive and probably spell trouble for Democrats in the midterm elections.

These figures, when incorporated into Professor Fair’s equations, suggest that Democrats will win 46.7 percent of the two-party vote for Congress with Republicans gaining the remaining 53.3 percent of the vote—a large margin for Republicans. These figures should be adjusted slightly for the roughly 3 percent of the vote usually cast for third parties, yielding a prediction of the generic vote of 52.1 percent for Republican candidates and 44.9 percent for Democrats, or a 7.2 percent margin for Republicans.

This forecast is more than double the 3 percent margin in the generic vote recently reported in the Real Clear average of polls (48 percent to 45 percent), and sets up an interesting test as to which method is more accurate—the polls or Professor Fair’s model.

The question arises as to how many seats Republicans are likely to pick up this year, assuming a seven-point victory in the national generic vote. That calculation is complicated due to the growing number of safe seats the parties have created for themselves through re-districting and the diminishing number of competitive districts that might swing toward Republicans in a “wave” election. In addition, the number of seats a party picks up or loses in the preceding presidential election will influence the gains or losses in the succeeding midterm election. The presidential party usually picks up seats in the presidential election but then loses some or all of them in the succeeding midterm election. As it happened, Republicans gained thirteen seats in the House of Representatives in the 2020 election, an unexpected result in view of the presidential outcome, bringing their total to 212 seats (as against 223 seats for Democrats). Those surprising pickups for Republicans in 2020 will reduce the number of seats they stand to gain in this year’s elections.

There have been two occasions since the 1990s in which Republicans led the generic congressional vote by the margin by which they are forecast to win it this year: in 1994, when Republicans won by a margin of 6.8 percent and picked up fifty-four seats; and in 2010 when they won the generic vote by the same margin and picked up sixty-three seats. In both cases, Republicans gained majorities in the House by virtue of those landslides. On the other hand, Republicans won the generic vote in 2014 by 5.7 percent, but gained just thirteen seats, in large part because they had already picked up sixty-three seats in 2010 and managed to hold on to them in the 2012 election. There were fewer vulnerable Democrats to take out.

In 1994 and 2010, Republicans went into the elections holding just 176 seats (1994) and 179 seats (2010), compared to the 212 seats they currently hold. In those earlier elections, Republicans had more opportunities to gain seats than is the case today. Their victories in 1994 and 2010 brought them to 230 and 242 seats (respectively) in the House, as against 218 required for a majority. The Republican high-water mark in the post-war era stands at 247 seats, the number held after the 2014 elections. They may surpass that number this year, but probably not by much because they have already picked off many of those vulnerable Democratic seats.

The forecast then, based upon Professor Fair’s model, is that Republicans will win the generic congressional vote by seven points and will pick up around forty seats in the House, giving them a gain of some fifty-three seats (when combined with the 2020 results) over the two election cycles, comparable to their gains in 1994 and 2010. This will increase the Republican total in the House of Representatives to around 250 seats, surpassing the 247-seat record for the entire period dating back to the 1930s. This would give them a margin of roughly sixty-five votes over Democrats in next year’s session of Congress, a much more impressive majority than Democrats have held over the current term.

Implications for Senate Elections

The economic model does not forecast the outcomes for Senate elections, and Professor Fair does not attempt to extend it to those contests. For one thing, those races are shaped by the personalities of candidates as much as by underlying economic conditions. There are also fewer Senate seats in play, usually between thirty-three and thirty-six Senate seats contested in each cycle as against 435 House seats.

Much depends also on the nature of the election cycle, and how many seats each party must defend in a particular year. In 2022, there are thirty-four seats up for election, but twenty-one are held by Republicans and thirteen by Democrats, leaving Democrats with fewer seats at risk. By comparison, in 1994 the situation was reversed, with Democrats defending twenty-two seats and Republicans just thirteen. In that election, Republicans (with a 6 percent margin in the generic vote) picked up eight Senate seats, or close to 40 percent of the Democratic seats at risk. In 2010, the balance of seats at risk was nearly even, with nineteen Democratic and eighteen Republican seats contested. Republicans won the generic vote by almost three points and won six more seats in the Senate out of the nineteen Democratic seats contested, or just under one-third of those seats. With just thirteen Democratic seats in play this year, Republicans are unlikely to pick up as many seats this time around as they did in 1994 or 2010.

Yet, if this pattern holds for this year’s Senate elections, then Republicans might pick up as many as a third of the twelve Democratic seats contested—or a gain of three or perhaps four seats. This forecast once again goes against the forecasts based upon polls which until just a short time ago gave Democrats a good chance of picking up a seat or two in the Senate and holding on to their majority. A Republican gain of as many as four seats would necessarily involve a few surprising victories in states currently thought to be safely in Democratic hands. It is not difficult to foresee how Republicans might pick up three seats, one each in Nevada, Arizona, and Georgia, but it is hard to tell where that fourth pick-up might come from in view of current polling. Perhaps in Washington state or New Hampshire or Colorado—though current polls do not point toward Republican victories. Nevertheless, that is the rationale for the model: it points to results not clearly visible in the polling.

Overall forecast for the House and Senate based upon economic metrics

This, then, is the forecast, relying upon current economic figures and ignoring the polls altogether. Republicans will win the generic congressional ballot by seven percentage points, pick up around forty seats in the House of Representatives, and perhaps three or four in the Senate, giving them solid majorities in both houses of Congress—and setting up a fierce clash between Congress and the Biden administration over the next two years.