How Unemployment Affects Midterm Elections

How Unemployment Affects Midterm Elections

By Sean Trende - November 25, 2009

It’s become something of an article of faith within the lay punditry that high unemployment and a weak recovery will seriously weigh down the Democrats’ performance in the 2010 midterm elections. People vote with their pocketbooks, the adage goes, and their pocketbooks are likely to be awfully thin in November of 2010.

On the other hand, pundits with technical backgrounds have noted that predictive modeling shows a weak correlation at best between almost every measure of the performance of the economy and midterm gain. Seth Masket, a professor of political science at University of Denver, notes that neither the unemployment rate nor the change in the unemployment rate (in other words, how much unemployment has increased or decreased from the previous year) correlate with seat loss. Masket does note a statistically significant relationship between real disposable income growth and midterm loss, but observes that even this relationship is weak.



Karl Rove has approached this from a slightly different perspective. He notes that neither the April unemployment rate nor the September unemployment rate correlates heavily with seats lost. Rove also looks at GDP growth for the second and third quarters of each midterm election year, and finds only a weak correlation there as well.

Common sense nevertheless tells us that when the economy is pinching people, they’ll take it out on the party controlling the levers of government. One can argue, then, that we’re really seeing are the limitations of regression analysis more so than a weakness in the relationship between midterm elections and economic performance. Let me illustrate what I mean using unemployment as an example (though the same argument could be applied to various economic data).

First, consider the fact that we have only had 15 midterm elections since 1950. This is barely data; it’s more of a good collection of anecdotes. This is compounded by the fact that if a relationship between unemployment and midterm election results exists, it is almost certainly exponential. In other words, below a certain point, voters probably don’t delineate that much among different unemployment rates. But once you cross a certain threshold, people start to worry increasingly about each point increase in unemployment. For example, I’m guessing if unemployment jumps from 4% to 8%, people get concerned. But when it goes from 8% to 12%, people start to freak the heck out (for what it’s worth, this is also a technical problem for most regressions you will read about, which generally assume variables have linear relationships to one another).

This severely limits our analysis in other ways.  As it happens, we don’t have any observations where unemployment is above 10 percent (which seems at least possible in November 2010), and have only one observation where unemployment is above seven percent.  In ten of these fifteen midterms, the unemployment rate is between 3.8 and 5.8 percent. In other words, almost all of our midterm elections took place at unemployment levels where people probably aren’t distinguishing much between different employment rates. A regression analysis isn’t going to reveal any relationship there, even if some type of relationship becomes evident at higher rates of unemployment.  We just don't have the data to say "yes" or "no."

The one "high unemployment" midterm we've experienced since 1950 suggests that high and rising unemployment hurts the party in power significantly.  In 1982, unemployment was at 9.6 percent and rising. The Republicans managed to “only” lose 26 seats. But they only had 192 seats to begin with; this represented 14 percent of their caucus. If the Democrats lose 14 percent of their caucus in 2010, they will lose 35 seats.

And at the end of the day, there are an awful lot of things that affect an election. Is the President pursuing an unpopular war and controversial policies at home (1966, 2006)? Then it probably doesn’t matter that the economy is blazing ahead. Is the President waging a successful war and getting ready to take out a longtime nemesis (2002)? The public is going to be more forgiving of the sluggish growth in real disposable income and rising unemployment. The bottom line is that every election becomes something of an explainable, unique event – in other words, they’re almost all outliers.

The truth is, you’re not going to find a particularly good variable to explain the influence (or lack thereof) of perceptions of the economy on peoples’ voting choices. Just because it’s hard to define this relationship, however, doesn’t mean the relationship doesn’t exist. Common sense, and a more qualitative look at the data imply a correlation.

Consider the worst midterm drubbings since we started to compile good economic data (around the Great Depression), defined as those elections where the President’s party lost more than 10 percent of its seats. Those years are: 1974 (25% seat loss), 1958 (24%), 1946 (22%), 1938 (22%), 1994 (21%), 1930 (19%), 1942 (17%), 1966 (16%), 1982 (14%), and 2006 (13%). The bolded elections are those that immediately followed a recession, or occurred within a recession ("immeditately followed" is defined as those midterms held after a recession that ended in the third quarter of the year before the election or later). Most really bad midterm election years do, in fact, follow recessions or depressions.

But does the inverse relationship exist? Are bad recessions inevitably followed by rough midterm elections? Look at the data from a slightly different angle. If we examine recessions since 1929 from most severe to least severe (in terms of GDP decline; we see a similar relationship with peak unemployment), we get Aug1929-Mar1933, Feb1945-Oct1945, the present recession, May1937-June1938, Nov1973-Mar1975, Aug1957-Apr1958, and July1981-Nov1982. Every substantial recession here results in the President’s party losing 10 percent of its seats or more. If we follow the dataset further, we find two recessions that weren’t followed by rough midterm elections for the President’s party: the Recession of 1953-54 and the early 2000s recession. But these were relatively mild recessions; it is understandable why they may not have had the same results for the President’s party.

The economic data become more problematic before the 1930s. Nevertheless, the great contractions in American history that also occur close to midterm elections – 1920-1921, 1893-94, 1839-1843, 1873-1879, 1913-1914, 1857-58 – all of them involve disastrous results for the party in power (respectively, -25%, -57% (not a typo), -49%, -48%, -21%, -37%).

In other words, a bad recession occurring close to a midterm election isn’t a necessary condition for a disastrous midterm election, but it seems to be sufficient. Because of the problems described above, regression analysis is going to have a hard time picking up on this, especially since we don’t have a good, all-encompassing variable for “the public perceives a bad economy.” But the relationship is pretty obvious nevertheless.

To be fair, there are reasons why this year’s Democrats might avoid blame for unemployment this time. The most obvious reason is that this recession didn’t begin on Barack Obama’s watch. Even if unemployment still is marching upwards, it started going up on his predecessor’s watch, and his party will probably be given at least some credit for this. Additionally, if unemployment begins to decline precipitously – which is possible – people may perceive that the economy is improving. But the latter appears increasingly unlikely, while the former argument is weakened by polls showing the public increasingly places the blame for the state of the economy on the Democrats.

The bottom line is that I don’t think Democrats can find much solace in the lack of a statistically significant relationship between various measures of the health of the economy and midterm election results. If voters think the economy stinks in 2010, common sense and a review of the historical record point toward substantial seat losses for the Democrats. Combined with a President whose numbers are drifting below 50%, and a Democratic party that is dramatically overextended into red territory, these factors increasingly point toward a Republican wave of some magnitude in 2010.

Sean Trende is senior elections analyst for RealClearPolitics. He is a co-author of the 2014 Almanac of American Politics and author of The Lost Majority. He can be reached at Follow him on Twitter @SeanTrende.

Why We Can't (Or Won't) Govern
Robert Samuelson · November 17, 2014
Marriage Is Pro-Growth
Larry Kudlow · November 15, 2014
Obama Touts Economic News as He, GOP Leaders Meet
Alexis Simendinger · November 8, 2014

Sean Trende

Author Archive

Follow Real Clear Politics

Latest On Twitter