What Is a Wave Election, Anyway?
Democrats look poised to have a good showing on Tuesday. They are almost certain to gain seats in the House, and may break even in a Senate cycle where they have to defend 10 seats in states Donald Trump won. They’ll also add a number of governorships to their tally, along with a slew of state legislative seats.
But will it be a wave election? This concept is something that analysts refer to routinely, but there’s no generally accepted definition of it. We know that it is a really bad election for one party, but sometimes the lines become difficult to trace. Everyone agrees, for example, that 2010 and 1938 represent wave elections, however that is defined. But what about years like 2006 or 1982, where the shifts were less dramatic, or even internally contradictory (in 1982, Republicans gained Senate seats despite losses in the House).
In a way, the concept of a wave is irrelevant. Years like 1954, where Democrats barely gained seats but still flipped the House, are probably more consequential than a major wave year like 1922, where Democrats gained over 70 seats but failed to capture the chamber. Whether 1982 is a wave election is irrelevant, as Democrats gained enough House seats to stop Ronald Reagan’s domestic agenda.
Yet waves still capture our imagination, in part because we tend to interpret them as nullifying a presidential mandate. So the 1994 GOP wave forced Bill Clinton to declare that the era of big government was over (however temporarily) while the wave of 1966 stalled the Great Society. In a presidential year, waves can be seen as inaugurating new eras, such as Barack Obama’s win in 2008 or Ronald Reagan’s in 1980.
So, we start with a definition: A wave election represents a sharp, unusually large shift in the national balance of power, across multiple levels of government. To measure this, I turn to a metric devised by David Byler and me, which measures the power that a party has in the government at a particular point in time. It’s a combination of the party’s share of the presidential popular vote, Electoral College, the House popular vote and seats in Congress, the makeup of the Senate, the makeup of governorships and the makeup of state legislatures.
What I’ve done is tracked the shift in this index going back to 1856. In other words, I’ve tracked how the index moves from election to election. The idea is that in wave elections, you should get unusually large swings in partisan power across multiple levels of government, and hence large swings in the index. As a technical side note, the average shift is 0.44, with 33 percent of the observations within one standard deviation of the mean and 5 percent of the observations within two standard deviations. This is consistent with the frequently expressed view among political scientists that elections are fundamentally random and don’t favor one party or the other over time.
In any event, the question then becomes: How do we decide whether a shift is a “large” shift? A common statistical tool is the standard deviation, which is basically a measure of how spread out a data set is. For example, the average American male is about 70 inches tall (5 feet 10 inches). But that doesn’t tell us that much about whether someone is unusually tall or short. It only tells us whether someone is above average or below average. The standard deviation helps us with that calculation. One standard deviation from the average can be thought of as unusually tall, and two standard deviations from the average can be thought of as very unusually tall.
As it turns out, the standard deviation for height is three inches. So, using our rule of thumb for standard deviations, we might say: A man who is 6-1 is tall, and a man who is 6-4 is unusually tall, while a man who is 6-7 is extremely tall. On the other side, 5-7 is short, 5-4 is very short, and 5-1 is extremely short. We can quibble about the cutoffs, but this seems about right.
Applying that rule of thumb to our data, there is only one election that is three standard deviations out from the average: 1932. Indeed, this is the granddaddy of all elections, where a deep, durable Republican majority was swept away.
Loosening our requirement to include elections that are two standard deviations out, we add three others as waves: 1860, 1894, and 1874. These represent two canonical realigning elections, plus one that probably should be in that bucket (1874). They also represent extreme events in American history: The 1860 election saw the demolition of the Democratic Party outside the South on the eve of the Civil War. The 1874 election saw the rebirth of the Democratic Party, which gained 94 House seats at the onset of the so-called Long Depression. The 1894 election saw Democrats lose an astonishing 127 House seats, as the aftermath of the Panic of 1893 continued to roil the nation. All of these elections had ramifications far down the ballot as well.
We might be tempted to stop here, and say that a wave election should just be an election that is two standard deviations out from the average (the equivalent of a 6-4 male) and that our discourse to the contrary is misplaced. But this doesn’t really get us to what most observers are really talking about with wave elections. After all, according to this approach we haven’t had a wave election since 1932, and none of us talks about wave elections that way.
So perhaps we should go with 1.5 standard deviations. This is the equivalent of making our cutoff for a very tall man be about 6-2 ½. If we do this, we don’t seem to lose anything in explanatory power, as most of what is added falls cleanly in our definition of a wave: We add 1920, 1938, 1994, 1922, 1912 and 2010. We’ve added only elections that people talk about as waves, without adding anything that isn’t a wave.
If we take the next step of one standard deviation from our average, we add a lot of elections. The first few are pretty clearly wave elections, but after that it gets iffy. We add (in descending order): 1958, 1948, 1966, 1974, 1930, 1870, 1890, 1910, 1980, 1856, 1968, 1942, 1882, 1964, 1862, 1954, and 1946. I’m content to leave the definition at 1.5 standard deviations from our average, but reasonable minds can vary: We might want to include midterm elections just outside of that range since presidential elections can vary a bit more than midterms do.
In any event, note that two widely discussed wave elections, 2006 and 1982, do not fall within our definition, nor does an arguable wave election, 2014. In fact, all three just miss a one standard deviation cutoff. In this case, we might just say that the commentariat got it wrong (especially 1982, where half of Republicans’ losses can be chalked up to redistricting), unless people are also willing to accept 1942 and 1954 as wave elections.
I might say, however, that this is only part of the story. There might also be elections that don’t shift things much, but that maintain an unusually strong showing for a party. For example, 1934 did not shift the national balance of power much, but it was still an unusual outcome, historically speaking for Democrats. For this, we turn back to the initial index. Rather than measuring the change in the index, let’s look at elections that result in unusually strong showings for parties, regardless of what their previous showing was.
If we do this, 1936, 1866, 1934, 1868, and 1864 show up as two-standard-deviation wave elections. This makes sense, as the 1860s were a time of generally significant Republican strength, while the early 1930s were a time of generally significant Democratic strength. Expanding to 1.5 standard deviations would cover a few wave elections above, but also add 1964, 1976, and 1862 as “maintaining” elections. Once again, at a single deviation, we add some things that seem to fit our “common” talk, but also exclude a bunch of elections that don’t seem to fit: 1872, 1974, 1958, 1870, 1960, 1912, 1978, 1904, 1940, 1962, 1928 and 1992.
Anyway, using both definitions, the 1.5 standard deviation cutoff seems to best encapsulate what we mean when we talk about a wave election, and if we add some wiggle room to that we’ve done a really good job – because it suggests that our public dialogue about elections is really on to something. The only elections I think are clearly missing are the 2006/2008 pairing, which can really be seen as a two-cycle wave election.
Regardless, this is the definition I will be using for the next week.