The wave definitely broke, but why?

Stu Rothenberg offers one definition of a "wave" election, which I think is pretty good:

"For me, the “political wave” metaphor evokes the image of a surging ocean wave that is much larger than normal and deposits debris that otherwise would not have made it ashore without the violent surf.

Politically, that translates into an election surge that is strong enough to sweep candidates who wouldn’t ordinarily win – because of the make-up of their districts or the limited funding of their campaigns, for example – to victory." -- Stu Rothenberg, Rothenberg Political Report

Let's agree on one thing: Republicans shellacked the Democrats last night, at least in the Senate. They gave them a "thumpin'". Swamped, dominated, crushed, emasculated, thrashed, blown out of the water: there are a lot of words for it, but at least an eight-seat pickup in the Senate, the loss of only one governor when they had at least six incumbents at risk, and the securing of what could very well be the biggest House majority the Republicans have ever had since the 1920s indicate a wave, or something close to it. 

Democrats wish the GOP wave last night were this small.

What definition of a "wave" should we use? A simple one is to simply look at the magnitude of gains by a single party. From this perspective it was very clearly a wave in the Senate--2010, which has generally been agreed to be a wave election, was only a six-seat pickup for the Republicans, or a loss of 31% of the Democrats' Senate seats up for election that year. This year, it looks like Democrats are losing 9 seats, or 42% of their Senate seats up for election this year

One interesting definition of a wave I'm tempted to use, apart from Rothenberg's pretty good one, is an election where the polls were biased in the direction of only one party. In this case, no Democrat in the Senate or gubernatorial races performed better than their average in the polls. Republicans in Virginia and New Hampshire kept races there much closer than they should have been. Joni Ernst's victory 9-point victory margin in Iowa was much higher than the near-tie polls predicted. Mark Pryor came very close (within 4 points) of Blanche Lincoln's 21-point loss in 2010. Kentucky was called for McConnell almost immediately after polls closed; both McConnell and Pat Roberts defeated their challengers by double digits, despite polls showing at best a mid-single digit gap (and at worst a tiny tiny lead for the challengers). David Perdue avoided a runoff in Georgia; Thom Tillis knocked off Kay Hagan in North Carolina; Sam Brownback and Rick Scott trailed in the polls and held on; Bruce Rauner came from behind to win every county in Illinois except Chicago's Cook County; Tom Corbett kept Tom Wolf's lead in Pennsylvania in single digits despite polls showing a 10-20 point lead for the Democrat; Paul LePage won quite comfortably in Maine; Republicans picked up friggin' Maryland, of all places, and Vermont's gubernatorial is now thrown to the state legislature since Governor Peter Shumlin could not win an outright majority.

However, a definition based on polls is a bit flawed because the polls can be flawed. Looks like they were biased this year--but not, as Democrats in Colorado and other places kept saying, toward Republicans. In fact, this year's set of polls was one of the most Democratic-biased set on the recent record (equaled only by the 2002 cycle). In addition, under this definition 2010 wouldn't count as a wave: polls that year were actually slightly biased toward Republicans.

Why the bias, though? My first instinct was that, looking at relatively low vote returns in Illinois's Cook County (Chicago), Baltimore, and North Carolina's Mecklenburg County (Charlotte), that polls overestimated the black vote--but it didn't quite explain the closeness of races in Vermont, a 94% white state, or in Colorado, where the electorate is only 3% black. Only in Arkansas did African-American share of the vote decrease from 2010 to 2014, but even there the black vote went much more heavily to Pryor than it did to Lincoln four years earlier.

A more plausible possibility is that polls overestimated the vote by women. In spite of (or possibly because of) Mark Udall's single-minded focus on abortion and contraceptive rights in Colorado, the electorate was six points more male this year than it was in 2010. Of course, polls also could have overestimated women's support for Democratic candidates. In Illinois the electorate was two points more female (and it didn't help that Governor Pat Quinn performed even more abysmally downstate than he did in 2010), and in Florida it was a whopping eight points more female (51-49, compared to 55-45 in 2010). Nationally, the electorate was two points more female than in 2010, and the gender gap was slightly less pronounced--women voted 4 points more Democratic than men did this year, compared to a 6-point gap in 2010. I compare exit polls and not polling done during campaign season because 1) we don't have access to the weighting methodologies used by most pollsters and 2) those weighting methodologies are usually based in part on exit polling from previous elections. Why the female vote was slightly depressed and slightly more Republican this year compared to 2010, I'll leave pundits to decide. Speculation is their job, after all.

What went well

For all we got wrong in Kansas and in Florida and in Maryland and in Illinois and in North Carolina (which is why we give probabilistic judgments, not definitive calls, in the first place), we did get a lot right. But so did a lot of other people. One thing we did get uniquely right, though:

• Our Michigan Senate point estimate was spot on. In the last post before Election Night, our model predicted Gary Peters would prevail over Terri Lynn Land 54.5-41.4. The actual result was 54.6-41.4. (All that this really means is that we got lucky, but it's a good feeling nonetheless.)

In addition, our bellwethers from the same post did pretty well. In particular:

• Arkansas Senate. Tom Cotton won Hot Spring County 56-39 while winning the whole state by the same margin, 56.5-39.5. (Yell County went a bit more Republican than the state, 60.2-33.8.)
• Iowa Senate. Joni Ernst carried Jasper County as well as the state of Iowa with 52% of the vote (though Bruce Braley performed one point better statewide than he did in Jasper).
• Louisiana Senate. It was actually our presidential bellwether, Concordia Parish, that performed most accurately; Mary Landrieu won it 42.6-41.3 compared to 42.1-41.0 statewide. (St. Martin Parish was won comfortably by Bill Cassidy.)
• North Carolina Senate. While carrying Lenoir County 49.0-47.7, Thom Tillis won the entire state 49.0-47.3.

There were some others that were within four or five points, but those were the closest.

Thanks for watching.

Oh, never mind, we've got gubernatorial races in Kentucky, Mississippi, and Louisiana next year? Good to know.