If you'd asked most political people a week ago who'd be celebrating his victory in the Kentucky gubernatorial race Wednesday morning, they'd have probably guessed it would be the Democrat, Jack Conway. Kentucky has a long tradition of electing Democrats to the highest position in the state; since 1971, only one Republican had won the position (Ernie Fletcher in 2003).
But also, the polls! Surveys conducted in late October showed Conway with a decent lead over Republican Matt Bevin -- suggesting that the race would be close, but Conway was likely to win.
Conway didn't win. Conway got beaten, badly -- as did three other Democrats running in Kentucky's six statewide elections.
So how'd the polls get it so wrong? And, as an important corollary, should we, once again, be worried about the polling guiding the coverage of the 2016 election?
First, let's consider the polls, as compiled by Huffington Post Pollster. There were only six since October 1, including an internal poll from Bevin's campaign. (Independent Drew Curtis is indicated with a yellow line.)
Conway over Bevin, until one last Internet-conducted poll on October 28. Then, the big Bevin victory. (He won by almost nine points.)
In the grand scheme of things, though, there really weren't a lot of polls. The last one by a well-known firm was done by SurveyUSA, and it was completed on October 26.(We'll note: That’s one of two polls conducted by well-known firms since October 1.)
If you compare SurveyUSA's numbers with the final results, a pattern becomes clear: Either SurveyUSA underestimated the number of Republicans likely to vote, or everyone who was undecided voted against the Democratic candidate.
That second option isn't terribly likely. It's safer to assume that SurveyUSA underestimated Republican turnout. When pollsters conduct surveys, you'll remember, they typically start with a broad sample of adults or voters and try to weed out people who are unlikely to vote. That process gets more difficult in low-turnout contests, and can result in misreading who is likely to vote. Turnout this cycle -- as estimated by the successfully reelected secretary of state, Alison Lundergan Grimes -- was about 30.7 percent. That's not bad for an off-year general election in Kentucky, but it's low.
If pollsters misread who those voters are going to be, the poll can be weighted wrong across the board. Clearly, Republicans voted more heavily on Tuesday than SurveyUSA expected, meaning that their results were shifted more toward the Democrats than they should have been.
The lesson for 2016? First, that the presidential race will have an advantage that the Kentucky gubernatorial race didn't: Lots of polls. While lots of polls can lead to pollsters mimicking the herd so as not to be an obvious outlier, in general more polls means a better picture of the race. In the national Republican primary, for example, we saw 12 polls conducted between October 1 and November 2 -- and that's including only well-known pollsters. In the presidential race, we'll have more polls from more pollsters that use more-expensive (and more-proven) live phone calls.
Polling is not and will never be foolproof. More polling and more eyeballs before Election Day reduces the risk of getting a result wrong. In Kentucky, both those things were in fairly limited supply.
Scott Clement contributed to this report.