March 2, 2016
There were 22 contests (11 Democratic contests, 11 Republican) yesterday if you don’t count whatever the GOP was up to in Wyoming. No surprises. None. Sure, Ted Cruz did a little better than expected, Marco Rubio a little worse.
If everything was exactly what everyone figured, no need to waste money with the actual voting. But this was very, very close. There was one ludicrous Democratic survey in Minnesota last month. Otherwise, even the states with very limited data usually had the right clues.
Let’s take a quick tour of the map:
Alaska: We predicted this incorrectly. Trump did not win by 10 points over Cruz. Instead Ted finished narrowly ahead of The Donald. How is this a victory for the pollsters? Because surveys consistently showed Marco Rubio had no constituency in Alaska.
That was the key lesson, and it was visible in the data. Many of us decided Cruz had negative momentum and rounded him down (or didn’t give him his normal boost for turning out voters), but that’s bad analysis, not bad data.
It looked like a Trump-Cruz race and it was. As soon as it became clear Ted was doing well in Oklahoma and Arkansas, it was easy to project a very close outcome in Alaska
Minnesota: Of all the states voting yesterday, Marco Rubio led in one poll in one state. Care to guess where that was? One state had Donald Trump at serious risk of finishing third. Both proved out.
The Star-Tribune/Mason-Dixon poll from January was absurd. It had Hillary 30 points up on Bernie. He won relatively easily. This is the sole polling embarrassment of the evening.
FiveThirtyEight’s demographic analysis correctly showed this as one of Bernie’s better targets.
Colorado: There was no polling from 2016, only from November when Hillary had a larger lead nationwide. FiveThirtyEight indicated this was a strong target for Bernie, and it turned out that way.
Oklahoma: Late polling showed Trump pulling ahead. The betting markets had The Donald as a big favorite as the morning started. This was a bit of a miss. Still, Cruz and Rubio had polled well overall.
This was in a bucket with competitive states like Arkansas, not definite Trump states like Alabama, Georgia and Massachusetts. This wound up being the least accurate of the recently polled states. The result was very similar to Iowa, where Cruz did several points better than projected, Trump did a few worse, and the outcome flipped as a result.
Bernie led in one poll over the last few days outside of Vermont. It was here. Even those showing Hillary ahead had a large number of undecided voters. She failed to crack 50%. Even the undecided numbers were predictive. The two candidates combined for barely 90% of the final vote.
Sanders exceeded his poll margin, but they correctly showed this was by far his best non-Vermont primary state.
Texas: Polls showed Cruz surging late. Polls showed Rubio at definite risk of falling below the 20% cutoff. Polls showed Hillary beating Bernie by a 2:1 margin. Exactly as indicated.
Arkansas: Another victory for the polls. Hillary eviscerated Bernie as expected. An early February poll showed substantial strength for Cruz. Rubio was competitive. A late Survey Monkey poll put Trump ahead by a little over Cruz. That’s what happened.
Alabama: Projected as Hillary’s best southern state and Trump’s best southern state, it was both. One of several places where pollsters correctly showed Cruz and Rubio battling for a distant second and showing one or both at risk of falling below 20%. In this case, Marco wound up below the line.
Tennessee: We didn’t have any data for months, then got a couple scraps, which proved extremely accurate. This was the Trump/Clinton state surveys said it was. Rubio and Cruz fought for second and to stay above 20% just like polls indicated.
Georgia: See above, only with more consistent data. It looked slightly less overpowering for Trump than Alabama and Tennessee and was. Hillary won by the huge margin the polls said she would.
Massachusetts: Bernie closed well to make it close, but the final several polls had Hillary ahead by modest margins. Her victory didn’t surprise anyone who was watching the polls.
Meanwhile Trump was within a few points of his final average, as was Kasich. Rubio was exactly on it and Cruz was within a point. Very, very accurate overall.
Vermont: There were two surveys taken on the Democratic side. Bernie’s average margin was 75%. He won by 72.5%. Think we can call that good, especially with the limited sample.
Only one GOP poll taken over the first half of February. Trump pulled 32% and wound up at 32.7. Rubio polled at 17% and wound up at 19.3%. Cruz fell from 11% to 9.7%. Over 20% of the voters surveyed were undecided or had chosen a candidate who was no longer available.
They all picked Kasich. He moved from 10% to 30.4%. Without polling in the second half of the month, nobody was going to catch this, but the survey clearly identified an available block of votes that could make the primary close if they migrated to the same place.
Pollsters are regularly criticized these days. There is a margin of error, and a race seemingly going to one candidate by a few points can easily wind up going the other way. Primaries are much harder to survey than general elections.
With several candidates, many of whom are acceptable to an individual voter, there’s plenty of room for minds to change. Even when pollsters are right on it, history shows multiple times where the numbers moved 20 to 25 percent over a long weekend.
Ronald Reagan reversed things almost overnight on George H.W. Bush in New Hampshire in 1980. Bush returned the favor there against Bob Dole eight years later. More recently, Mitt Romney recovered against Newt Gingrich in Florida in 2012 quicker than you can blink an eye.
So far, the assembled surveyors are having a good primary season. Even when we’ve had limited polling in a low-turnout caucus state, outcomes have fallen within the range of what was indicated. They’ll miss on something occasionally, just because it happens, but so far, it’s safe to pay attention to them.
Good job guys.