February 2, 2016
Donald Trump lead every Iowa poll for the final two weeks of the race. Donald Trump did not win Iowa. Ergo, polls suck.
J. Ann Selzer’s Iowa Poll, conducted for the Des Moines Register and Bloomberg, is considered the gold standard. Everyone fawns over her and her data. Selzer’s final poll had Marco Rubio a distant third at 15 percent, with just over half of Trump’s support. Marco almost caught him at the finish line. Ergo, polls suck.
Not a single poll had Hillary and Bernie in a dead heat. Hillary was 4 ahead in the final Real Clear Politics average. CNN had Bernie up 8, Loras College had Hillary up 29! Ergo, polls suck.
Emerson polled the GOP at the last minute and wound up with Trump 27, Cruz, 26, Rubio 22, awfully close to the final numbers, closer than anyone else. At the same time they published a poll on the Democratic side. Hillary +8. If the same pollster has widely divergent effectiveness, who can you trust? Ergo, polls suck.
It’s easy to blame the polls. Especially this cycle, when polls have determined the placement of candidates in GOP debates. They can create self-fulfilling prophecies. They can also contribute to wall-to-wall Trump. As great a showman as he is, at 6% in the polls, we wouldn’t have heard as much from him.
I love polls and find them tremendously useful. Let’s look at what they did accomplish in the Hawkeye State:
- For weeks, they indicated Trump and Cruz were the two candidates with a chance to win the GOP caucus. One did. The other finished second.
- They noticed Rubio was closing strong. They noticed he was clearly ahead of the governors. They noticed he was pulling well ahead of Ben Carson.
- Averages put Rand Paul and Jeb Bush ahead of the trailing group of Carly Fiorina, John Kasich, Chris Christie, Mike Huckabee, and Rick Santorum.
- In an 11 person GOP race, each candidate wound up in their assigned tier.
- The two candidates who outperformed their final poll averages were among the most popular and were far and away the most common second choices.
- They correctly showed Sanders making this a real race and correctly indicated Hillary was hanging in.
- They consistently indicated Martin O’Malley would not reach viability.
How much more than this could we have possibly expected?
You do need to interpret the polls. When O’Malley is at 4%, you have to adjust those numbers out and use underlying data to speculate who those supporters will caucus with when his group dissolves at most precincts.
You do need to adjust a candidate like Cruz up slightly to account for evangelical grass roots champions traditionally over-performing in the Iowa caucus.
When you run into a Black Swan like Trump, it’s helpful to accept there’s a margin of error. For all the talk about his voters only showing up in the polls, he was all of 4 points shy of his RCP average. The final two individual polls gave him an average of 23.5%. He wound up with 24.3%. The narrative fooled us, not the data.
No matter what, the polls aren’t going to perfectly call the result, regardless of the quality of analysis or algorithm. Nate Silver has managed to solve the past two general elections, but he’ll happily tell you primary season is a different sport.
They give a great outline. We need to fill in the rest with our imagination and the understanding there is some degree of chance (see Hillary winning delegate equivalents by coin flip).
New Hampshire pollsters have given us plenty of clues. Tomorrow morning, we’ll begin to sort through them.