Friday, July 15, 2016

Not So Unusual a Question

Every election cycle a handful of journalists discover that asking survey respondents to predict who is going to win the election can be far more interesting, and accurate, than the usual asking of who they support and adding up the results. A HuffPo piece posted online Thursday afternoon is the latest. It includes this breathless headline and subhead:
Unusual Polling Question Reveals Which Candidate
Is More Likely To Win In November 

We usually ask voters which candidate they plan to cast a ballot for.
But asking which one they think will win actually reveals more.
Asking survey respondents to predict who is going to win, that's "unusual?"

Not really.

The ANES has been asking this question in every presidential election year since -- wait for it -- 1952. Go to the ANES core question list and search for "Who does R think will be elected president in November" and you'll see it listed for all those years. I've analyzed this question extensively for decades, so I kinda know this question, its accuracy, and even its theoretical strengths and weaknesses.

A note to the HuffPo author and everyone ... yes, this question is accurate, but not always so. Recent example, Brexit. More U.K. folks predicted Remain would win than Leave, and we all know how that turned out (my breakdown of that question here). The article does a nice job linking to several of the "Who's gonna win?" questions so far this election cycle, and notes that Hillary Clinton scores significantly higher than Donald Trump. One recent question, for example, has it Clinton 54-26 over Trump in prediction.

But these are national surveys. I'm not saying they're wrong, I'm saying it's too early to pay any attention to the "Who's gonna win?" question, plus as we all know the U.S. presidential election is by state, not nationally, though we can take a lot of guidance from national polls. As an aside, ANES also often -- but not always -- asks for a prediction of a respondent's state as well.

That caveat aside, let's look at some data. First off, people tend to believe their own candidate will win. Three-fourths of Mitt Romney supporters believed he would win in 2012. See the graphic below. As you can tell, if you add predictions of victory by the eventual winners and eventual losers, the result ranges in the 70s or 80s, percent-wise. Indeed, it's gone up in the last few elections, which in itself is kinda interesting.

So how accurate is the "Who's gonna win?" question? Very. The last time it was off was the hotly contested 2000 election and in some ways it was right in that Al Gore did indeed win the popular vote, just not the Electoral College. In general, as research shows, the question is accurate, though the percentages are often higher on prediction question for the winner than they are for the traditional counting up of preferences. For example, two polls in November 2012 had Barack Obama over Romney with 57 and 55 percent of the vote, respectively. Obama won with 51.1 percent of the popular vote. So there is a bit of inflation here. It's better at predicting a winner than the degree to which a candidate will win.

No comments: