Monday, July 17, 2017

The Expectations Game

I'm always fascinated not so much by who people say they're going to vote for as who they think is going to win. Generally, people say their preferred candidate will win and you can see that below. I'm playing with some 2016 election data. Of those who said in the pre-election survey they preferred Clinton, 96.2 percent predicted she would win. Of those in the pre-election survey who preferred Trump, 75.6 percent predicted he would win. For you statistical nerds out there, that's a X2 of 1399.7, p<.001. In other words, a huge association. See the table below for a summary.

Who Will Win

Who Voting For


I excluded the handfuls of people who preferred or said they would vote for Gary Johnson and Jill Stein. Their numbers are too small to matter, at least when comparing the preference-expectation link. 

We call the above findings wishful thinking, a body of research that consistently demonstrates that people tend to believe their sports team or candidate will win, even when that candidate or team is behind. 

When time allows I'll look at the predictors of wishful thinking in the 2016 election. Who were the 52 Clinton supporters who predicted Trump would win? How do they differ, if at all, from the 1,333 who said Clinton would win? And vice versa for Trump. We do know that affect, as in emotion, plays a huge role. The more you care about an outcome, the more likely you are to engage in wishful thinking. Education and knowledge tend to, at least somewhat, make people more accurate. The role of the media is kinda mixed. In my analysis of 2012 data, I found that watching partisan news, such as Fox News, made you more likely to inaccurately predict Mitt Romney would win, even after controlling for lots of other factors such as caring about the outcome. It'll be fun to see if watching MSNBC, for example, has the same effect on Clinton supporters in 2016 that Fox had on Romney supporters in 2012.

Wait. I do have one quick analysis to share. The only MSNBC program in the data, that of Chris Matthews (Hardball), there seems to be an effect. Among those who watch Hardball and supported Clinton, not a single one predicted Trump would win. For those who didn't watch Hardball and supported Clinton, 4.1 percent predicted Trump would win. That's not a powerful effect, but it is suggestive. There's a similar result for watching CNN's Anderson Cooper, but not quite as strong.

Again, when I have time I'll build a multivariate model and see what separates the accurate from the inaccurate.

A final word. The overall expectation in the numbers above was that Clinton would win -- and she did -- at least in terms of the raw popular vote, and that's what this poll reflects. Most respondents are not doing a state-by-state Electoral College analysis in their heads when answering a survey question like this. The same happened in 2000 when George W. Bush lost the popular vote to Al Gore but won the Electoral College. The ANES data that year had more folks expecting a Gore win. And he did, in terms of the popular vote. A more nuanced analysis here would include the state the respondents live in and whether they were accurate predicting their state versus the national outcome. Yes, I have those data.

Data source: ANES 2016 survey

No comments: