Saturday, November 5, 2016

Final Thoughts on Polls Before Presidential Election 2016

I don't think many people took me seriously when I predicted that Trump and Clinton would be close in the polls close to Election Day. And yet, here we are. In fact, the sine wave has already reversed and by Election Day, Clinton could be significantly ahead again. A few final thoughts before the election:

First and foremost, I've said for a while, this is a really Presidential weird election year, and I think the least likely to match outcomes between polls and results.

Last week, Trump saw the biggest rise in the last 30 days of any election of any candidate since 2004; Obama's rise in 2008 was about 2.2 heading into the election, and no other years featured any rises in the last month. Despite that rise, the outcome of an election has never changed in the last 5 days.

People ask me why I follow the RealClearPolitics polls and no others. It's because, statistically, their average should be more accurate than any one poll, especially ahead of the election. Here are their results for the last three Presidential elections, with final Popular Vote at the top and the one-day-before-election RCP poll average below.



They correctly called all three one day before the election. In 2004 & 2012, the winner took even more than they predicted; in 2008 it was about the same. As such, the RCP average on Monday should determine the winner (this is all about Popular Vote; Electoral is a different thing).

This year though, I'm worried about their accuracy for the first time. For some reason (that didn't happen in previous elections), the LA Times Poll is way off from any other poll. It's so far off that in statistics, we would call this an "outlier" (and in many cases, ignore it). But RCP takes every legitimate poll, so they include it. Here's what it has looked like for the last four months:



Compare that to the RCP average:

 

While the average has predicted that Clinton would win for all but a few days during that time, the LA Times poll is opposite -- insisting that Trump would win on most days. This could give Trump voters false hope -- but more importantly could skew the entire poll average enough to make the race look less close than it is, for all voters.

If this poll proves to be way off from final results, I believe that RCP should remove it from its list of legitimate polls. There's obviously something wrong with their methodology if they are that far off (and incorrect in the end). If it's correct, then it's a brilliant poll which we should all watch closely next time. LA Time also polls people on who they *think* will become president (regardless of who they are voting for). This, interestingly, looks much like the RCP polls, which I think makes the LA Times voting poll even more suspect.

2 comments:

  1. Jesse, good points as always, but I will raise one point about the LA Times Daybreak Poll: even though it shows their respondents as more likely to vote for Trump, those very same people think Hillary will become President... for me anyway, this can be easily explained by the Mainstream Media Bias for Clinton which would give the "average american voter" (and the poll respondents) reason to believe Hillary will win, thoughts?

    ReplyDelete
    Replies
    1. That could be... though it still doesn't explain why their respondents are so Trump-heavy.

      Delete