Looking back at the 2016 Trump vs. Clinton election, There was some interesting shy Trump voter effect. I'm not worried about the actual results in 2020 if this time the polling misses the same voters it did in 2016. Trump's support was often understated.* Below are some of the key states showing how the polls got it wrong leading up to the 2016 election by month. Interestingly, the polls in 2016 often understated* Clinton support, albeit typically by a smaller percentage. That all could be from the impact of third party votes switching late, or cheating. But the polls having Biden +15 are likely experiencing the same mistakes they did last time.
*Note, where the number is negative, it indicates the polls have overstated the candidate's support in that month. Where both candidates' numbers equal exactly zero, it indicates there was no polling in that time frame.
In Florida: Trump highly over-performed his polling right into November. Hillary Clinton in October polls was actually doing worse than her polling, though by November the polls were back to understating her actual results.
In Georgia: Polling generally improved month over month for both candidates, but Trump was hurt more by being understated throughout the late polling cycle.
In Iowa: The polls massively underestimated Trump but were fairly close on Clinton.
In Michigan: The story is similar to Iowa, in October short-selling his support by almost 8% while overselling Hillary's by 1%!
In Minnesota: In October the total error of support for Clinton and Trump was 8%!
In North Carolina: The polls were off by a combined support error of 4.6% in October and even more in November.
In New Hampshire: There was an interesting reversal of error in November but in October the effect was a +8 shift towards Hillary.
In Ohio: Trump was again slighted. You get the idea and pattern forming among all the states here, right?
In Pennsylvania: Much the same as other states, the Trump shy voters are not getting picked up just like elsewhere. Unless, just like elsewhere, the differential that largely evaporates in November was fictitious to begin with. Did the pollsters suddenly figure out how to spot shy Trump voters in November 2016? Or did they not want to be labelled as untrustworthy, and had to start reflecting the truth? The former is more likely true because there are pollsters who largely got it right in 2016 (Trafalgar) that are showing a tighter race than other pollsters who are still showing large Biden leads today. Remember, these graphs are the pollsters in the aggregate by month, not differentiating between correct and incorrect pollsters. A lot (if not most) of the same players are showing the same type of results this time around - both the Trafalgars and the Ipsoses of the polling universe.
In Texas: Here's one they hurt Hillary, dramatically (as well as Trump). But most notably in every month but October. October is THE month this time around, since there are only 2 days worth of potential polling in November, and more realistically, just 1 day.
In Virginia: The bias probably doesn't matter because Biden is obscenely ahead in the state. However, it raises the question - are the polls (either deliberately or through error) overstating the Biden lead even more strongly that they did for Hillary Clinton? Or is Biden actually really that far ahead? If he is, the contest is over. But if not, what is the ratio of Hillary oversampled voters in 2016 to Biden oversampled voters in 2020? I have in all of my assessments so far assumed it's 1-1. Interestingly in my analyses so far, if the Biden bias is only 85% of Clinton's, he still wins the election. At 90% it's not decided yet and at 95% or more, Trump wins.
Finally in Wisconsin: The error was clear. In November they finally got Clinton support right, but there was still a massive undercounting of Trump support.
So what's it all mean? It means GO VOTE. For Trump. Nothing matters - not the polls, not the early voting split, nothing matters except the final tally. I have statistical knowledge and I'm saying the polls don't matter as much as you are being told. If all the polls were done properly, they might. But there's a lot of bias at work, both political and technical. We won't understand it until after the election, and then and only then, is there a chance for pollsters to start correcting their mistakes of 2016 and 2020 (if the latter are still suffering from the same issues as the former). But that's not the election. The election is you turning out and making a difference.
No comments:
Post a Comment
Disagreement is always welcome. Please remain civil. Vulgar or disrespectful comments towards anyone will be removed.