Author Topic: Polling Analysis  (Read 158 times)

yossarian22c

  • Members
    • View Profile
Polling Analysis
« on: November 05, 2020, 02:47:37 PM »
Final results aren't in yet but wanted to take a look at the final 538 predictions vs outcomes. Trump outperformed the polls in many places but in others they were relatively accurate. Some of the vote totals may shift a little more towards Biden as mail in votes are counted but just doing a quick check on where polls were okay and where they missed.

I'll look through the swing states and compare the vote so far to the final 538 average of polls. I'll give the Trump share of the poll and error.

State:    Poll           Vote         Poll Error
Texas    50.3%      52.2%         1.9%
Ohio     49.8%       53.4%         3.6%
Iowa     50.0%       53.2%        3.2%
Georgia 49.2%      49.6%         0.4%
NC        48.8%      50.1%         1.3%
Florida   48.5%      51.2%         2.7%
Arizona  48.1%      48.1%         0%   *Significant outstanding vote
Penn      47.3%      50.3%        3%    *Significant outstanding vote
Nevada  46.2%      48.5%        2.3% *Significant outstanding vote
Michigan 45.5%     47.9%        2.4%
Wisconsin 45.4%    48.8%       3.4%
Minn      44.6%      45.4%       1.2%


3-4 Percent would be an average margin of error for an individual poll. The big issue is that Trump outperformed the polls across the board. The upper midwest seems to have the worst errors. Everywhere else except Florida seemed to measure Trump support within about 2%. 2% is a big deal when it changes who the winner is, but in terms of statistics an error under 2% is reasonable. Taken as a collective consistently missing in one direction by about 2% is more problematic. Also since 538 is doing a meta analysis on multiple polls we should consistently see errors of less than 2% and some in favor of either candidate if there isn't any type of sampling bias. Maybe it really is people in the upper midwest support Trump but lie to pollsters about it. Hard to say, years without Trump on the ballot the 538 polling averages have been nearly perfect. Years with him on the ballot we've seen the consistent error away from Trump.

My quick analysis is there is something going a little wrong with polls in the upper midwest and Florida. The rest of the sun belt looks to have been predicted about as accurate as you can expect. Even if ideally you would have overestimated Trump's support somewhere as well. But IMO measuring a population statistic within 2% is "passing."

Also as a note we sometimes think the polls are worse than they were when we measure the gap between the candidates because it doubles the error. So a poll that had Biden 50% Trump 48% (B+2) and the actual result is Biden (48%) Trump (50%) (B-2) is really measuring the values to within 2% but the difference between the margins (+2 to -2) is 4%.

TheDeamon

  • Members
    • View Profile
Re: Polling Analysis
« Reply #1 on: November 05, 2020, 03:09:09 PM »
If you'll remember, I had previously been posting on the topic of the "shy trump voter" and some polling operations were estimating about 10% to 5% split on "unreliable reporting" of voting intentions among Trump vs Biden voters.

Assuming a perfect 50/50 split between trump and biden voters, where the 5% of unreliable biden voters immediately cancel out 5% of trump voters, that still means 5% of Trump voters would be reporting intentions to vote for someone other than Trump.

Incidentally, 5% of 50 works out to 2.5

Which is about where the margin of polling error consistently hit in the above example. So you can take your pick, if their methodology was in error. Or people were misleading the pollsters.

JoshuaD

  • Members
    • View Profile
Re: Polling Analysis
« Reply #2 on: November 05, 2020, 03:35:33 PM »
https://electoralmap.net/2020/

Got it pretty close to right, it seems. Arizona's the only big surprise to his prediction.

I was watching 538, RCP, and this site, and was surprised that he got it most right. He was the only one showing a close race through the past 4-5 months, and it looks like his analysis was best.

TheDeamon

  • Members
    • View Profile
Re: Polling Analysis
« Reply #3 on: November 05, 2020, 03:43:13 PM »
Have to wonder how much of the "Arizona surprise" was a consequence of Californians moving to Arizona over the past 8 months as a consequence of the Covid lockdowns.

DJQuag

  • Members
    • View Profile
Re: Polling Analysis
« Reply #4 on: November 05, 2020, 05:42:56 PM »
Have to wonder how much of the "Arizona surprise" was a consequence of Californians moving to Arizona over the past 8 months as a consequence of the Covid lockdowns.

Or a proposition being up for voting, that being, "Should we allow recreational cannabis?"

It's anecdotal but based upon the people I saw growing up that is something that would have motivated them to get into the polling booth. Them being younger voters who, since they were there anyway, signed on to "Duck Trump."

Younger voters swing wildly liberal yet they're pansies who can't be arsed to actually vote on the day of. Well, for the most part. It's why the political establishment as a whole discounts them and why Bernie couldn't get past the primaries.

In a battleground state all it can take is a small nudge and I have a feeling this surprise may well have come out of that proposition.

DJQuag

  • Members
    • View Profile
Re: Polling Analysis
« Reply #5 on: November 05, 2020, 05:47:57 PM »
And I'll just say it's pretty funny that the "legalise it," crowd somehow never came under the umbrella of the, "Screw the feds, state's rights, people have inherent choices about their own bodies and choices in life," crowd.

LetterRip

  • Members
    • View Profile
Re: Polling Analysis
« Reply #6 on: November 06, 2020, 06:15:30 AM »
Legalize pot has been on the ballot in AZ for many elections.  I doubt it impacts turnout.

Grant

  • Members
    • View Profile
Re: Polling Analysis
« Reply #7 on: November 06, 2020, 08:22:09 AM »
Have to wonder how much of the "Arizona surprise" was a consequence of Californians moving to Arizona over the past 8 months as a consequence of the Covid lockdowns.

I think it has more to do with McCain having become more powerful than Trump can possibly imagine. 

DonaldD

  • Members
    • View Profile
Re: Polling Analysis
« Reply #8 on: November 06, 2020, 08:37:10 AM »
I think it has more to do with McCain having become more powerful than Trump can possibly imagine.
LOL.  We see what you did there.