June 8, 2020

Polling filters and Trump's true chances.

I have created an Excel spreadsheet to track the presidential polling and project the 2020 election winner based on filtering out polls that are of questionable value.  It's not bias, it's applying standards across all pollsters in the RCP polling, on a state by state polling basis.  It's a cumbersome process to maintain but worth it to understand what's really happening this election year.

I apply several criteria (note that these are adjustable in my spreadsheet so I can be more or less stringent.  However I try to be reasonable and not change the settings); 

  1. How recent is the poll?  Right now I include polls from the last 45 days but realistically it should be 30 days, and closer to the election, less and less.
  2. Sample Size.  Any poll that uses less than 500 participants is small enough to start having statistical significance issues.
  3. Sample Type.  Are they polling likely voters, all registered voters, or all adults?  Likely voter polls tend to be more accurate than registered voters and all adults is a pointless poll.  There is a caveat to this - some pollsters may skew how they determine likely voters either on purpose or in error.  This can have a positive or negative effect for either candidate. There are additional ways to compensate for this.
  4. Margin of Error.  What is the +/- on the margin of error for the poll.  The larger the Margin of Error, the less valuable the poll.  This is tied to a number of factors, most notably, the sample size. A poll with a +/- 2% is far more trustworthy than one with a +/- 5%.  I prefer to use 3% but am currently allowing all the way up to 4.5% margin of error just to try to include as much information as possible. This will drop as we get closer to the 2020 election, but 3% is a good reasonable measure.
  5. Lastly, I've used Five Thirty Eight's rating of the pollsters.  They rate pollsters from A to F (and ones they have banned).  I only accept any pollster they rate as a B or higher.  They also have a partisan rating that indicates if the pollster due to ideology or methodology shifts the poll towards Democrat or Republican.  I adjust those pollsters results accordingly.
My filtered results .
With all of these polling filters, I go state by state for anything resembling battleground states and scrape the polls in realclearpolitics.com poll data.  I then include/exclude polls on a state by state basis and take not an average, but a weighted average of votes to determine who will win that state.

Using this approach, every swing state is undecided.  Most in fact have no polls that qualify as admissible yet.

That's good news and bad news for president Trump.  Good news in that what you hear as far as job approval, and national polls vs. Biden, the president looks to be in bad shape but those are not realistic views.  It's bad news in that he still has a lot of work to do to lock up a re-election.  I think he's in better shape than Biden, it's just that I cannot prove it yet using my tracking methodology.

No comments:

Post a Comment

Disagreement is always welcome. Please remain civil. Vulgar or disrespectful comments towards anyone will be removed.

Related Posts Plugin for WordPress, Blogger...

Share This