Jump to content

One Thing We Can All Agree Om


Recommended Posts

This is from his final forecast, and I don't see the bit where he's wrong:

 

Quote

But what’s tricky about this race is that — because of Trump’s Electoral College advantage, which he largely carries over from 2016 — it wouldn’t take that big of a polling error in Trump’s favor to make the election interesting. Importantly, interesting isn’t the same thing as a likely Trump win; instead, the probable result of a 2016-style polling error would be a Biden victory but one that took some time to resolve and which could imperil Democrats’ chances of taking over the Senate.

https://fivethirtyeight.com/features/final-2020-presidential-election-forecast/

Link to post
Share on other sites

He's wrong in that the issue isn't with polling error (which generally refers to an unbiased measurement error) but with polling bias. You can reduce polling errors by weighing many polls, but dealing with polling bias is far trickier, and not something silver has really taken on. 

Link to post
Share on other sites
27 minutes ago, Orik said:

He's wrong in that the issue isn't with polling error (which generally refers to an unbiased measurement error) but with polling bias. You can reduce polling errors by weighing many polls, but dealing with polling bias is far trickier, and not something silver has really taken on. 

I thought his model includes poll "quality" which is their attempt to deal with bias (as well as poor sample size, live polling, etc.).

Link to post
Share on other sites

He uses quality to weigh polls (which is how you minimize measurement error), and he has a backwards looking, mean reverted bias number for each poll, but the bias is not just a feature of the poll, it's a feature of the population that depends on the degree of shame, anti-establishmentism, etc. which in turn are numbers you need a separate poll to establish (or if you've got common sense you could tell where they're going to be higher given the current climate) 

As an extreme example, you can run many polls in Israel asking questions that would generate right or left leaning biases, but all polls will result in nearly zero votes for the ultra-orthodox parties. Of course this is well known and pollsters have found ways to deal with it by just looking at the number of registered voters by neighborhood, but things are murkier when there's a significant group that is now being described as "shy" (plus of course if there's an effective and selective mechanism in place to dissuade certain other groups from voting)

 

Link to post
Share on other sites
15 minutes ago, Anthony Bonner said:

so do you think the sample is bad or do you think people lie? I have no idea. I'd guess sample, but that's just because I can't imagine being willing to answer a poll and then lying.

Is it that the pollsters have gotten too far into weighting different parts of their sample and tweaking it?

Link to post
Share on other sites

The part that isn't simple is predicting the level of shyness and its correlation with party or candidate affiliation. Camil Fuchs did extensive work on this after having the same sort of error in the Israeli elections and has developed a social media sampling strategy that's allowed him to correct the biases completely in the last couple of campaigns. I think it's harder in the US because everything gets collapsed into a binary decision. 

Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

Loading...
×
×
  • Create New...