538 did a pretty admirable job this time of presenting uncertainties around polling averages. I followed pretty closely & the potential for error was made very clear. But @zeynep has good criticisms rooted in the way most people view models -- and polling itself has big problems.https://twitter.com/NateSilver538/status/1325858783923417088 …
-
Show this thread
-
Replying to @JonMurray @zeynep
Polling is in big trouble. I think it struggled to balance rural voting with suburban vs city. If you look at something like the senate race in Maine Gideon won all of the large cities and then got really outran in the more rural areas.
1 reply 0 retweets 0 likes -
Replying to @mcconnellj @zeynep
Yeah, that's one of the really stark examples. One way I think about it is that polling is both art and science. But the more nonresponse rates go up, the more you have to pad the solid science part of it -- probability theory -- with the art of weighting etc. It gets unwieldy.
1 reply 0 retweets 1 like -
Replying to @JonMurray @zeynep
I totally agree. Assuming you had great response rates, and a good sample polling should still be a great indicator. I am just not sure we have either of those anymore.
1 reply 0 retweets 0 likes -
Replying to @mcconnellj @zeynep
That's one of the points made by
@zeynep in interviews/comments. When response rate is 30-40%, you can work around by reaching out to more people. When it's down to 6% or so, you start missing entire groups (espec when trust in institutions is way down). Can't model out of that.1 reply 0 retweets 1 like
Right, and especially since we have both unreachable segments and people who are resistant to the polls because of cultural mistrust. I don't see how we fix this, structurally, without finding a new way to randomize.
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.