He’s essentially the pre-eminent expert on polling, polling errors, and best practices in that regard.
i would not call Silver this anymore; he’s an increasingly partisan libertarian with weird hangups and a stick up his ass about things he wishes he understood (like COVID-19), and it’s almost assuredly part of why he’s out of a job at 538. he’s also increasingly being lapped by people like G. Elliott Morris and an armada of Twitter-based election analysts and prognosticators. at the bare minimum he’s absolutely not the only guy in town on this, and some of the people i just described like Adam Carlson actively dispute he’s even using the data in this article correctly because they’re the ones who made it. he’s also being challenged on his points here by professors like Matt Blackwell, specific polls of Black voters and data about minority youth voters. in general: the case is not nearly as clear-cut as he makes it seem and deferring to him on this would be a bad case in which to do it, because he’s probably wrong.
Only time will tell, but I remember when the same sort of claims were made about him calling the 2016 race a toss up. Although it does look like I have some links to read.
Silver was correctly giving better odds of Trump winning than basically any other prognosticator in 2016, but i think even he would agree with the point that past success is not strictly indicative of future results. that he got 2016 correct does not mean his analysis today will hold up well (and indeed there are a few reasons to think he is a worse, more partisan, more rigid-in-approach-and-analysis prognosticator now than he was in 2016, some of which i touched on here).
circumstances have also changed, in any case: polling is struggling to methodologically keep up and capture representative samples in a way they simply wasn’t true in 2016. we saw subsample results like what we’re seeing now in a few 2022 state-level polls and they were wrong then, and additionally no elections have yet convincingly or reliably demonstrated the kinds of shifts being implied currently. even discounting special elections, it seems very clear for example that Republicans did not get 20% of the black vote in Virginia in 2023 based on their House of Delegates results. by far the most compelling argument i’ve seen that accepts current polling data at face and not as error is the argument by Nate Cohn that Democratic weakness is among non-voters–but that obviously invites the question of: why should we assume these people will vote in 2024?
i would not call Silver this anymore; he’s an increasingly partisan libertarian with weird hangups and a stick up his ass about things he wishes he understood (like COVID-19), and it’s almost assuredly part of why he’s out of a job at 538. he’s also increasingly being lapped by people like G. Elliott Morris and an armada of Twitter-based election analysts and prognosticators. at the bare minimum he’s absolutely not the only guy in town on this, and some of the people i just described like Adam Carlson actively dispute he’s even using the data in this article correctly because they’re the ones who made it. he’s also being challenged on his points here by professors like Matt Blackwell, specific polls of Black voters and data about minority youth voters. in general: the case is not nearly as clear-cut as he makes it seem and deferring to him on this would be a bad case in which to do it, because he’s probably wrong.
Only time will tell, but I remember when the same sort of claims were made about him calling the 2016 race a toss up. Although it does look like I have some links to read.
Silver was correctly giving better odds of Trump winning than basically any other prognosticator in 2016, but i think even he would agree with the point that past success is not strictly indicative of future results. that he got 2016 correct does not mean his analysis today will hold up well (and indeed there are a few reasons to think he is a worse, more partisan, more rigid-in-approach-and-analysis prognosticator now than he was in 2016, some of which i touched on here).
circumstances have also changed, in any case: polling is struggling to methodologically keep up and capture representative samples in a way they simply wasn’t true in 2016. we saw subsample results like what we’re seeing now in a few 2022 state-level polls and they were wrong then, and additionally no elections have yet convincingly or reliably demonstrated the kinds of shifts being implied currently. even discounting special elections, it seems very clear for example that Republicans did not get 20% of the black vote in Virginia in 2023 based on their House of Delegates results. by far the most compelling argument i’ve seen that accepts current polling data at face and not as error is the argument by Nate Cohn that Democratic weakness is among non-voters–but that obviously invites the question of: why should we assume these people will vote in 2024?