He posted an essay on the topic a while back. He gets a lot of shit for getting things wrong using the same data everyone else was using and who also got things wrong; I think the second time, he really started to look into the problems with polling and, in particular, how other predictors were performing.
He was unfortunately the face of the polling failures in 2016, but he’s a first class statistician. People easily forget part of the reason he was so vilified is because, until 2016, 538 was the reliable source for predictions, which speaks as much to how good he is as the subsequent failures. Something went really wrong with polling in the mid-2010s.
Something went really wrong with polling in the mid-2010s.
It became much easier to ignore pollster calls. People were dropping land lines entirely, and gained the capability of automatically ignoring unknown callers. When I was younger, if the phone rang, you answered it. Voice calls simply don’t carry that kind of importance with younger generations, so the respondents to pollsters are going to skew towards older people (and probably some other demographics).
The problem with cold-call polling is that the respondents are necessarily “people who answer the phone for an unknown caller” and then “actually interact with that caller.” How do you weight your polling to consider the opinions of people you cannot get responses from?
The lack of FCC enforcement has virtually made it a necessity. I get anywhere between 8-12 scam calls a day, and it’d be more if I actually answered them. I use the google call screener these days, but yeah, if I answered ever unknown call, I’d go barking mad.
He posted an essay on the topic a while back. He gets a lot of shit for getting things wrong using the same data everyone else was using and who also got things wrong; I think the second time, he really started to look into the problems with polling and, in particular, how other predictors were performing.
He was unfortunately the face of the polling failures in 2016, but he’s a first class statistician. People easily forget part of the reason he was so vilified is because, until 2016, 538 was the reliable source for predictions, which speaks as much to how good he is as the subsequent failures. Something went really wrong with polling in the mid-2010s.
It became much easier to ignore pollster calls. People were dropping land lines entirely, and gained the capability of automatically ignoring unknown callers. When I was younger, if the phone rang, you answered it. Voice calls simply don’t carry that kind of importance with younger generations, so the respondents to pollsters are going to skew towards older people (and probably some other demographics).
The problem with cold-call polling is that the respondents are necessarily “people who answer the phone for an unknown caller” and then “actually interact with that caller.” How do you weight your polling to consider the opinions of people you cannot get responses from?
The lack of FCC enforcement has virtually made it a necessity. I get anywhere between 8-12 scam calls a day, and it’d be more if I actually answered them. I use the google call screener these days, but yeah, if I answered ever unknown call, I’d go barking mad.