I don't know that I'd call Whit Ayres a "highly partisan low-quality" pollster. Dude has been in the game for quite a while and isn't exactly a fan of Trump...
I don't disagree with the face value of the rest of what you wrote, but that said, many of the instances of modern polling shortfalls that you pointed to *have* fallen within the m…
I don't know that I'd call Whit Ayres a "highly partisan low-quality" pollster. Dude has been in the game for quite a while and isn't exactly a fan of Trump...
I don't disagree with the face value of the rest of what you wrote, but that said, many of the instances of modern polling shortfalls that you pointed to *have* fallen within the margin of error of polls. The percentage splits on polls are a *range of values* based on the margin of error but so often people just go with the single-value reporting that polls give and don't look at the error margin. If you actually +/- the margin of error values onto the single-value outputs, those value envelopes almost always contain the same values as actual outcomes, but people get stuck on the single-value outputs of polls without factoring in the margins so it looks like the polling is always wrong.
For example, take this ABC/WaPo poll released Nov 7th, 2016 (just before election) that goes 47/43 for Hillary:
The margin of error on that poll is 2.5 points. If you add that 2.5 points to Trump's margin and take it away from Hillary's you get a 44.5/45.5 result instead in favor of Trump--which is within the polling margin of error:
All of that said, Hillary still netted more votes than Trump did, so technically the poll was right with respect to national voting, it's just not reflective of the electoral college. Trump got 62,984,828 votes, Hillary had 65,853,514. That's pretty close to a 47-43 split (the percentages don't add to 100, so multiply combined vote totals by 1.1 for that missing 10% and then find 47% of that value, which is 66M--close to Hillary's actual numbers) and certainly within the margin of error of that poll in either direction.
I don't know that I'd call Whit Ayres a "highly partisan low-quality" pollster. Dude has been in the game for quite a while and isn't exactly a fan of Trump...
I don't disagree with the face value of the rest of what you wrote, but that said, many of the instances of modern polling shortfalls that you pointed to *have* fallen within the margin of error of polls. The percentage splits on polls are a *range of values* based on the margin of error but so often people just go with the single-value reporting that polls give and don't look at the error margin. If you actually +/- the margin of error values onto the single-value outputs, those value envelopes almost always contain the same values as actual outcomes, but people get stuck on the single-value outputs of polls without factoring in the margins so it looks like the polling is always wrong.
For example, take this ABC/WaPo poll released Nov 7th, 2016 (just before election) that goes 47/43 for Hillary:
https://abcnews.go.com/Politics/clinton-trump-campaigns-end-close-unpopular-poll/story?id=43344414&mc_cid=0aa18f39a7&mc_eid=d103da7bcc
The margin of error on that poll is 2.5 points. If you add that 2.5 points to Trump's margin and take it away from Hillary's you get a 44.5/45.5 result instead in favor of Trump--which is within the polling margin of error:
http://www.langerresearch.com/wp-content/uploads/1184a162016ElectionTrackingNo16.pdf
All of that said, Hillary still netted more votes than Trump did, so technically the poll was right with respect to national voting, it's just not reflective of the electoral college. Trump got 62,984,828 votes, Hillary had 65,853,514. That's pretty close to a 47-43 split (the percentages don't add to 100, so multiply combined vote totals by 1.1 for that missing 10% and then find 47% of that value, which is 66M--close to Hillary's actual numbers) and certainly within the margin of error of that poll in either direction.