the pollsters definitely get that landlines are old news and most people don’t even answer unknown numbers on their cell phones. The same applies to text requests for political surveys. Response rates — or, rather, non-response rates — are awful. But pollsters know all of that and they’ve come up with pretty smart ways to deal with it. Without getting too far into the weeds, it comes down to increasingly sophisticated ways of modeling the electorate, using those models to weight the results, and in so doing backing out a representative sample from the data.
Just quoting this here because I’ve seen this point made many times.
I do think they’re trying, but its gotta be complex and error-prone to extrapolate like that, especially if there’s some confounding factor that correlates to the likelihood of receiving an answer.
Yeah but that’s why these people are professionals. Of course it’s complicated to do this but statisticians do it all the time. If there’s a known confounding factor (e.g. young people don’t answer calls) then it can be adjusted for. I know polling isn’t perfect but I find these points are less “I have a technical point about the problems with extrapolation/interpolation” and more “This poll doesn’t show what I want so there must be a problem”
I think it is telling how difficult it can be to accurately predict based on polls and all other available data by how rare it is for a professional analyst to make an accurate prediction on something like a federal election 12 or 8 months away.
There were plenty of predictions that got the winner right, but the amount that accurately predicted the ~51/47 are much fewer and I believe the rarity is an indicator that accurate prediction is not as clear and straight forward as some may expect even with statistical training.
For me it’s not so much about the specific number and more “Who will win” and it definitely seems like polls do that quite well, even 8 months out etc as you mentioned before.
I think expecting precise accuracy is quite a high bar. The only real test is the election if that is your stance, and by then it’s too late to do anything about it
Just quoting this here because I’ve seen this point made many times.
I do think they’re trying, but its gotta be complex and error-prone to extrapolate like that, especially if there’s some confounding factor that correlates to the likelihood of receiving an answer.
Yeah but that’s why these people are professionals. Of course it’s complicated to do this but statisticians do it all the time. If there’s a known confounding factor (e.g. young people don’t answer calls) then it can be adjusted for. I know polling isn’t perfect but I find these points are less “I have a technical point about the problems with extrapolation/interpolation” and more “This poll doesn’t show what I want so there must be a problem”
I think it is telling how difficult it can be to accurately predict based on polls and all other available data by how rare it is for a professional analyst to make an accurate prediction on something like a federal election 12 or 8 months away.
Really? What about 2020
Almost all correctly predicted a Biden win. We don’t need the precise number just the general lay of the land
There were plenty of predictions that got the winner right, but the amount that accurately predicted the ~51/47 are much fewer and I believe the rarity is an indicator that accurate prediction is not as clear and straight forward as some may expect even with statistical training.
For me it’s not so much about the specific number and more “Who will win” and it definitely seems like polls do that quite well, even 8 months out etc as you mentioned before.
I think expecting precise accuracy is quite a high bar. The only real test is the election if that is your stance, and by then it’s too late to do anything about it