Pollsters do not call people at random. They try to reach a representative cross-section of the voters in, say, the state in which they are conducting the poll in terms of age, race, income, gender, and a variety of other factors. That’s why they ask you all of those preliminary questions.
So if, for example, the state in question is 60% white, they try to make sure that 60% of the respondents are white. Same with the other factors. It is more difficult these days because fewer people answer calls from unknown numbers and, if they do, fewer agree to take the time to answer the pollsters’ questions. Pollsters claim they keep making calls until they get the number of responses they need in each demographic.
One of the key problems in 2016 was reaching Trump voters because, apparently, they were less willing to participate in polls or harder to reach or something. So if, for example, the pollster needed to get 20% of its responses from white, male, blue-collar workers, those in that group they actually reached may not have been representative of that specific demographic. The pollsters checked off the appropriate box, but the results were inaccurate because the responses of this sub-group were not really representative. This is one of the things they are trying to fix for 2020.
Another thing that apparently happened in 2016 is that some number of undecided voters broke for Trump at the last minute. It isn’t that they lied to pollsters – they may have honestly said they were undecided when polled – but when they went into the booth, they pulled the lever for Trump.
In terms of lying to posters, that may occur with some of the demographic questions (how much money do you make?), but I do not believe it occurs to any significant degree with respect to the key question of who are you going to vote for. That just makes no logical sense to me. If that were true, the national polls in 2016 would’ve been inaccurate. They weren’t. The below NYT article discounts the existence of the so-called “shy” Trump voter.
My own surmise is that state polls are more likely to be inaccurate than national polls. For one thing, the smaller the voter base, the more likely the results will be inaccurate. If a town has 10 voters and you only poll two of them, your results may not reliably predict the outcome of the election. If the town has 1 million voters and you survey 200,000, your results are likely to be more accurate than they would be in the small town.
Another reason is that I suspect state polls are less well-funded and put together with less sophistication than national polls. Accordingly, they may be less accurate.
To my mind, the key thing pollsters have to try to do a better job of this year is getting responses from Trump voters. I think that was the main problem in 2016, though not the only one. Whether pollsters will be able to improve their performance this time remains to be seen.
Like them or hate them, accurate or inaccurate, polls are the best information we have as we careen towards November 3.
https://www.nytimes.com/2017/05/31/upsh ... trump.html