It will boggle the minds of everyone who’s ever taken a statistics class, but nonprobability polling is the future. Conventional (probability) polling rests on the assumption that a representative sample can allow estimation of the population. Nnonprobability surveys, which do not rely on random sampling and instead recruit through ads, pop-up solicitations and other approaches.
In a simple sense, consider what we see when Donald Trump gets 80% of the vote on the Drudge Report debate polls. The enthusiasm we show in these “non scientific” suveys is still a valid indication of public opinion. As is the fact that Democrats do not vote for Hillary in anywhere near the same way.
>Conventional (probability) polling rests on the assumption that a representative sample can allow estimation of the population
There’s nothing wrong with the math; the problem is that current polling, as this article shows, does not _get_ a representative sample! Current methods only get the people who own land lines; are near them when pollsters call (often mid-day); and are willing to endure the poll.
That’s on top of any polling bias wherein companies ‘adjust’ the results to match a prior turnout or defuse a result they’re not paid to obtain.
There are ways to correct for that, but they are complex and you can add a lot of error to the result.