About Polls

Polls, polls polls, so many come out every week we get quite conflicting forecasts.

With the number of polls going on you would think we are all being inundated with people asking for our opinion.  Who are the people in the polls?

Since I don't answer the phone is I don't know the caller and don't click on questionable links I know I'm not in there.

In fact it is a big generally undisclosed issue with polls about who they are reaching and how representative that group is.  They publish a margin of error, but I doubt that's as accurate as it should be based on the sample problems.

A perfect poll would get  a random snapshot of the people of this country.  That would includes all demographic categories and be verifiable.  This is essentially an impossible task so the pollsters have to adjust for this in some way.

I'm not questioning the statistical skills or dedication of the pollsters, although some do have a point of view, but they tackle an impossibility and have to adjust for the unreachable.

So ultimately it becomes a bit of a Scientific Wild Assed Guess (SWAG) meaning its correct if all the uncertainties sort of balance out.

What is actually surprising is how close they come, much of the time considering beside the people they can't reach, the possibility that the people they do reach are lying to them.

Now most American are honest and reliable, but some aren't and giving the opportunity they may take delight in trying to hack a poll.

Its hard to say who is underrepresented in the polling although there have been some allegations that certain Trump supporters are embarrassed to tell the pollsters they support him.  This doesn't seem like any of the Trump supporters I have met, but of course it is possible.

Now some sites average the polls together to get a less variable outcome.  Of course there are statistical issues with this as well, but it is likely to be more accurate than any individual poll.

With two months to go it is likely that we will be bombarded with hundreds of polls and many will show significant variations.  They are indicative of what they sample and as a statistical person I find them interesting, but remember they are trying to predict using flawed samples and statistical adjustments.  They may be presented by the media as conclusive when in fact they are SWAG.
This is a good discussion on how to read polls.

http://fivethirtyeight.com/features/13-tips-for-reading-general-election-polls-like-a-pro/