OK, I looked around on that site I used, one I'm not familiar with. The averages do in fact include results from generic polls, some of them from years ago.
I will now agree that UT is much closer than the figures I cited (Mormons seem to have a strong aversion to Frump), but I dunno about the other states you pointed to. Texas is perhaps in play as well, but it looks like Hillary will win by double digits in NJ. I am completely certain that Trump will lose in MA. And who cares about WY?
Hi, I'm the guy who runs that site.
First of all, thanks for the link.
Second of all, I'll just briefly explain the "old polls". They aren't actually old polls, they are the results of the previous presidential elections. The basic idea is that in states that have had limited polling of the actual candidate combination in questions I use the previous election results to "jump start" the average based on the overall propensity of the state. As actual polls come in, the election results "roll off". I always use at least five data points to construct the average (more in some specific cases), just on the principle that you should never trust one or two polls in isolation.
In the case of states like Utah, what this means is that you start out assuming it will be a heavily Republican state, since the last five presidential elections varied from R+21.1% to R+48.0% (Average R+36.7%) and then only change that gradually if evidence comes in to contradict that. With three of the five data points now actually Clinton vs Trump, that average is now down to Trump +16.6%. The actual Trump vs Clinton polling has moved the average significantly toward Clinton, but it will take a couple more polls showing it very close or Clinton ahead for the model to fully "believe" it and actually show Utah as a competitive state. Basically, since Utah has been SO Republican in the past, it takes a decent amount of new evidence to move things to where the model actually classifies the state as a swing state.
One more poll showing a close race would probably do that, two definitely would. Utah has been sparsely polled, so the process takes awhile, unlike states that are traditionally "swing states" that have been polled a lot more and where the average has reflected only "real polls" for a long time now.
On my national summary page [
2016 Electoral College - Clinton vs Trump - National Summary ] if you look down at the state breakdown, the states with no parentheses around the average are the ones where the average is 100% polls between the two candidates with no filler from election results. For those that still use election results, the number of parentheses tells you how many. For instance Arkansas has not been polled AT ALL for Clinton vs Trump, so has five pairs of parentheses and is based only on the historical elections from 1996 to 2012... while Minnesota has one pair of parentheses, so it is based on four actual polls, plus the 2012 results... and Iowa has no parentheses, so is 100% based on actual Clinton vs Trump polls. (You can also see this easily on the state detail pages of course, but the national view lets you easily see which states have more or less polling.)
This method seemed like a good compromise between saying "unknown" until there were five actual polls (which would be really unsatisfying and prevent building a good national picture), or basing an average on only 1 or 2 polls when that was all there were (which would give too much weight to individual polls).
It is all explained in my FAQ, but I know most people don't read that.
In any case, hope you find the site useful even if it takes looking a little carefully to interpret correctly sometimes.