North Carolina:
- PPP - Biden +2
- Gravis - Trump +3
- CNBC - Biden +2
- Civtas Harper - Trump +3
- NYTimes - Biden +9
Michigan:
- Trafalgar Group - Biden +1
- CNBC - Biden +2
- NYTimes - Biden +11
Arizona -
- CNBC - Biden +1
- NYTimes - Biden +7
Pennsylvania
- CNBC - Biden +2
- NYTimes -Biden +10
Obviously polling doesn't look good for the President right now, nor do I expect it to change much in the near future. In fact, if history is our guide, most pollsters won't provide us with their real honest to goodness polling results until a week to ten days before the election. In 2016, it wasn't until the last two weeks or so that Hillary's larger leads started evaporating into the three to five point leads that nearly everyone showed at the end.
As I pointed out the other day, Hillary still held double digit leads in a double digit amounts of polls released within 30 days of the election. I don't see any reason why we will not be seeing these same double digit Biden leads until November actually rolls around. Then, of course, all of the pundits will feint confusion as to why all of the polling converged (again) in the last stages of the election cycle.
Harry Enten has been working overtime trying to convince everyone that Biden is running ahead of the pace that Clinton was running. He seems to take stock in the idea that a 48-38 polling lead is significantly better than a 46-36 polling lead, as it puts Biden closer to where he needs to be. I neither agree or disagree with this theory at this point. I see little use in polling taking place in the middle of significant events that will likely not be what we are talking about in October and November.
I would also offer that Enten misses the larger point. In 2020 (especially in the state polling) there was quite literally two different subsets of polling. If you graphed out the polls, you would have seen almost no polling showing the race where the cumulative average (mean) was. What you saw was two groupings of poll results. One subset of polls showing Clinton doing considerably better than the average and one subset showing Trump doing considerably better than the average.
My personal experience in watching these sorts of polling is that generally the answer is not found in the middle, but rather one subset or the other subset will ultimately be correct and the other wrong. The difference between the two will almost always be assumptions about who will show up to vote. It seems that every year there are pollsters who make assumptions that "this" is really the year that the electorate will shift dramatically. Then there are other pollsters who keep themselves tied closely to historical demographic turnouts and poll accordingly. Those assumptions lead to wildly different results.
Once you eliminate those assumptions and focus on the raw numbers, the cross tabs were not all that different in 2016 between the pollsters. Based on my own assumptions about the demographic outlook (tied exclusively to historical averages and not opinion) I was able to use cross tabs (and only cross tabs) guess the final result within a couple of tenths of a percentage point. Moreover, my spreadsheet did not move up and down all that much over the course of time. It never got any higher than about a four point Hillary lead, even when she was leading by double that in national averages.
We are already seeing this separation in many of the State polling (as pointed out in my opening numbers). You see many pollsters showing the race very close (as it was in 2016) and then you will see many others showing Biden with large leads. The answer, however, is likely not to take an "average" of those two subsets. Accuracy, in this case, will be to figure out which subset will be right, and plan accordingly.