Read on Twitter
Let's talk about what's happening under the hood on some of the poll averages right now https://twitter.com/NilesGApol/status/1291583616867274755?s=20
RCP is a particularly peculiar case. Yeah, it's a simple polling average, but even a simple poll average would include the Morning Consult and Ipsos polls. There's no plausible reason to exclude those but allow Rasmussen and HarrisX or whatever.
So, even a 'simple' RCP-like average should obviously be better for Biden than this. But even if we add in all the recent polls to make a better RCP-like average, we'd still find Biden doing worse than earlier in the month.
One reason is a change in the composition of recent polls. There's just one live interview national survey among the last *50* recorded by FiveThirtyEight.
And in June/July, live interview surveys were quite a bit better for Biden than the others.
And in June/July, live interview surveys were quite a bit better for Biden than the others.
Here, for instance, is what the poll averages look like if you break the polls down by method. Black = all; blue = live; red = online; green = IVR.
The live interview average is still at Biden+9.4, with the drop below 10 attributable to the one live poll of the last two weeks
The live interview average is still at Biden+9.4, with the drop below 10 attributable to the one live poll of the last two weeks
The rigorous polling averages--like the one you see here, or FiveThirtyEight--do give more weight to the live interview polls. But there's only so much you can do.
Despite those efforts, IVR polls would probably still have more weight in the average--the live polls are too old
Despite those efforts, IVR polls would probably still have more weight in the average--the live polls are too old
You can see that we have these moments in mid-July and mid-June when the live polls rise to nearly match the online polls. But right now, they're only about 15% of the estimate, even though they're considered the best polls.
Another thing that rigorous averages do is adjust for 'house effects'--persistent bias in a poll.
But these house effects don't go as far as you might think, since they're calculated with respect to the quality weighted average--which still includes a *ton* of non-live polls
But these house effects don't go as far as you might think, since they're calculated with respect to the quality weighted average--which still includes a *ton* of non-live polls
As a result, it's not like the House Effect drag Rasmussen all the way back to the live polls. Instead, it drags Rasmussen back toward the average, which is something more like a Morning Consult and YouGov.
Similarly, a Monmouth poll would be adjusted back to the right.
Similarly, a Monmouth poll would be adjusted back to the right.
Here, for instance, are House Effects by 538 pollster rating (where 0 is their "A+" and 1 is an "D-")
As you can see, there's a decent relationship between pollster quality and a Biden House Effect. The small number of A+ polls means they can't anchor the adjustment.
As you can see, there's a decent relationship between pollster quality and a Biden House Effect. The small number of A+ polls means they can't anchor the adjustment.
Now I'm not saying that the House Effects *should* be calibrated at A+ = true and adjust all others to it.
My main point is that if there were as many Monmouth polls as Morning Consult or something, you'd have a different House Effect adjustment too
My main point is that if there were as many Monmouth polls as Morning Consult or something, you'd have a different House Effect adjustment too
At the same time, none of this means that there *hasn't* been tightening. If we go back to this chart, there's still tightening across all mode. https://twitter.com/Nate_Cohn/status/1291731444411367425?s=20
But the online polls show *very* little movement, and there are really no live interview polls at all--that big drop is mainly due to one poll, and may or may not hold up with more data. The real movement here is on the IVR side, both in terms of relative weight and trendline
So while I think the data is consistent with tightening, I don't think it disproves stability either. It is entirely possible that another round of round of ABC, Monmouth, CNN type national polls would show double-digit leads for Biden and move the average quickly back his way
The live interview state polls are also rare, and they don't add a ton of extra clarity. IMO, two Monmouth polls of GA/IA were maybe more like Biden+8 than their prior Biden+12 national result. Those Quinnipiac polls, OTOH, sure look like their last Biden+15 poll.
Anyway the polling drought will come to an end. Virtually all will want to weigh in before the DNC, which starts 8/17. In fact, that's the reason for the drought: if you're polling next week, you wouldn't poll over the last few.
So I'd guess we've got a lot coming.
So I'd guess we've got a lot coming.
