Do the polls show a health bounce? Looking at 538’s polls-only national polling trend, I cannot discern one. Maybe there’s an inflection point around September 12, but it preceded the polling coming in with samples that included September 11. And if it’s an inflection point, it’s barely distinguishable from a half-dozen others in the slow creep going back to mid-August.
Why isn’t the better analysis that Republicans are returning to Trump after a rough July?
That’s possible too. We’ll see.
Nate Silver just tweeted that there looks like a very small (1 point) shift in Hillary’s direction, nationally, in the last week.
If you aggregate all the samples from those polls when calculating MOE, it’s still probably bigger than net 1.
New Monmouth poll, rated A+ by 538, just came out putting Clinton +5 in Florida. Taken Sep 16-19th, making it well post 9/11 collapse and pre-Trump Institute shenanigans story.
I know, I know. Don’t get worked up about individual polls.
It’ll be interesting to see what that does to the FLA map.
538 rates Monmouth A+ and adding a +5 to this:
9/14 Clinton +1
9/9 Clinton +2
9/12 Trump +3
9/7 Tie
Should be interesting.
I don’t think the Monmouth FL poll is brand new – I think it was already worked into 538. And I think it’s the one I was thinking of earlier, not the St. Leo poll.
That’s correct. They put the Monmouth FL poll in yesterday.
Does anyone know anything about St. Leo? Judging from the adjustments made 538 seems to believe they’ve been leaning overly blue, but do they have any type of history that makes them reliable once adjusted?
I guess I have to just trust him but it’s weird; on the Florida polling:
his “adjusted” number for A+ rated Monmouth, with a “mean reverted bias D+.9”, moves it from Clinton+5 down to Clinton +4
but his “adjusted” number for A rated Siena, with a “mean reverted bias R+.3”, moves it from Clinton +1 to Trump +1.
Edit:Sorry, Monmouth is only D+.6
It has nothing to do with the pollster rating. It does involve house effects, but there are also a half-dozen other factors. They explain the methodology on the page, I think.
But the adjustment is based on 538’s judgment of their bias, no? But for one poll there’s a full point adjustment against this percieved .6 bias and in the other there’s a two point adjustment in the same direction as this .3 bias.
538 today has Nevada as extremely pale pink, Florida as only slightly more pink, and North Carolina and Ohio as less pale pink, but still not red. Surprisingly, Rhode Island is only pale blue.
Sorry I meant to add, I know 538 also makes adjustments to “equalize” polls wrt likely voters/registered voters and 4-way/2-way but I think both of the ones I mentioned were the same in that respect. Basically, I am squinting hard at his adjustment of the Siena numbers.
There’s also a “trend line adjustment,” and “house effects adjustment”–the latter being independent from the pollster’s rating.
The trend line adjustment is described as follows:
The house effects adjustment, afaik however, is exactly what the mean reverted bias is about, which I already talked about. So we are left with the trendline adjustment being the sole culprit for a rather large difference in how these polls, taken within a week of each other, were adjusted.
I don’t really understand your criticism.
We have two polls. Assuming the “mean reverted bias” is the same as house effect, then we know that piece of the adjustment for each (.3R and .6D). Assuming the likely voter and third party adjustments are the same, and that neither poll reflects a convention bounce, that leaves trend line to explain the differential adjustments.
The adjustments, breaking out the house effect, are:
An additional 2.3 for Trump (9/10-9/14)
An additional .4 for Trump (9/16-9/19)
This implies that the trend line adjustment for 9/10-9/14 was +2.3 Trump, and the trend line adjustment for 9/16/9/19 was +.4 for Trump. Why do we think that’s wrong? Also note that they round the adjustment numbers, so the difference might be even smaller than it seems.
Well I think it might be “wrong” because 2.3 is a lot more than .4 in this context, i.e. that the trendline is given such a weight as to give such different adjustments to polls taken within 10 days of each other. I’m sure he’s got more math on his side than me but it just strikes me as dubious.
With rounding artifacts, there could be less than a point difference between them, which doesn’t strike me as an obvious impeachment of the method since the polling periods didn’t even overlap.
But I have to confess that I don’t really understand the trendline adjustment or why it’s beneficial.