In my review of the Silver Bulletin model, I focused on the weirdest and largest adjustment Nate Silver makes to the raw polling numbers: the “convention bounce”, and argued that it is the single most obvious flaw in his model. I am - to put it mildly - not alone in this opinion.
But I’ve now come to think that the adjustments are both weirder, and less transparent than most realise - including me, until a few days ago.
Not only is there an adjustment which - according to the published methodology - subtracts a few percentage points from each candidate after their convention and then diminishes as we get further out, but there is another adjustment too. One that gives a positive effect to Trump (never to Harris) and is growing with time. And this one is - I think - undocumented.
Let’s look at what this bounce adjustment does - here are two snapshots of Silver’s “bounce” adjustments: from the first time these numbers were published on 1 September, to the most recent.1
The explanation of the “convention bounce” adjustment has been given many times - here’s Silver' describing it, somewhat wearily.2
So, the convention bounce adjustment should be a negative adjustment to Harris. And you can see this, all the adjustments are negative for Harris on both dates (blue bars), with the adjustments coming down on the second read on the 10th (coming down at different rates, but this can be put down to the dates of the polling).3 You could also expect a small residual negative effect for Trump - but much smaller - because the Republican convention was longer ago, in July. And you can indeed see that in a few states.
But there are also positive effects for Trump. And that’s less expected. According to the multiple descriptions Silver has given, there should be no positive effects from a convention adjustment - just a “shave” or “haircut” on the polling percentages for each candidate. And yet there these adjustments are.
What’s more, these positive adjustments for Trump are growing over time. By the time of the second read, he is benefitting from a positive adjustment in the vast majority of states. And what is more, for most of the states - and in particular in swing states - this mysterious adjustment is growing faster than the convention bounce is shrinking.