That Ain’t Cricket

On October 9th, Wikipedia posted six statewide polls by two Republican polling firms — Fabrizio, Lee & Associates and McLaughlin & Associates. Fabrizio is Trump’s pollster.

All six polls were of likely voters; all six had 3.5% margins of error; and all six were to Trump’s advantage — a five point lead in Georgia, three points in Nevada, and one point leads Michigan, North Carolina, Pennsylvania and Wisconsin.

Fabrizio’s poll results must have brought tears of joy to Trump’s cheeks. He was getting what he paid for — the ego gratification that he was winning, winning, winning.

Or was Trump being played?

My earlier discussion of “weighting on recall vote” is easily recalled. The polls are being adjusted to match the actual results from four years ago. Yikes!

And yet, the margins back then do not reflect the margins now. Back then, the margins were +11,800 in GA, +33,600 in NV, +154,000 in MI, -79,000 in NC, +80,500 in PA, and +20,700 in WI.

So, with this new weighting on recall vote, the battleground states all flip to Trump? Neat trick if you can pull off. 

And how convenient for his pollster. It must mean a bigger payday? Or the polling equivalent of taking a Sharpie to a hurricane map??

NYT’s Nate Cohen hasn’t explained who does gold standard polls and who does gold-digger polls. So we’re left with guess work.

Here’s mine: The initial  idea was that the pollsters somehow missed a hidden vote for Trump, mostly from the non-battleground in 2020. The “weighting on recall vote” was suppose to correct for that hidden vote. Instead, it allows pollsters to craft a fudge factor — that Sharpie thing, again — by expanding the margins to meet the candidate’s fantasies.

How else does Trump’s less than one percent loss in Georgia turn into a five point win? Are tens of thousands of Georgians ‘remembering’ they voted for him… when they didn’t vote at all?

Until the pollsters publish their four or five step weighting processes, we’d better stick with election statistics. That way Kamala Harris wins every battleground state except North Carolina. And she wins all sixteen of the mega-Democratic states not named Texas, Florida, Ohio, or Missouri.

But the real problem with “weighting on recall vote” is that adding long-term recall — four years after the votes were cast  — increases the average absolute error by at least 9 percent. That’s according to Australian political scientists whose long-term recall experiment covered 33 months, not 48 months. The Sharpie thing strikes,  again and again.

If the accuracy of the voters recall declines in a linear fashion, then 48 months ought to impose a 13 percent average absolute error rate. 

Whenever we try to correct polling errors, particularly non-existent ones, we make the problem much, much worse. So I’m sticking with election statistics and I don’t care what Trump’s pollster uses to keep his client happy. 

But whatever he micro-dosed him with this past week had him swinging add swaying with Kristi Noem, the dog killer, for 39 minutes.

And that’s just ain’t Cricket.