Even aggregating polls as RCP does doesn't help if you include, as RCP does garbage polls. "Garbage in, garbage out" as the saying in survey research goes. The Cook Political Report swingers poll is an example (and is included in RCP's averaging): stale when published, small sample sizes, extravagant margins of error, each result within the margin. Meaningless! If you include polls with historically demonstrable bias, like Rasmussen, you're including biased garbage. RCP does that, too.
And if you're a bright, shiny thing and decide that you can correct for bias, as Nate Bronze does, then you're going to skew the averages for being too clever by half.
I have thought for some time, and have experience, albeit very remote, in doing polls, that the safest path to get the most accurate read is to pay attention to the highest-quality polls only. But in our 15-minutes-is-an-eternity times we political junkies, and those who care about the continuation of the experiment called Democracy in America, grasp at flotsam and jetsam as a drowning man. Even with high-quality polls you have to be parsimonious. There's a difference between HQ and highest quality. When NYT got rid of Nate Bronze they subbed in Nate Conehead and teamed up with Siena College to conduct state polls. The Times/Siena polls are universally considered high quality. Last cycle the last poll in Iowa that Times/Siena conducted showed the state in play for Joe Biden. Then the gold-standard Iowa Poll came out. It showed trumpie with, from memory, an 8-point lead. Final result in November, trumpie by 8 (or whatever it was). They hit it on the dot.
I willingly show my ignorance--This is all prelude, by the way, I haven't gotten to the main point of this post yet--here on another polling matter. It was my understanding that you cannot average away the margins of error of different polls conducted at different times by different firms by lumping them all together. Example: Cook's swing polls had a M.o.E. of 4.7%-4.9% depending on the state surveyed. Cook took each of those individual state polls (conducted by two firms, one Dem, one Rep, joined as one (?)) and lumped the survey sample sizes together to calculate one average all-swings poll with an infinitesimal M.o.E.--and their average all-swings result was still within that M.o.E. I say I am showing my ignorance because all poll aggregators do this averaging so I must be wrong.
Okay, so that's the intro. Now to the main subject. Good Chief WaPo has his own bright young thing who aggregates polls too. His name is whatever his name is. This individual's calling card however says that he only averages the highest quality polls, thereby avoiding the RCP garbage, and he does not attempt to correct for house bias, Nate Bronze's conceit.
...we have chosen to use only the highest quality polls for our model, and not many polls that were released in the last few weeks met our standard.
He acknowledges that,
...the polls underestimated Trump in the last two presidential elections...
But that is as interpretative caveat, he does not attempt to correct for it.
But it’s important to remember that polling errors can go either way.
And then caveats the caveat by citing to 2012 polls,
...(not a cycle we usually associate with an error at all)...
I mean, dude. 2012? Mitt Romney and Barack Obama? 2012 is ancient history. You can't caveat your caveat to 2012 in 2024 when we have the same candidate in '24 who caused the underestimation in '20 and '16.
Using these excellent methods, Chief WaPo's guy has trumpie leading in five of the seven swings-- including Meeshagan (!)--with Kamala leading only in Pennsylvania and Wisconsin. Ergo, doom and gloom:
...if you count up the electoral votes and award them to the candidates leading in those states, Trump comes in at 283 and Harris at 255.
BUT:
According to our model now, Harris has become the slight favorite. ...
The reason Harris is now the favorite is because Harris has closed the gap with Trump in Sun Belt states enough to open a second path to the presidency.
"Closed the gap": That means Kamala still trails in the Sun Belt, which WaPo's guy's polls show: -3 in North Carolina, -3 in Nevada, -3 in Arizona. Note to WaPo brave: electoral votes are not distributed within states proportionately, it's winner-take-all. The "second path" for Kamala leads to DEFEAT at this particular moment in the rent fabric of space-time.
It really shouldn't be this hard but we lumpenproletariat must needs do what we will not do on accounta we're addicted to the polling narcotic: research the best polls by polling firm (I like Fox), choose the polls that canvass likely, as opposed to registered, voters, and have the lower margins of error and caveat that trumpie will be underestimated if he polls less than 46% and overestimated if he polls 48%.
