As of a week ago, FiveThirtyEight's polling averages included 19 polls that were conducted entirely in October. Of those 11 were from right-wing firms and news outlets, 7 were from non-ideological pollsters, and 1 was from a Brazilian pollster that's not well known. One of the right-wing pollsters included in that average, Quantus Insights, responded "you're welcome" on Twitter when a MAGA influencer celebrated the 538 polling average shifting to a slight Trump lead.
You may not have noticed, but tons of right-wing polls have come out in the last month or so, way more of them, in fact, than non-ideological polls, and they've got Trump performing significantly better than non-right-wing polls, Harris leads by about 1.7% in the average of non-ideological polls while Trump leads by 1% in the right-wing polls. We've actually seen this before, in 2022 when a huge raft of right-wing polls came out in October, just before the election, and showed GOP Senate and House candidates doing much better than other polling and, in fact, much better than they actually did when the votes were counted.
Now, poll aggregators like FiveThirtyEight will tell you that they're able to weight for the kind of thing and that it doesn't make a huge deal in their models, but that really depends on what you consider a "huge deal". FiveThirtyEight, for example, calculates that the right-wing polls shift their average by between 0.1% and 0.8%, almost entirely toward Republicans and, look, I get it, from a data perspective, you'd always rather have more data than less. Even if it's bad data, weighting and adjusting can allow you to still get some reasonable value out of it. From a statistical analysis perspective that's pretty much negligible, but most people reading these polling aggregators, even other journalists, aren't all that educated in statistical analysis.
Still, we have to acknowledge what it does at the most basic level. That 0.8% shift was in Pennsylvania where 538 currently has Trump ahead by 0.3%. In other words, without the flood of right-wing polls, they would instead have Harris ahead by 0.5%. And look, again, I get it, from a statistical point of view, both of those results are toss-ups, well within the margin of error, and you can't make any solid conclusions about the state of the race from either of those results except that we don't know who's going to win. But look at the media coverage.
Even reputable news outlets are writing stories about Trump being ahead in the swing states and right-wing news media is all but crowing it from the rooftops. Whether they like it or not, poll aggregators are now driving news coverage that completely ignores what their results actually mean.
The long-story-short here is that it appears likely that polling averages are being manipulated by a flood of right-wing polls that look far better for Trump than non-ideological polls do. Whether this is to drive favorable news coverage or to provide an excuse to later challenge unfavorable election results is unclear, but what is clear is that this isn't just data that's bad because someone did some shoddy work, it's deliberately bad and it's designed to throw off the aggregate data.
If that's not a good enough reason to question whether it should be part of the aggregate data and to question the results of aggregate data that includes it, I don't know what is.
Hat tip to The New Republic whose story I got most of this data from.