For the first time since 538 published our presidential election forecast for Vice President Kamala Harris and former President Donald Trump, Trump has taken the lead (if a very small one) over Harris. As of 3 p.m. Eastern on Oct. 18, our model gives Trump a 52-in-100 chance of winning the majority of Electoral College votes. The model gives Harris a 48-in-100 chance.
My reading of Nate Silver’s article suggests that the OC (original commenter’s) comment is right though. Quote from the article:
the movement could just be random variation in the polls — if Harris really is ahead nationally by 3 points and in the Blue Wall states by about 1 point we’d expect her to have better and worse weeks that vary around that average.
Sounds like phantom momentum to me. And Nate also agrees with the part about there being Republican bias in the polls,
First, are polls from Republican-aligned firms more favorable to Trump this cycle? Yes, … Harris is ahead by 3.0 points nationally in this simple average. But when you look at only Republican-aligned firms, she’s up by only 2.0 points. Removing those polls from the average brings Harris up to a 3.4 point national lead. These aren’t huge differences, but they’re not nothing. Combined with a similar pattern in state-level averages, polls from Republican-leaning firms could push polling averages — and by extension forecasts — rightward.
This last sentence is important so I’ll repeat it,
Combined with a similar pattern in state-level averages, polls from Republican-leaning firms could push polling averages — and by extension forecasts — rightward.
Of course Nate believes, as you state, that he’s able to account for it by adjusting for the house effects and such. Which would overcome the flood.
He then seems to go on and justify that his house effects are accurate by comparing with pollsters whose averages are excluding Republican polls (thus avoiding the bias completely) and saying that he winds up with the same result as them.
However, it’s really interesting to note that most of the polling averages he compares with don’t include as many GOP polls.
In fact it’s Nate’s own average that is the lowest in favor of Harris. In fact I think 538 is the only other one that does even include those GOP polls.
And somehow these are the ones that show the GOP candidate with a lead.
In fact, VoteHub - the one using only high quality nonpartisan polls - actually has Harris winning the Electoral College currently., 270 vs 268: https://polls.votehub.com/
Electoral College average
Harris 270, Trump 268
National Average
Harris +2.3
Tipping Point (MI)
Harris +0.1
Electoral College Bias
R +2.2
Now Nate can easily justify this as a tiny difference within the margin of error. And he’s be right, of course. But I feel that this shows, even after all the hard and brilliant work by Nate and folks, the flooding by the GOP polls seem to be off by just enough to push things over the edge. Ignore them for more accurate data, and the picture looks different.
This might be the best comment reply I’ve had on lemmy. When I’m not half cut coming home from soccer, I will dig into this and either agree or give you a worthy response.
Pleased to have been able to contribute. And, no rush here, but happily awaiting your reply - either way I’m bound to learn something new.
Hi there! Sorry for the delay, I rarely Lemmy on weekends. And made the foolish mistake of posting something that took off towards the end of last week.
Anyhow, I think your analysis is really interesting! Admittedly, I’m a cheapskate and haven’t paid for the Silver Bulletin, so I’m quite curious to see what Silver’s model would have done taking only the same polls as VotHub does.
That being said, I think that even the VoteHub models are showing the same sort of movement in the polls, as recently as Oct 6 VoteHub showed the two candidates seperated by almost 3.5% nationally, which is pretty significant. However, even their model now shows Harris’ lead cut by half, down to 1.7% And this is from a model that only accepts non partisan polling etc! Looking over their recent polls, of the four polls in the 3 Blue Wall states in the last week, none showed Harris ahead. (At least on the 20th, there was one that showed Harris up by 4 in Michigan.)
Now for the part that would fascinate me about the Silver/VoteHub model comparison (I’m not going to lie, this conversation is making me want to drop the $30ish, just to see and of course for election night/week/month) is that even using the exact same inputs, for a race this close, you could easily see different outcomes by different modelers. Some put more weight to the possibility of uncorrelated polling errors between states, others between correlated polling misses by certain demographics (and of course, models might split those demographics differently, e.g., figuring that a Florida Latino may be more likely to be Cuban than of Mexican heritage and thus respond very differently to accusations of socialism etc.) All that to say, you could very easily put the exact same polls into different models and you ought to see somewhat different projections.
Again, I really appreciate you diving into the piece and coming back with solid evidence, you are exactly the type of person making Lemmy a place I want to hang out.
