Another bit to add, specific to the Simon Rosenberg bit.
The 538 article about how the ‘22 polls were very good does discuss the Red wave narrative in the media and how right-aligned polls fed it.
Some of the media certainly did see those polls and use them to tell that narrative. But they did NOT have an impact skewing the aggregations far off rightward. Overall the lean of the error ended up very slightly to the Democratic side in fact.
And those pollsters poor performance in that cycle are now taken into account in how they are weighted going forward.
Specifically regarding 2024 general election considerations at the “individual state polling” level, Rosenberg’s point is subtler than the bolded. His point regarding the here-and-now is that a glut Republican pollsters are skewing the NC and PA general election polling average further rightward than what they should be. Whether the average leans left or right is not material to his point. Whether a Harris +0.5 (for the sake of an example) in NC/PA should be a Harris +2.5 is material to his point.
In yesterday’s column, Rosenberg is specific that seven Republican pollsters have “flooded the zone” in recent North Carolina polling. These seven are viewable on 538’s North Carolina general election page – might have to scoll down and press the “View more polls” button once or twice.
Sept 1-3 - Patriot Polling (Trump +3)
Sept 11-12 - Trending Politics (Trump +2)
Sept 11-12 - Trafalgar Group (Trump +2)
Sept 11-13 - American Greatness (Trump +3)
Sept 11-17 - Fabrizio, Lee & Associates/AARP (Trump +3)
FiveThirtyEight does not list Fabrizio & Lee as a Republican-partisan pollster, and I would expect that this oversight is not a lone instance. IMHO, the bias weighting used by poll aggregators like FiveThirtyEight is a blunt, imperfect tool at best.
Of course it is. Given that the “skew” goes one direction or the other as much as it does from election cycle to election cycle, aggregation is a very blunt imperfect tool. Trying to “unskew” though is almost certainly a recipe for making it worse.
FWIW if you’re having to click down several times to get to polls you’re hitting ones that have lost their weight just by losing recency bias. 538 has NC Trump+less than 1. Pretty much what it should be looking only at NYT, Marist, and MC. If there is an impact of that “flooding” it is pretty minimal.
I’d personally be curious if median polling performed less poorly, eliminating outliers on both sides. Doubt it though.
Jay Kuo, Morning Consult and (really) Fox News Polling to the rescue!
Kuo covers a lot in the piece linked below, but this is an unsung takeaway:
Another thing that has got to worry the Trump campaign is party loyalty. As Fox reported, “By a 4-percentage point margin, more Democrats back Harris than Republicans back Trump. One in five non-MAGA Republicans supports Harris(my emphasis - b).”
Read that last sentence again. Hello, Never Trumpers and Haley voters!
Could this poll be an outlier? It’s technically not . Many other high quality pollsters now put the race in Georgia at even to slightly favoring Harris. While this doesn’t mean she will win the state (we should never assume that when it’s this close), it does mean that her campaign should continue to put Trump on the defensive there and charge toward the finish line full out.
1006 total polled. I can’t find where they determined who was a “non-MAGA Republican” but given that previous polling puts the vast majority of Republicans as MAGA, let’s be generous and say that is 20% of self identified Republicans. “Non-MAGA Republicans” gets to a very very small n.
Bottom line in the cross tabs is that 92% of Republicans choose Trump, 5% Harris, the remainder Oliver and Stein. Not a big surge.
Yes only 1% of Democrats say Trump with West and Stein each getting 1%.
I can’t pinpoint the bolded, either, although it might not be derivable from what’s publicly released in the crosstabs. However, in trying to sort this out, I did run across some other interesting information. From your link (margin of error for both subgroups is 4.5% as shown on page 1 of your link):
Nearly one-in-five (19%) of self-professed conservatives (not necessarily Republicans) plan to vote for Harris. That, to me, actually seems a lot more impressive than “one in five non-MAGA Republicans”.
Six percent of 2020 Trump voters plan to vote for Harris. Even if you assume the entire margin-of-error hit and knock that down to 1.5%, that’s still over 36,900 former Trump voters straight-up flipping to Harris – not staying home or voting third party. Adjust that 36,900 figure down, if desired, to account for lowered voter turnout.
Of course 4% of those who voted for Biden say they are voting for Trump. So net maybe a 2% difference, clearly swamped by MOE.
I get that my picking at the insignificance of the numbers can be annoying, but we really do need to be honest with ourselves and not squint until we see what looks good.
A poll that is up 3% in Georgia is great to have. Better than it not existing. With the polls we have there very well may be some very nice surprises! And there may very well be some very nasty surprises! Both are very true statements. I’m expecting the nice side for a variety of reasons and will be horrified but not shocked by nasty ones.
Regarding Republican-poll zone flooding … the way it’s done is not that one pollster drops a bunch of polls in quick succession. Instead, it’s a battery of pollsters - like the separate seven pollsters thinning out Harris’ aggregate NC number, shown above.
But why? It’s not like the polls actually change the election result. It certainly didn’t help them in 2020 (if in fact it was a conscious effort to “flood the zone”). Is to generate (or, conversely, deflate) enthusiasm? Drive some sort of press narrative? Support a “Democrats stole it” messaging?
Ultimately no amount of “red wave” polling is going to change the overall narrative (and likely reality) of a very close election. If all of those “biased” pollsters disappeared the overall story would be very much the same - perhaps more like Clinton in 2016 than the toss-up scenario we have now but not substantively different.
I guess it depends on one’s media-consumption silo … but in the stuff I’m exposed to, Republican-poll zone flooding is not the least bit questioned or controversial.
This one pertains to the upcoming Washington state governor’s race:
And yet the 538 NC number is right where it would be if the only polls considered were the most Marist, Morning Consult, and NYT/Sienna polls.
If the goal of this “flooding” is to impact that number, to skew the polling “average” then it is doing a very poor job of it. (RCP possibly the exception as they seem to actually do averages, rather than aggregate with any sophistication. But even they are only impacted slightly.)
Yes, there are people on each side who will glom on to the poll that tells a story - one that panics them or one that sells clicks or one that might drive excitement and funding because suddenly it is close! (A single poll.)
Sure just as your silo is sure that there is rightward skewing going on, the other silo only reads the polls they like. They may think Trump is doing better than the aggregate results would say. And he may be. If these rightward polls are supposed to drive fundraising though, they have failed there too. Trump’s fundraising is way down compared to past cycles and compared to Harris, including in small donation percentage.
Will those in that silo be more likely to believe a Trump stolen election lie because the polls they saw said he was going to win? Nah. They would believe it no matter what polling had been saying, Trump said it, they “heard” the polls were skewed to Harris, the fix was in, whatever.
That Washington state bit was from a poll on March??? All I can say is that Ferguson has been ahead in every poll listed on 538 since. So not very effective flooding. And a lightly polled race line that one would be the easiest sort to impact.
Keep in mind that the 538 article about 2022 polls being historically accurate included zero polls from September.
“We analyzed virtually all polls conducted in the final 21 days1 before every presidential, U.S. Senate, U.S. House and gubernatorial general election, and every presidential primary, since 1998,2 using three lenses — error, “calls” and statistical bias — to conclude that 2022 was a banner year for polling.”
So if right-leaning pollsters were putting their thumbs on the scale in 2022 to manipulate the media from June through September, it wouldn’t show up in 538’s measure of 2022 accuracy.
True. We can however look to see if its forecast had been far off earlier on looking here. It really wasn’t.
Eyeballing it they went with the House popular vote being R+6 in June, gradually down to maybe R+2.5 in October before popping back to R+4 final week or so. Final was 51 to 48%, R+3. Nothing that looks like an R skew along the way.
Their Senate numbers had a clear D skew most of the cycle, from mid August into October anyway, only forecasting a narrow R majority in the final weeks.
Nah. No evidence that there was successful GOP skewing earlier in the cycle. If they were trying, and maybe they were, they were really bad at it.
