As I have noted elsewhere, Republicans seem to be hard at work manipulating the media and trying to suppress Dem enthusiasm by claiming premature victory. One tool of choice seems to be putting out poll after poll of battleground races with pretty hard-right leans. And I’m not the only one who has noticed this:
But how bad is it?
Well, I took a look at every Senate, Governor and House poll in the 538 database in the 3 months leading up to November 8. For each, I excluded it and computed what the other pollsters looking at the same race said at the same time (3 week running average) and computed the relative bias.
To be clear, just because Trafalgar, say, is 5 points more republican than the “average” doesn’t mean that they’re wrong. However, it does mean that a bunch of Trafalgar polls thrown into the mix are going to produce a very rightward swing. And the rightwing pollsters poll a lot.
Plotted above is the relative bias (sometimes called “house effect”, where negative means more republicans, obviously) as a function of how many polls they have in the database. The most prolific are also among the most right-leaning. To wit:
Pollster
|
Number of Polls
|
bias
|
Trafalgar Group
|
48
|
-5.09
|
Emerson College Polling Society
|
42
|
-2.34
|
Data for Progress
|
30
|
-1.75
|
Cygnal Political
|
28
|
-2.07
|
InsiderAdvantage
|
27
|
-2.9
|
Note that Trafalgar, Cygnal and InsiderAdvantage are all Republican oriented pollsters (whether or not 538 labels them as such), so not only are they right-leaning, they have a serious RW agenda. Data for Progress is nominally dem-aligned, so I don’t know what to tell you about that.
Ranked by most RW (in terms of bias) we get:
Pollster
|
Number of Polls
|
bias
|
Wick
|
10
|
-5.46
|
Trafalgar Group
|
48
|
-5.09
|
Rasmussen (Pulse Opinion Research)
|
6
|
-4.95
|
co/efficient
|
11
|
-4.03
|
While I don’t know the political orientation of Wick, they’ve jumped into the fray to post some truly bizarro outliers. Co/efficient seems to be Republican oriented, working with “Big Dog Strategies,” “Missouri First Action” and republican candidates.
As for the middle of the pack (relatively unbiased), we get a bunch of nationally-recogznied, unaffiliated pollsters like Suffolk, Marist, YouGov, Siena, and so on:
Pollster
|
Number of Polls
|
bias
|
YouGov
|
16
|
-1.84
|
Data for Progress
|
30
|
-1.75
|
Beacon Research/Shaw & Company
|
11
|
-1.22
|
Suffolk
|
18
|
-0.6
|
Susquehanna
|
5
|
-0.54
|
Echelon Insights
|
19
|
0.04
|
Siena
|
15
|
0.15
|
Marist
|
16
|
0.74
|
OH Predictive Insights / MBQF
|
5
|
1.17
|
SSRS
|
9
|
1.33
|
Amber Integrated
|
6
|
1.37
|
PPP
|
13
|
1.61
|
Again, this can’t be stated strongly enough. It may be that Trafalgar et al. are the most accurate, but I’d be surprised if they were totally unbiased (in a statistical sense). Even in 2020, and during pretty much all of the last 5 years, Trafalgar was 2-3 points to the right of the final outcomes. It would be strange if, after getting all of these RW clients, they adjusted their algorithm to be more Dem friendly.
But all that aside, recognize that a lot of the recent “shift” in the polling isn’t due to a change in fundamentals or even a change in public response; it’s just a change to which pollsters are being averaged in.