Partisan nonresponse bias is far more nuanced than people realize...
Here's what we see when we isolate SMS survey respondents in red and swing Senate states this cycle and match them to their modeled party in the voter file.
It doesn't take a rocket scientist to see that Democrats are responding at a higher clip than Republicans in 9 of the 10 most recent states we surveyed.
Notably, those are the same states where Republicans have beaten the polls in 43 of 58 races since 2018.
Last year, we at Quantus, along with the rest of the polling industry, missed big in New Jersey due to the inverse effect where engaged Republicans were more likely to respond to surveys and taint the swing / lean cells in the opposite direction.
New Jersey Republicans answered at a higher rate last year and still lost the Governor's race by 14.
The reality is that the small, highly engaged slice of each party that responds to text surveys tells you very little about statewide turnout and the depth of a candidate's support.
Oh, and this nonresponse bias is actually worse in the cheaper collection modes like SMS and online panels too, which is what a large amount of public polling has shifted to for survey collection over the last 10+ years.
Translation: Frequent and reliably Democratic voters (voted D in 4 of 4 elections) are answering polls much more often than frequent Republican voters.
Some pollsters have compensated for this since 2016, but many still havenβt β and that will inevitably skew polling averages.