An email/text poll is a terrible methodology. Reputable pollsters use a mix of cellphone/landline/online outreach to get as accurate of a read as possible, because they're more likely to reach more voters, and a more diverse spectrum of voters.
You can't build a poll around the assumption that everyone has a cellphone or internet access.
So I guarantee that this poll has under-sampled likely voters over age 65, and the A/A vote. You know, the voters most likely to favor Biden. Having to use RIM modeling, "third-party voter-file data" and other exit polls to "correct" your poll's self-admitted bias toward white, highly-educated voters only makes the poll more suspect, not less.
Yes, the poll admits that it's raw sample was biased toward white, highly-educated voters:
The weighting efficiency was primarily driven downward by the raw sample being tilted towards those with a higher education, and slightly tilted towards white voters. RIM models correct for this by appropriately weighting groups that are under or overrepresented in the sample. Population estimates for the demographics were obtained through a combination of third-party provided voter-file data and exit polls.
Mm hm. Well, if you don't name those "third-party" and exit poll sources, it's difficult to know if the bias was accounted for properly.
I downloaded their spreadsheets and will take a closer look at the full data for representative sample group percentages and the wording of their questions. They could have done a decent enough job, but that their results are so out of whack with what known reputable pollsters like UNF are finding doesn't give me confidence in this firm.