Is this a weighted or unweighted poll?
I’m curious. I know it uses a large number of calls—something like 3000. If your call sample is large enough, you shouldn’t have to cook the results for predicted poll representation (absent some other systematic polling error).
In that case, it could be the most accurate of polls.
And, it's not a telephone survey. It's a tracking poll of about 3200 people (randomly selected) conducted via the web. Each day, 1/7th of them are asked to answer the questions.
It's a rolling average of the past 7 days.
The details are here: https://uas.usc.edu/documents/uas/UAS%20Weighting%20Procedures.pdf
“D” + 8%, and they include all outliers. Nontheless they claim a provable “95% Confidence Factor” for all results outside the MOE.