Thanks.
So, the systematic error I worried about lies in whether people who are reachable through an internet poll might be more likely to vote Democrat or Republican. To be determined.
Actually, they do address that. They provide a tablet and broadband access if necessary, once a respondent is selected.
More info here: https://uasdata.usc.edu/recruitmentoverview/1
This poll could have a response bias as it is requires a bit more work on the part of a respondent to register, compared to a one time phone survey, perhaps balanced by the financial reward (the reward would also be a potential issue). They then continue to receive a financial reward each time they participate (which would be 20 dollars once a week).
They allow all members 18+ from a survey’s household to participate. This will multiply the financial reward and could lead to households with more than 1 member being more likely to register to begin with.
In any case the trends are great, any polling is imperfect, and traditional phone surveys have their own issues. It’s also interesting because it tracks the same people over time, once per week, rather than being a new random sample each day. I’m looking forward to a Trump victory and a liberal meltdown!