• 13 Posts
  • 2.5K Comments
Joined 2 years ago
cake
Cake day: August 22nd, 2023

help-circle











  • There’s no real solution for selection bias if you don’t have other respondents of that group. With something like race or education, you have their demographics and can upsample those that do respond. But it the group is specifically defined by not wanting to respond to polls and that comes with biases to the poll questions, you don’t have anything to upsample.

    Now whether such a group is really a distinct entity out there that can’t be kind of approximated by people who share other traits is the question. If white conservatives have a spectrum of trust in pollsters and the non-responders would just answer questions the same you’re fine. But it those with low trust are also more anti-vax or some sort of distinct population like an insular community, you couldn’t just approximate them with people who did respond.



  • Swing voters are not really the sole political deciders. They matter extra because they effectively count as two votes, but base turnout is often a larger effect than the actual swing voters.

    About 15% of Biden’s voters did not vote, 5% switched to Trump and 1% voted for someone else. That’s compared to 11% of 2020 Trump voters, who sat it out, 3% who switched to Harris and 1% who went for someone else.

    So of 2020 voters, Harris lost a net 4% to the couch and 2% to switching. You can count the switchers twice because they were a lost vote for Harris and a gained vote for Trump, so that’s basically a wash. Trump then won a net 1% of people who didn’t vote in 2020 (which coincidentally is roughly the same size as an individual candidate’s 2020 voters). So doing better with any of those groups could have swung the election.