Voter population modeling seems to be the real challenge, and it always seems to be blamed when the overall pulling averages aren't that accurate. I wonder if one's enthusiasm for responding to polls or pollsters is part of the modeling process? For example, if I was really excited about candidate x I might be more likely to respond to the poll. How's that accounted for?
Similarly, in some states really understanding the details of some smaller demographic groups could swing an election. For example, in Texas perhaps the behavior of Black men under 40 who traditionally voted Democratic but in the last two elections voted for republicans could determine the outcome of the senate race. Do posters try to model that with the tiny sample sizes they're going to get or at this level of detail is that too small to deal with so it just turns out to be part of the margin of error even though it could be very important?
Those are really intelligent questions. The enthusiasm question is a crucial one, because while unenthusiastic votes count just as much as those of rabid fanatics, unenthusiastic individuals are less likely to bother to vote. It has to be hard for pollsters to make an accurate assessment of just which unenthusiastic individuals will vote.
Not even sure you have to go to the other end of the enthusiasm gradient to the mostly unenthused voter. If I'm gung ho for Annie Andrews I guess I'd be mostly gung ho to answer a poll, if they don't interrupt my dinner. But if I'm a reliable voter but in the "you have to vote for someone/it's my duty" category, which seems like 50% of the voting population based on comments I here on election days when I'm working polls, then wouldn't I also be far less likely to spend 15 minutes talking to a pollster?
Poll of Latinos in 17 battleground districts: Latino voters across a survey of 17 competitive House districts shows Democrats leading 58%-35%. The margin is 12 points higher than the self-reported 2024 vote. In three Texas districts, Latino voters went from an even self-reported 2024 split to a Democratic 56%-36% lead. Trump’s job approval in these 17 districts is 36%-63%.
(M/E ±2.5) (1,500 Latino RVs; Conducted 8/6-17 by Hart Research for TelevisaUnivision
Good timing! I was just polled the other day about the race for mayor in my hometown. They started off by asking demographic questions. I wonder if that was so that they can categorize responses from people who don't stay around for the whole interview and get something out of it. They started off with the horse race question : Karen Bass or Nithya Raman? and asked about the level of support (strongly, somewhat, neutral, somewhat oppose, strongly oppose. I am undecided, maybe leaning towards Nithya...
They started losing me when they got to the message testing part of the interview. Would (an issue or argument) make you more inclined to support or oppose someone? Yeah, it makes me more likely to oppose the pollster! It used to be that pollsters would do focus groups where they paid the participants for their time and opinions. Now they expect us to give them unlimited time for free. I always ask how much of my time they want and they consistently underestimate it. "Ten to fifteen minutes" means at least twenty. Of the last few times I have been called, this was the closest I have come to completing an interview. When twelve minutes turned to nineteen I hung up. I am still undecided, leaning towards Nithya, though neither candidate is amazing or horrible.
Polling is broken, maybe never to be put back together. There can be some useful information gleaned from polls, but there is a whole lot of noise for relatively little gain. End of rant...
I love following the elections in New England with all their tiny towns and grants and wondering things like, with only about 130 potential voters, why is Hales Location still unreported as of 12:15
No surprise indeed. UNH polls have always, since I have been following elections in 2005, been noisy and prone to wild swing. They have serious quality control and sampling issues. Primary poll was way off on the Dem side, and I think the GE result was also off. But so tedious seeing even the NYT and Pappas's campaign all braying loudly that Dems are risking their odds of losing the Senate by "neglecting" NH (a small, cheap state where outside Dem groups have already booked 10 million in ads) and that Pappas is down to Sununu, when I will, if not eat, at least lick my shoe if Sununu comes closer than 5 points to winning the race in this environment and with a trade war with Canada ongoing.
Very helpful and clear criteria. Another reason I love the Downballot.
Agreed. It also doubles as good info to know about which polls are legitimate or not.
Voter population modeling seems to be the real challenge, and it always seems to be blamed when the overall pulling averages aren't that accurate. I wonder if one's enthusiasm for responding to polls or pollsters is part of the modeling process? For example, if I was really excited about candidate x I might be more likely to respond to the poll. How's that accounted for?
Similarly, in some states really understanding the details of some smaller demographic groups could swing an election. For example, in Texas perhaps the behavior of Black men under 40 who traditionally voted Democratic but in the last two elections voted for republicans could determine the outcome of the senate race. Do posters try to model that with the tiny sample sizes they're going to get or at this level of detail is that too small to deal with so it just turns out to be part of the margin of error even though it could be very important?
Those are really intelligent questions. The enthusiasm question is a crucial one, because while unenthusiastic votes count just as much as those of rabid fanatics, unenthusiastic individuals are less likely to bother to vote. It has to be hard for pollsters to make an accurate assessment of just which unenthusiastic individuals will vote.
Not even sure you have to go to the other end of the enthusiasm gradient to the mostly unenthused voter. If I'm gung ho for Annie Andrews I guess I'd be mostly gung ho to answer a poll, if they don't interrupt my dinner. But if I'm a reliable voter but in the "you have to vote for someone/it's my duty" category, which seems like 50% of the voting population based on comments I here on election days when I'm working polls, then wouldn't I also be far less likely to spend 15 minutes talking to a pollster?
Yes, for sure.
Poll of Latinos in 17 battleground districts: Latino voters across a survey of 17 competitive House districts shows Democrats leading 58%-35%. The margin is 12 points higher than the self-reported 2024 vote. In three Texas districts, Latino voters went from an even self-reported 2024 split to a Democratic 56%-36% lead. Trump’s job approval in these 17 districts is 36%-63%.
(M/E ±2.5) (1,500 Latino RVs; Conducted 8/6-17 by Hart Research for TelevisaUnivision
https://www.axios.com/2026/09/08/poll-latino-voters-trump-gains-fade?utm_source=newsletter&utm_medium=email&utm_campaign=newsletter_axiosam&stream=top
Good timing! I was just polled the other day about the race for mayor in my hometown. They started off by asking demographic questions. I wonder if that was so that they can categorize responses from people who don't stay around for the whole interview and get something out of it. They started off with the horse race question : Karen Bass or Nithya Raman? and asked about the level of support (strongly, somewhat, neutral, somewhat oppose, strongly oppose. I am undecided, maybe leaning towards Nithya...
They started losing me when they got to the message testing part of the interview. Would (an issue or argument) make you more inclined to support or oppose someone? Yeah, it makes me more likely to oppose the pollster! It used to be that pollsters would do focus groups where they paid the participants for their time and opinions. Now they expect us to give them unlimited time for free. I always ask how much of my time they want and they consistently underestimate it. "Ten to fifteen minutes" means at least twenty. Of the last few times I have been called, this was the closest I have come to completing an interview. When twelve minutes turned to nineteen I hung up. I am still undecided, leaning towards Nithya, though neither candidate is amazing or horrible.
Polling is broken, maybe never to be put back together. There can be some useful information gleaned from polls, but there is a whole lot of noise for relatively little gain. End of rant...
I'll answer text or online polls but I am not talking to a person for 20+ minutes over issue questions lol
Between Strength in Numbers and The Downballot, I now have parameters for judging polls. Thanks for the detailed explanation.
Justice Brett Kavanaugh today turned down a request to pause a ruling by the Missouri Supreme Court that bars the state from deploying the new map. This dispute should be over for the year. https://www.scotusblog.com/2026/09/justice-kavanaugh-turns-down-request-from-missouri-to-use-congressional-map-expected-to-aid-repu/
Now let's hope the citizen's referendum passes.
NH Sen: No surprise but it looks like Pappas is running away with it and will be our nominee. Currently, with 22% counted, he is up 62%-36%.
I love following the elections in New England with all their tiny towns and grants and wondering things like, with only about 130 potential voters, why is Hales Location still unreported as of 12:15
No surprise indeed. UNH polls have always, since I have been following elections in 2005, been noisy and prone to wild swing. They have serious quality control and sampling issues. Primary poll was way off on the Dem side, and I think the GE result was also off. But so tedious seeing even the NYT and Pappas's campaign all braying loudly that Dems are risking their odds of losing the Senate by "neglecting" NH (a small, cheap state where outside Dem groups have already booked 10 million in ads) and that Pappas is down to Sununu, when I will, if not eat, at least lick my shoe if Sununu comes closer than 5 points to winning the race in this environment and with a trade war with Canada ongoing.