NYT’s Sample Versus Early Vote

At this stage in the Presidential race should start to show some herding towards what we’re seeing in the early vote. But NYT’s last poll suggests NYT was inconsistent in doing so.

This table attempts to compare how many people in NYT’s state-by-state samples said they had already voted, as compared to the demographics of those who’ve actually voted.

Given Ann Selzer’s surprise and the known herding we’re seeing from everyone else, I wanted to see how closely NYT’s sample matched current voting trends.

The first thing this shows is that only in PA is NYT’s sample close to capturing the current number of people who’ve already voted — but that’s likely because it was in the field later than other states (save Michigan), just from October 29 to November 2. By comparison, the Georgia and Nevada polling started on October 24 and Arizona started October 25 — all before Trump’s Madison Square Garden event.

Most of Georgians and North Carolinians have voted, but only half of those polled by NYT said they had.

In one state — Georgia — NYT had too few men who’ve already voted. That suggests their result, showing Harris winning by a point, may undercount Trump supporters.

But then there’s MI. To come up with their tied result, NYT first found too few women who had already voted, both in terms of the total who have voted, but also in terms of the mix. And then they said that the likely voter split would be 52-47, when it’ll probably be at least 53-47.

Nevada, which is one of the states I think Trump is most likely to win, is what puzzles me most of all. It’s has a near-even split on known gender, plus 11 percent where gender is not known (likely due to their automatic registration). From that, NYT comes up with a near-even split on its already-voted sample, as if those unknown voters split evenly. Its final Likely Voter model says more men will vote than women. That’s not going to happen. Even still, it shows Harris winning by 3.

Please let me know of any errors you catch, as I’m sure I’ve made a number.

Update: I’ve added a column showing the NYT’s 2020 exit poll on gender. It suggests NYT’s likely voter model for this year is significantly undercounting women in MI, NV, and NC (but could hypothetically be overcounting them in WI).

Legend

In this table, blue is male, pink is female, and grey is unknown.

Reported Early: The reported early gender split, from NBC.

% of 2020: The known total early vote compared to the total vote from 2020.

NYT already: The percentage of each gender in NYT’s sample who say they’ve already voted.

NYT LV: The gender split in NYT’s likely voter model.

NYT result: The topline results of NYT’s poll.

NYT 2020 exit: The gender split in NYT’s exit polls.

2020 total: The total votes cast in 2020.

Current EV: The total early votes cast, from Election Lab’s numbers.

 

image_print
27 replies
    • Purple Martin says:

      Speaking as a Color Critic only, observe male is a soft Baby Blue, while female is a vivid Muscular Magenta.

      In the artist’s palette…

      “Magenta is vibrant and bold, often associated with creativity, passion, and a touch of mystery.”
      “Baby Blue is a soft pastel evoking calm and tranquility, but often associated with depression and detachment.”

      Hmmm… [stroking-chin-emoji] 8-)

      Reply
      • grossman says:

        Male color is cyan, not Baby Blue. From colorpyschology.org, “Cyan is a bright, refreshing color that sits between green and blue on the color spectrum. It’s a shade that evokes feelings of clarity, freshness, and tranquility.”

        Reply
        • Rayne says:

          Okay, this ^ is the last comment on color which is fairly arbitrary in its selection.

          The point of this post and the thread is the NYT’s sample versus early vote, which means the focus is on data and analysis, not color interpretation.

          Let’s get back on topic, thanks.

    • JonathanW says:

      I’m color blind, so I truly cannot identify the “blue” and the “pink”, they both seem to be different shades of the same color to me (purple, fwiw). I believe the top row is men, middle is women and the 3rd row, when it exists, is unknown. I hope this helps anyone out there who is color blind like me. In my job I’ve had to ask people to use symbols in addition to colors in spreadsheets or else I can’t tell what’s going on.

      Reply
        • JonathanW says:

          Thanks for asking! A lot of people ask me if I only see shades of grey, and the answer is no. I can really only speak for myself on this, but my understanding of the issue is a question of saturation of color, especially red. So, in your chart, both colors looked purple but I could tell what was grey. But I could identify red if it was just a big blob of red by itself. Examples of things where I get really confused outside of spreadsheets: rock climbing walls, charts with colored lines on a white background, back in high school and college, playing with electric circuits, trying to figure out the little bands of color on the resistors….

          In my job people show me spreadsheets with red for delayed, green for on track, and I had to train people to use blue for on track and red for delayed. Oh, and I’ve occasionally purchased clothing that really clashed for normal people but I thought really went well together!! :)

        • JonathanW says:

          Oh and I should add that when I say “both looked purple” I mean that I could see they weren’t the exact same, but I couldn’t match the word “blue” and “pink” to either one with confidence. I thought they were different shades of purple, one is clearly lighter to me and the other darker.

        • Savage Librarian says:

          I’m not color blind. But I would prefer gray. The colors are much too intense for me to look at the chart. It is jarring, unsettling. I can’t gaze at it without feeling uncomfortable.

        • JonathanW says:

          I don’t want to speak for other people with color blindness but one thing I do at work is use a lot of green, yellow and blue, as well as grey, white and black. I personally can pretty easily identify those colors in contrast with each other.

  1. marc sobel says:

    Shorter: NYT right thumb on the Poll scales doesn’t talk to the left thumb on the Poll scales.

    (intended as a comment, not an analysis)

    Reply
    • grizebard says:

      Of course. But with Trump in the mix, there was always going to be turbogrift, right across the board, to the very last writ.

      Always worth keeping in mind though that eventual sore loser Trump is far better in the long run than (temporarily) happy winner Trump.

      Reply
  2. Amateur Lawyer At Work says:

    AZ is a little suspect with the NYT Already showing more men than women. Same as NV, looks like NYT decided that “unknown gender” would be a 50-50 split. A split more with more women voting would narrow the 3% GOP advantage down into true toss-up land.
    However, outside of “that’s bold/shocking/non-savvy”, I haven’t seen them address why their model disagrees with Selzer’s model in Iowa.

    Reply
  3. Shadowalker says:

    Not sure if this is relevant. But PA does not have in person early voting. Much of what we saw last week was Republicans trying to shoehorn early voting using absentee ballots, so they were applying for a ballot after passing all requirements, filling out the ballot and then putting the ballot into the drop box. That could be why the early vote count is so low comparatively.

    Reply
    • emptywheel says:

      Oh, the early vote is much lower. It is either absentee or in-person absentee.

      And what I’m trying to say is PA is the only state where NYT captured a remotely real reflection of already cast votes.

      It appears they boosted the value of the men who had voted and very slightly backed off likely gender split. But that’s not the weirdest state here.

      Reply
  4. RockyGirl says:

    Copying my comment from the previous post since I think it’s also relevant here:

    I just finished working 16 days of early voting. We had over 12,500 voters come through just our voting center in that time. Huge numbers of women, black voters, and young voters. We don’t register by party so can’t tell that way, but I saw lots of Democratic sample ballots in their hands, way more than Republican ones. Lots and lots of first-time voters too. We are very blue, but the Dem enthusiasm as compared to the other side was palpable.

    Reply
  5. Ithaqua0 says:

    Thoughts and meditations:

    Their per-state sample size averages about 1,100, so the margin of error on the vote % is about +/- 3%.

    I wonder if women are a higher proportion of non-responses to the Siena survey. You’d only need about 5% fewer women than men choosing to respond (perhaps declining by saying they hadn’t voted yet,) and if Siena made no adjustment for non-response, you’d get roughly the observed gender effect, which looks real to me too, in the sense of being really wrong rather than just luck-of-the-draw wrong.

    Reply
    • Rayne says:

      LOL just realized that describes my household. My spouse will answer the phone to play cat-and-mouse with callers, whether scammers in India or political surveyors.

      I won’t answer the phone unless it’s one of a preprogrammed ring for approved persons or a caller ID matching someone from whom I am expecting a call.

      I have to wonder how many women won’t answer a landline for security reasons.

      Reply
  6. Garrett_09NOV2008_1320h says:

    How accurate, actually, though, are the figures about “the demographics of those who have actually voted,” coming from the voter data outfit TargetSmart, that your NBC cite is using?

    (NYT/Siena uses a different, competing voter data outfit, L2.)

    Consider Michigan. That I know of, party affiliation is not recorded or included in official data files there.

    By what Dark Data Arts the voter data outfits TargetSmart, or L2, use to link voters to their party affiliation, or to other demographics, is not information they disclose.

    Since they are not transparent about their methods, how good their methods are, is not something we can really assess.

    When there are discrepancies (say between quasi-random NYT polling samples, and the “actual” numbers from these firms), how certain can we be in knowing on which side and by how much the actual error lies?

    There is a recently increased scrutiny and suspiciousness, I think, of the methods and results of opinion pollsters, and of the poll aggregator forecasters like Nate Silver.

    But the less visible less publicized parts of the system, such as TargetSmart and L2, deserve some scrutiny and some suspiciousness too.

    [Welcome back to emptywheel. Please choose and use a unique username with a minimum of 8 letters. We have adopted this minimum standard to support community security. With only (~294) comments at this site to date as “Garrett,” you will not be grandfathered to keep your existing username. Because your username is far too short it will be temporarily changed to match the date/time of your first known comment until you have a new compliant username. /~Rayne]

    Reply
  7. bloopie2 says:

    I hadn’t realized the significance of early in person voting. When we went to do early in person voting yesterday at our town hall, they said that in the two weeks that option had been available (on a daily basis!), some 35% of the electorate had come in to vote. They also said that this was the first year for it, so I have got to believe this will be a thing going forward now. It’s wonderful that you’re no longer no tied to that arcane “takes a day to travel there Tuesday”.

    Reply
  8. earthworm says:

    moderators – please help cleanup with this link.
    the results are being updated every 10 hrs, according to my source.

    https :// aaruaaru. com/model

    [Moderator’s note: the above link has been “broken” with blank spaces to prevent accidental clickthrough. It is a AI-based prediction system which relies on other AI-based data including chatbots; it is NOT a pollster. We do NOT allow AI-generated content here; in this case, each query generates an exorbitant amount of CO2 to obtain results which may be easily distorted by data not transparently obtained and understood by persons who look at the results. Any other comments containing and/or using AI content will be binned; this comment has been left up for the purposes of explaining to readers here what Aaru is and does and why it will be rejected in future comments. /~Rayne]

    Reply
  9. earthworm says:

    Rayne,
    thank you, for the helpful explanation and all the other things you do too.

    Reply

Leave a Reply

Want to join the discussion?
Feel free to contribute!

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.