The Hard Voter Data Indicating Democrats Will Outperform the Polls & Win Big: In Data (& Black Women & Latinas) We Trust

One of the best data guys in politics back in 2022 who was quite uniquely bullish on Democratic turnout for the midterms and was proven right is drawing attention to an even better situation for Democrats in 2024

(Traduce en español/translate to Spanish) By Brian E. Frydenborg (Twitter @bfry1981, Threads @bfchugginalongLinkedInFacebook, Substack with exclusive informal content) September 6, 2024; see my related July 5 article, Biden at His Worst Is Better than Insurrectionist Trump at His “Best”: 13 Reasons to Keep Calm & Carry On (Biden’s Already Gaining Ground!) ; because of YOU, Real Context News is approaching two million all-time content views, but I still need your help, please keep sharing my work and consider also donating! Real Context News produces commissioned content for clients upon request at its discretion. Note: when decimal percentages are given, averages are rounded to the tenth; when not given, they to rounded to the nearest full number; *correction appended to fix a date.

Tom Bonier/DataSmart/CBS News/The Daily Report with John Dickerson/The Atlanta Voice

SILVER SPRING—Just before the 2022 midterm elections in the wake of Roe v. Wade’s half-century precedent protecting reproductive abortion rigths being overturned by Dobbs v. Jackson Women’s Health Organization, I had been following the excellent analysis of Democratic political expert Tom Bonier.  He was then busy was pointing out both that women were registering in far high numbers than usual for midterms and, later, that the early vote—which had been favoring Democrats in recent election cycles—was also significantly higher than the previous midterms.  Because of the hard voter data Bonier so skillfully presented, I drew the conclusion in detail before the election that Democrats would outperform their polls and had a real chance to hold both the Senate and the House.  Democrats ended up not holding the House (barely, in large part thanks to 4 Republican flips of Democratic seats in New York, including by the now famous George Santos) but my analysis was correct: with turnout high and many more women voting than usual in a midterm, the pollsters were off in many races and underestimated the vote for Democratic candidates.  Back in 2022*, in the run-up to the midterms and commenting on the new registration surges, Tom was making it clear that he’s “never seen anything like it.”

That midterm, Joe Biden tied for the fifth-best midterm performance for his own party for his first midterm among all modern presidents, with the four doing better than him in the House having significant historical advantages when Biden was at disadvantage, and Biden is tied for tenth out of all presidents in the House and seventh in the Senate (excluding the aberrations that were the Reconstruction midterms and John Tyler, who was partyless).

My beautiful charts…
Wonderful New Data for Democrats

Now, Bonier is at it again with even far more encouraging and unprecedented data for Vice President Kamala Harris, her running mate—Governor Tim Walz—and down-ballot Democrats.  In what he is calling the Harris “Effect,” for the week beginning July 21st—the Sunday of which saw President Joe Biden formally withdraw as the Democratic Party’s nominee for the presidency—many constituencies that heavily lean Democratic in 15 states have seen massive surges in voter registration surges compared to the same time-period in the las presidential election in 2020, which Joe Biden won.

Tom Bonier/TargetSmart/Twitter
  • Young black women are up in voter registration by over 175%
  • Young Latinas are up almost 160%
  • Black women are up over 98%
  • Black voters overall are up nearly 85%
  • Young women overall are up over 84%
  • Latinas overall are up over 78%
  • Young voters overall are up nearly 75%
  • Hispanic voters overall are up over 68%
  • Democrats are up nearly 50%

But male voters—who strongly favor Trump—are just up over 18% and Republicans are just up 8% (not all states released party affiliation, so the Democratic/Republican registration was modeled by Bonier’s firm when that information was not available).  These were the included states: Wisconsin, Michigan, Nevada, Maryland, Delaware, Vermont, Rhode Island, Ohio, Kentucky, Missouri, Arkansas, Louisiana, Oklahoma, New Mexico, and Wyoming.  Since then, North Carolina, New York, and Pennsylvania (including 262% increase for young black women!) have been added and new data from Georgia analyzed by The Atlanta Journal-Constitution, with all confirming and continuing the trend (and even as I am proofing this, new data is in the process of being added from 19 other states that are confirming these overall trends again and again, but have not yet been presented as added to the overall demographic averages across all states analyzed).

Furthermore, Bonier noted that his political data outfit TargetSmart’s research “found that surges in voter registration are predictive of increases in overall turnout from those groups of voters (not just the new registrants in those groups).”  That’s huge, because this means these surges in registration are indicators of much more than just the individuals registering in record rates, individuals representing groups that are heavily pro-Democratic.

This is a dramatic imbalance that, if it holds, means that polling in these 39 states (38 looked at my TargetSmart and Georgia looked at by The Atlanta Journal Constitution)—including most of the swing states—will undercount support for Democratic candidates, perhaps significantly and perhaps more than in 2022.  This would mean if we look at these close averages, the election might end up being not as close as the polls are indicating they are: Kamala Harris might not just win most swing states, but could even do so convincingly, limiting the ability of bad-faith actors to disrupt the transition to a would-be Harris-Walz Administration or challenge election results and vote certifications.

2022 Midterms: Polling Past as Prologue for 2024?

Getting into the Weeds

I am not sure what changes pollsters have made and may yet make to their methodology from the midterms and previous presidential election for this election cycle.  Are they looking at these numbers and making adjustments?

I am inclined to think perhaps not much just by my gut, but beyond that, let’s look at what happened with the polls in 2022.

In 2022, Senate polls in the last three weeks before the midterms were 0.3% biased in favor of Republicans (also known as the GOP, or Grand Old Party), but 0.2% biased in favor of Democrats in the House.  However, this is actually quite misleading: as Nathaniel Rakich notes writing for FiveThirtyEight (a mecca for many things polling weighted whose averages are higher quality than those of Real Clear Politics), two main types of polls were included in the House calculations: polls for specific House district races and generic ballot polls, the latter being polls asking voters which party was preferred to control Congress, not about a specific House race.  Those generic ballot polls are generally far more accurate than the polls for specific House races (from 1998 and on, 3.9% average error vs. the 6.7% error for the district-specific polls).  In 2022, House polls were off overall by 4.0%, but the specific House race district polls were off by an average of 5.0% compared with 3.1% error margin for the generic polls.  And the polls for the House were overall relatively more accurate in part because a far larger portion of House polls in 2022 were the more accurate generic polls: 46% of all House polls when the average from 1998 and on has been only 21%.

As far as degree of error historically overall, since 1998, polls have been relatively close in the last three weeks before voting, averaging 6% error margins overall (9.2% for presidential primaries, 4.3% for presidential general elections, 5.4% for Senate polls, 6.1% for the House, and 5.4% for governors).  In the 2022 election cycle, the polls were even more accurate than usual and the best of all those examined starting in 1998, only being off 4.8% overall (off 4.8% in the Senate, 4.0% in the House, and 5.1% in governors’ races).  In spite of this relative accuracy, the polls were off in congressional House and Senate races, undercounting Democrats’ support as I suspected they would.

Women were 4% to 5% more of the electorate in 2022 than men, according to exit polls and the U.S. Census Bureau’s Current Population Survey, respectively (the two main sources of such information).  I tried looking under the hood of some of those final polls, but when I tried to find the details, some did not indicate their breakdown in gender as far as the sample and/or adjustments to the sample were concerned, though some did and seemed to have gotten it very close or right while others definitely underrepresented women (at least in the raw numbers of people interviewed, but even then, because pollsters don’t always get the exact portions they want for a likely voter model in terms of who responds, they adjust and weight their samples—including demographically including by gender—and it is not clear from the main press releases or crosstabs/methodology sections—if available—how these adjustments were made and if their projection for likely voters was the same or different from their raw sample, how accurate they were in modeling and predicting the portion of the electorate that would be female and how they would vote).

Some, perhaps most pollsters, would reveal their methodology upon request through individual channels but I confess I am pressed for time and resources in trying to track down methodology for two-year-old polls where the information is not as easy to track down online at this point in time, if it even is online, which can be difficult and time-consuming.  Under different circumstances in the future, perhaps I can and will.  Yet even the most wonky websites I have seen, including FiveThirtyEight, have not attempted compiling such a database…

Yet all things considered, given then-historic data on female voter registration in 2022 and because the polls were consistently off there, my hypothesis and one I feel good about given the situation with Roe being overturned is that women were either undercounted and/or the women that were counted were underestimated as far as their favoring Democrats (and I am thinking both, especially as new women registering after Dobbs were very likely motivated overwhelmingly by their loss of reproductive rights and would have been a much more Democratic-leaning group than women overall and who had registered prior to Dobbs).

Before getting into this next section, it should be pointed out, as Rakich does, that even some of the most accurate polls might get the race wrong in terms of predicting the actual winner, that the closer a race is the less “right” a often poll is in picking a winner: if a result is super-close, say, 0.5,% 1%, 2%, 3%, or even 4% or so, these results are often within the margin of error: the range above or below the level of estimated voter support for a particular candidate that the final result should fall under (in most cases) 95% of the time.  So, if a race is 47%-45% between two candidates and the margin of error is 3.2%, since the 2% difference is less than 3.2%, the race would be considered statistically tied.  But if the race was 50%-45% with the same margin of error, that lead would be considered more solid and safe.  And a poll can predict a winner who won by 5% but only have predicted a 1% win, while another poll could have been more accurate and have been off by less than 1% but predicted the wrong candidate.  In other words, polling is… complicated, and is really is about understanding about what the aggregate polling data means, not just screaming about one single poll.

And of course, one poll is just one data point, so it is the averages of polls over time and the polls closest to actual voting happening that matter the most, not one or a few polls.  But anyway, the point is, in very close races, pollsters should not be thought of as “off” if they predicted one candidate in their final polls as down 2 points who won by 0.5%, a 2.5-point-swing, if the margin of error was, say, 4%, meaning a 2.5%-swing either way would fall within margin of error, the way polling methodology is supposed to work.  So many polls could be super-accurate in close races and still get the winner wrong.  But what was interesting about 2022 is how many of the close races had polling biased against Democratic levels of support and were “wrong” even while often being relatively accurate.

Looking at predictions in 2022, FiveThirtyEight’s polls-only model had its average predictive outcome as 229 seats for Republicans to 206 for Democrats in the House, with the middle 80% of results landing from 250 to 208 seats for Republicans, and there were substantially more outcomes with Republicans doing much better than their average than Democrats doing better than their average.  For the Senate, the polls-only model predicted an average predictive outcome as a 50-50 Senate, with 80% of outcomes falling from 54 to 46 seats for Republicans.  For the overall popular House vote, the polls-only model had a 2.4% margin win for Republicans as the most likely outcome.  But keep in mind, these House models in a lot of cases included a lot of generic polls that were not a specific measure of the specific House race in question (this is in part because many House district races have little-to-no polling specific to them) and that, as mentioned before, the generic ballots favored Democrats relative to the specific district polls.

So, what actually happened?

What the Results Tell Us

In the end, the election results gave Democrats 213 and Republicans 222 seats in the House, a loss of 9 seats for Democrats compared with the results from the 2022 midterms, yet which, as I have noted, was a performance for the history books for Democrats, who kept 7 seats more than the polls-only model’s average prediction.  And in the Senate, Democrats won 1 more seat than the average of the polls-only model prediction.

The final popular vote outcome for the House was 50.6% Republicans, 47.8% for Democrats, a 2.8% margin for Republicans, the actual final margin being 0.4% higher than the 2.4% the model predicted

Yet, that year, the generic Congressional control preference polling—an important factor in the polls-only model—ended with 1.2% advantage for Republicans (smaller than the 2.8% actual margin, but as Rakich notes in footnote 6 here, Republicans had many more seats in House races where their candidates ran with no Democrat even running to oppose them than the reverse, depressing what the final popular vote would end up for being for Democrats by about 1% and suggesting that some of surge of Democratic women was strategically felt more in competitive races give the number of upsets we will get into in the next few paragraphs).  As of September 5 of this year, Democrats have a 2.6% advantage, a notable swing and another indicator Democrats could have an even better year than 2022 (although to be fair, on September 2, 2022, with the same number of days until Election Day that year—three days later than this year—Democrats were up 0.9%, so the generic ballot would be a serious indicator only if Democrats still end up with an edge towards the end and through early voting, beginning over the coming days and weeks in many states, and there is no reason to think generic ballot polls will naturally mirror patterns from 2022 and necessarily have Republicans favored over Democrats).  If this paragraph was a bit confusing, the big takeaways are that the model and especially the generic ballot average estimates underestimated the national House vote margin for the GOP, but the GOP had a lot more races with no Democrats running in them, meaning this is to be expected, and at least now the generic ballot polls are much better for Democrats than they ended up being in the end for Democrats in 2022, something that if it holds could be another good sign for Democrats.

As far as those upsets, specifically, the polls-only model favored Republicans in 13 House races in which Democrats pulled off upset wins compared with only 7 situations where the model predicted Democrats to win in the House and Republicans won instead (including the 4 surprises from New York state).  In the Senate, the model favored Republicans in two races that Democrats won.

FiveThirtyEight

If we break down these races by how close they were in polling, including governor races along with congressional ones, in the toss-ups (leader with a 50%-60% chance of winning in the model), those “tilting” Democratic—7 races in which Democrats were favored on average 55% of the time to win—only 3 of those 7 Democrats won (43%).  But in those “tilting” Republican—11 races in which Republicans were favored on average 53% of the time to win—only 3 of those 11 Republicans won (27%).

In the “lean” races (leader with a 60%-75% chance of winning in the model), those “leaning” Democratic—26 races in which Democrats were favored on average 68% of the time to win—23 out of 26 Democrats won (88%).  But in those “tilting” Republican—13 races in which Republicans were favored on average 67% of the time to win—only 8 of those 13 Republicans won (61%).

In the “likely” races (leader with a 75%-95% chance of winning in the model), those “likely” Democratic races—36 races in which Democrats were favored on average 88% of the time to win—all 36 Democrats won (100%).  But in those “likely” for Republicans—44 races in which Republicans were favored on average 86% of the time to win—40 of those 44 Republicans won (91%).

Thus, in key races, polling relatively favored Republicans much more in key races where Republicans were upset than the reverse, showing a significant undercounting of Democratic support.  In the competitive races (combining “toss-ups”” and “leans,” Democrats won 26 of 33 (79%) while Republicans won only 11 out of 24 (46%) and were upset in 4 races they really should not have lost(“likely”) while Democrats held all those seats.  Again, a historic performance for Democrats against very strong headwinds.

Thus, despite the narrative that the polls were historically accurate in 2022 and they certainly were in a relative sense, they were still consistently off in favor of Republicans in 2022 in many key races by underestimating Democratic support.  To get back to my question about whether pollsters will have adjusted much for this, this accuracy narrative might actually be leading pollsters this cycle to adjust less and question what they did in 2022 less, which might very well be setting up a repeat of pollsters undercounting Democratic support among voters, within the composition of the electorate, and level of Democratic support among women and other key groups, especially since the registration surges for strongly-Democratic demographics are even more dramatic and historic now after Harris’s rise than in 2022.  And as these are new registrants, a very high percent of the new registrants voted in the 2020 presidential election (81.3% in Pennsylvania, for example).

Conclusion: Expect Democrats to Overperform this Election

Given what happened in 2022 and the tendency of that cycles’ polls to miss a surge in Democratic turnout in dozens of close races and even a few that were not close in polling that led to key upsets, as well as the fact that polling now is better and that registration numbers are significantly better across a wide variety of states, feel free to add a few points to the numbers you are getting from polls for Democrats in most key races and attribute that the historic rise in voting registration of young voters, women, black voters, and especially Latinas and even more especially African-American women, then look at the relatively paltry numbers among groups that could favor Republicans.  Additionally, there are a number of other dynamics I felt would favor President Biden as a candidate over time before he dropped out, and most of those still apply to Harris now.  Because of all these factors and the hard voter registration data so wonderfully presented by Tom Bonier, now Kamala Harris, Tim Walz, and down-ballot Democrats can be quite confident in victory, especially now because people are responding disproportionately well when it counts most and that will count in the close races that will decide the fate of our republic in the face of insurrectionist Trump’s fascist, violent assault on American democracy.

In data we trust, but also in Latinas and African-American women.

© 2024 Brian E. Frydenborg all rights reserved, permission required for republication, attributed quotations welcome

Also see Brian’s eBook, A Song of Gas and Politics: How Ukraine Is at the Center of Trump-Russia, or, Ukrainegate: A “New” Phase in the Trump-Russia Saga Made from Recycled Materials, available for Amazon Kindle and Barnes & Noble Nook (preview here).

eBook cover

If you appreciate Brian’s unique content, you can support him and his work by donating here. Real Context News produces commissioned content for clients upon request at its discretion.

Feel free to share and repost this article on LinkedInFacebook, Threads, and Twitter. If you think your site or another would be a good place for this or would like to have Brian generate content for you, your site, or your organization, please do not hesitate to reach out to him!