Dem
52
Ties
3
GOP
45
Today, as promised, a LOT of questions about polls.
We were also going to run reader comments about polls, but: (1) this post is likely to generate a lot of comments, and (2) this post is already quite long. So, we'll run some reader comments about polls during the week.
J.M.A. in Round Hill, VA, asks: Okay, here's a polling question: What's the point?
You wrote: "[Pollster Charlotte] Swasey said polling is the worst system—except for all the others that have been tried."
Better for what? I struggle to see what the fundamental value of horse race polling is. I see the incentives for pollsters to do them (both legit and...less so), but what is the actual societal benefit to it, as opposed to people just going to their polling place and voting for their choices without a larger sense of who's likely to win?(V) & (Z) answer: Polls tell citizens which races are close, or reasonably so. This can help them make their decisions about where (or if) to donate, where (or if ) to volunteer, and whether or not to make a point of casting a ballot.
It is true that dishonest polls, or lopsided polls, can reduce engagement with the democratic process by conveying the message "there's no point." But we are confident that they encourage engagement more than they decrease it, because if it is a crapshoot every time a person donates/volunteers/votes, then that will undercut enthusiasm. It helps people to feel that, most of the time, their investment of time and money has a good chance of being useful. Donating money in Alaska matters; donating money in Wyoming does not.
D.J. in Glen Cove, NY, asks: Aren't polls supposed to be a way to detect cheating in an election? Why, when polls and "results" differed, especially for Mr. 11,780, weren't there calls for audits, recounts, anything rather than questioning the pollsters?
(V) & (Z) answer: Horse race polls are not intended to serve that purpose, nor should they be used in that way. If a horse race poll is off, it's entirely possible—likely, in fact—that the pollster made incorrect assumptions about the electorate, or other errors. A horse race poll being off, or many horse race polls being off, most certainly is not evidence of fraud. A poll tries to measure what the electorate wants on the day of the poll.
Exit polls, on the other hand, should give a very accurate sense of what the actual electorate looked like. And if there are good exit polls, and those exit polls are way out of alignment with the results, that could indeed be useful evidence of fraud.
M.H. in Salt Lake City, UT, asks: Why do you regard Harry Enten of CNN as "useless"?
(V) & (Z) answer: First, because Enten feels a need (perhaps at the behest of his bosses) to turn every poll into a BIG STORY. Sometimes there isn't a BIG STORY, which is why we sometimes write something like "Well, it's another pretty good poll out of Texas for James Talarico" for "Today's Senate Polls."
Second, because unlike his former boss Nate Silver, Enten rarely has an insight that goes beyond what anybody could come up with if they looked at the poll. Specifically, he almost never digs into crosstabs to see what interesting or deeper lesson a poll (or polls) might convey.
Third, because Enten, almost like clockwork, performs balance. There are many weeks where his Monday BIG STORY is "Good news for Republicans!" And his Tuesday BIG STORY is "Good news for Democrats!" Then the Republicans get Wednesday, the Democrats get Thursday, and Friday is a coin flip. Because of this (presumably Natalie Harp only shares the Republican days), Enten is one of Donald Trump's favorite psephologists.
D.E. in Lancaster, PA, asks: I'm going to go out on a limb here and guess that there are no regulations governing polling companies. Is there even a trade association recommending best practices? Do you all think there should be some kind of regulation? If yes, then what kind of regulations would you like to see? If nothing else, it seems that Wedgewood and SoCal Strategies defrauded their clients.
(V) & (Z) answer: The American Association for Public Opinion Research (AAPOR) is the primary professional association for pollsters and prognosticators. Here is what they do, per their website:
Membership in AAPOR is all about opportunity—the opportunity to learn from a diverse group of leaders in the survey and public opinion research field, the opportunity to network and exchange knowledge and the opportunity to improve how survey research is conducted and disseminated.In other words, it's a voluntary organization for those psephologists who wish to enrich themselves. AAPOR is obviously not a licensing body, and so obviously does not have the power to impose any sort of formal sanctions on bad actors. Although they do censure bad pollsters, which is roughly the same thing as when Congress censures one of its members. That is to say, it carries no penalty, but it's embarrassing and might be used as ammunition by the competition.
If a pollster who is being paid for their work were to fake their results, they would be defrauding whatever person or entity it was that paid them, and they could be held civilly liable. But if someone just cooks up a fake poll on their own and puts it out there, that would be hard to punish. There are First Amendment issues there, but there's also the problem of finding someone who has standing and who can prove they were somehow damaged by that particular poll. Wedgewood and SoCal Strategies were not working for specific clients, and so were not being paid for their "efforts." By and large, if a pollster is working for someone, then the poll is identified in the form [SPONSOR]/[POLLSTER]. For example, New York Times/Siena, or CNN/PF Global.
We think it would be useful if AAPOR were to get into the business of giving pollsters some version of a "seal of approval," like with foods that are certified kosher, or toothpastes that have the ADA seal, or movies that have been supervised/approved by the American Humane Society. Pollsters would not be required to seek such approval, but there would be benefit in doing so. And folks interested in that information (like, say, us) would be able to check and see if a pollster is AAPOR-approved or not. That would certainly make it harder for phony pollsters to break through.
A.L. in Highland Park, NJ, asks: You mentioned that prediction markets do not add any more value over polls. How about experts, specifically Charlie Cook and Larry Sabato? It seems to me that they tend to follow the polls, adding an extra bit of uncertainty over and above the statistical. Have they ever correctly called a race contradictory to the polls?
(V) & (Z) answer: We don't recall having looked back at Cook/Sabato/Nathan Gonzales after an election, and we don't recall anyone else having done so. Good idea for this year, though. We'll put that on the post-election to-do list.
And they are a useful data point. Their analyses give a good sense of which party, if either, has momentum. Further, if they, and the polls of individual races, and the generic congressional ballot, and the sitting president's approval are all pointing in the same direction, that's pretty good evidence that the direction is the correct one. Also, they can take qualitative factors into consideration quickly, before there is new polling. For example, Cook has moved the Ohio race for governor to lean democratic on account of the pushback on the despicable ad Vivek Ramaswarmy [sic] is running.
T.J.C. in Boston, MA, asks: What's the possibility that prediction markets can be manipulated to favor an underperforming candidate? A billionaire has the resources and no interest in actually winning the bet. I'm guessing it is a lot less labor intensive than creating a fictional polling house.
(V) & (Z) answer: We just don't see it. Not that many people follow the prediction markets, and those who do almost certainly suspect something funny is going on when a candidate surges unexpectedly.
T.K.B. in Seattle, WA, asks: You explained how FiftyPlusOne was able to look at the numbers and identify gaps, which in turn led them to run a comparative poll. Will this kind of detection even be possible in a year or two? I'm envisioning a scenario where an AI model is trained on a bunch of real polls to understand weighting and cross tabs, someone inputs the desired outcome, and the AI model generates underlying data and crosstabs matching the predetermined outcome. This is not dissimilar from today but with the weight of AI masking the obvious signals in the data.
At the very least, it would eliminate the obvious tells of made up data if all the obvious coincidences were made to look "normal." Arguably more importantly, the speed at which this can be detected, confirmation via another poll, outlier analysis, etc. will lag tremendously far behind the speed the lie travels and impacts betting markets (gross), news cycles, donors, etc. Back in the day, FiveThirtyEight graded pollsters. Do Experian, Moody's or some new equivalent need to start grading pollsters?(V) & (Z) answer: This strikes us as a real possibility; the next fake pollster is not likely going to be as careless as the ones who were outed this year.
That said, there's still a problem here, and it's the problem that actually outed Wedgewood. If there is a poll, someone had to do the actual contacts with respondents. And if a pollster can't explain credibly how they collected their data, then it's a giant red flag. Couple that with a lack of transparency about who is behind the polling outfit, and it would be near-fatal.
The upshot here is that everyone is just going to have to be more careful about vetting pollsters before they accept their polls as legit.
And Nate Silver isn't running FiveThirtyEight anymore, but he does still rate pollsters at his current site, Silver Bulletin. The New York Times goes a step further and when a new pollster pops up, a reporter calls the firm on the phone to ask questions about who the principals are, what their backgrounds are, and how they collect data. AI can't fake this easily.
P.R. in Arvada, CO, asks: Why? Why would someone set up a company to run fake polls? Saying that it is a social experiment isn't going to protect them from the lawsuits they will undoubtedly face along with the potential of getting charged with election interference. Surely there is no way they can get away with it. Right?
(V) & (Z) answer: For reasons we outline in the previous answer, we think it will be much harder to get away with fake polling in the future. That said, if a fake pollster limited their number of polls, and made sure to produce legit-looking results, then... maybe.
We can think of many reasons that someone might want to fake a poll:
- To sow the seeds of chaos
- To make some sort of point about horse race polling/election coverage
- To try to establish a "reputation" as a pollster, so that clients will begin paying for their work
- To try to influence one or more primaries or other electoral contests
- To try to trick a candidate or party into wasting money
- To try to influence prediction markets, so as to secure favorable betting terms
There are enough motivations that, once again, it points to a need for more aggressive vetting in the future.
R.E. in Birmingham, AL, asks: I understand that if one quality poll comes out 51% to 49% for one candidate, it falls within the margin of error. When you aggregate polls, to what extent does the margin of error narrow? I presume that more is better, but there are diminishing returns for each one you add. Can you talk about how that works without making me do any actual math?
(V) & (Z) answer: To a first approximation, averaging three polls each with 1,000 respondents has the same margin of error as one poll with 3,000 respondents. The margin of error (MoE) is inversely proportional to the square root of the sample size. So to get a 3% MoE down to 1.5%, you need to roughly quadruple the sample size.
But remember, sample size affects only statistical errors, not methodological errors. If you sample 10,000 people, the MoE will be 1% but it could still be off by a lot if, say, a quarter of the Republicans called refused to take the poll. Statisticians talk about "precision" to mean how reproducible a measurement is if you run it again. The bigger the sample, the higher the precision. In contrast, "accuracy" means how far off from the true mean you are. A poll could be highly precise (large sample, very reproducible) but also very inaccurate (because, say, many Republicans systematically refused to participate). To give a sports analogy, in darts, if someone can hit the same place on the target board within 1/8 inch over and over and over, the thrower is very precise. But if that spot is 3 inches from the bullseye, the thrower is not very accurate.
In statistics classes, professors often use the following analogy when discussing precision and accuracy. If you ask 100,000 Americans what the height of the emperor of Japan is, you will get a very precise (reproducible) answer. But it is probably not very accurate because none of the respondents even know who the emperor of Japan is, let alone how tall he is. At best, what you measure is a popular notion that Japanese people are short. A better example might be asking about the height of the president of Bulgaria, since few Americans have any preconceived notions of how tall Bulgarians are and probably don't know that the current one is a woman.
K.H. in Albuquerque, New America, asks: It seems like margins of error for the majority of pollsters hover around 3 or 4%. I would think a polling company could make a splash by increasing sample size and driving the margin of error much lower. I presume there's a cost-benefit tradeoff that makes this strategy not worth the bother. Is it time, money, or both that places the limits on reducing margin of error? Is there a theoretical and/or practical lower limit to margin of error?
(V) & (Z) answer: By and large, pollsters look to achieve 95% confidence with their numbers. In other words, that 19 out of 20 polls conducted identify the correct range of outcomes. And most polls are meant to reflect pools of voters larger than 100,000 people (which describes most mayoral elections, and all congressional, gubernatorial and presidential elections and primaries).
Under those conditions—95% confidence, 100,000+ voters being measured—then a pollster needs about 100 respondents to achieve a 10% MoE. They need about 400 respondents to achieve a 5% MoE, about 600 to achieve a 4% MoE, about 1,000 to achieve a 3% MoE and about 3,750 to achieve a 1.5% MoE. In other words, it quickly gets a lot more expensive to achieve only slightly better results. Most pollsters have decided the sweet spot at which cost/accuracy are best balanced is somewhere between 3% and 4%.
The numbers change a bit if the population is smaller, or if the target confidence level is something other than 95%. If readers want to play around with the numbers, there's a very usable calculator here. There is no lower limit, per se, because you could theoretically talk to everyone in the entire electorate. In that case, your confidence would be 100%, your margin of error would be 0.0%, and your bank account would be empty.
E.S. in Acton, MA, asks: Just how much money does it cost to commission a quality, state-level poll? (Obviously, once I win the MegaPowerZillionsBall jackpot, I will need appropriate things to spend my winnings on.)
(V) & (Z) answer: Depends on what kind of poll. Onlline polls cost between $1 and $4 per response collected. Robocalls cost between $1.50 and $5. Live-caller calls cost between $25 and $75. So, if we assume a fairly common 95% confidence, and a 3.5% MoE, and population of 100,000+ (which would mean that about 775 responses are needed), then the cheapest price would be around $775 and the most expensive would be about $60,000.
And when you win the lottery, we would be happy to oversee your polling project for a modest 40% administrative fee. We went to the Jared Kushner School of Business.
E.K.M. in Delaware, OH, asks: Instead of listing a race as Smith 47%-Jones 44%, why aren't there any pollsters using a format that would say Smith 43%-51%, Jones 40%-48%, instead of leaving it to the reader to spot or remember the margin of error?
(V) & (Z) answer: We are both teachers, and we think this would create more confusion than clarity. Anyone who is going to consume polls on a regular basis should have a basic sense of what MoE is, and how the polls are merely range projections and not exact predictions. Tomorrow we will have a graph of polls over time where the red and blue lines have light red and light blue bands around them to indicate the margins of error. Where the bands do not overlap, the difference is statistically significant.
B.B. in Dothan, AL, asks: Why don't you use the margin of error when aggregating poll results? Would that not provide more meaningful results?
(V) & (Z) answer: Maybe a little bit, but most political polls try to get about 1,000 valid responses for a poll. The difference in precision between 1,100 respondents and 800 respondents is so small compared to potential methodological errors that it is not worth even considering. We believe that when polls are off, it is due to bad samples (e.g., not getting enough Republicans to take the survey) rather than a sample that is too small. What separates the good pollsters from the bad ones is their secret sauce for trying to correct for under sampling some demographics.
D.J.M. in Salmon Arm, BC, Canada, asks: You have repeatedly addressed one of the risks of with gerrymandering, namely that basing this cycle's gerrymander on last cycle's voting patterns can blow up in the gerrymanderers' face (dummymander time!). Does the same problem exist for polls? For example, election polls were off because Latinos suddenly moved towards Republicans in the last election. Pollsters adjust for that movement and then Latinos move away from Republicans. Do pollsters make their adjustments based on the last election or do they modify in real time?
(V) & (Z) answer: This is indeed a problem for pollsters. They actually start with the last several elections (or more), and they try to make a reasonable guess about what this election's electorate will look like. And they do tweak their model as they collect data. That said, they are understandably leery of projecting dramatic swings in the makeup of the electorate, like "30% of Latinos are going to flip from Republican to Democratic." So, they may underestimate such swings when they happen.
J.A. in South Salem, NY , asks: How do polling firms deal with the shy voter? For example, people might not be willing to tell a person they will vote for Republican U.S. Senate candidate Ken Paxton (R-TX) but would be willing to vote for him in the privacy of the voting booth, and that behavior could skew polls.
(V) & (Z) answer: It's not an easy problem to solve. One option is to have checksum questions; if someone says they aren't voting for Paxton, but they do like guns, and they don't think corruption is a problem in politics, and they are voting Republican for every other office, then they could be lying about Paxton. Also, online polls have proven to have much less of a shy voter problem, since people are more willing to tell the truth to a computer than to a human being.
We do wonder how much of a problem the shy voter effect is right now, however. Part of Trumpism is wearing your MAGA credentials on your sleeve, not hiding them. The much bigger problem is Trumpists who just won't respond to polls because they don't like pollsters/the system/the deep state.
H.C. in Huntington Woods, MI, asks: I get texts to participate in polls all the time. I ignore them not wanting to click a link from an unknown source. Any suggestions for deciding which ones might be worth clicking?
(V) & (Z) answer: Those folks who are up to no good are very skilled at making themselves look legitimate. Our only advice is to copy and paste the exact message you got into Google, and to search for it. That should give you some webpages (for example, discussions on Reddit) where people are saying useful things about whether or not the message is fake.
A.W. in Lincoln, MA, asks: Now that many fewer people have landlines AND they don't answer calls from unknown callers on their cell phones, how do polling organizations effectively seek out "random" people? Does polling via websites or e-mail introduce too much selection bias (e.g., selecting for people who are online a lot, people who are "selected" via biased social media algorithms)?
(V) & (Z) answer: Some make lots of phone calls, even if they know only 1% of them will be successful. Some use multiple mediums, like phone calls and websites. Some use text messages. Some use e-mail. Some use postal mail. Some develop a regular base of respondents who are kept on board through some sort of rewards system. There is no one secret sauce. Some buy lists of registered voters, which many states sell, and randomly pick from the lists. This at least guarantees everyone in the pool is a registered voter.
J.B. in Bozeman, MT, asks: How does the U.S. polling industrial complex compare with that in other countries?
(V) & (Z) answer: The U.S. has more money than any other country. Also, it has major elections on a regular, 2-year cycle. This makes it particularly viable for a firm to make a living in the polling business.
That said, other Western democracies, along with some other nations beyond that, have fairly well-developed polling establishments. In many of those cases, the pollsters are academics who do polling as a public service/side gig.
B.S.M. in London, England, UK, asks: Do you have any data about which pollsters had the most correct predictions during the 2022 midterms and/or 2024 elections? In other words, who got it (the most) right? If you had to recommend one pollster to follow, who would it be?
(V) & (Z) answer: You can't judge a pollster based on one election, or even two. That said, there are a dozen or so pollsters who are widely considered to be at the top of the heap. Those include Siena, TIPP Insights, Monmouth, SurveyUSA, Public Opinion Strategies, Emerson College, Marist College and Echelon Insights. If you're only going to follow one poll, then it would probably be New Yotk Times/Siena. That's a high-quality pollster backed by the considerable financial might of the Times. Also, the person who manages the polling for the Times, Ruth Igielnik, has 17 years' experience in survey research.
E.M.H, in Oslo, Norway, asks: A short question that may warrant a long answer: What are the three most likely things polls may miss?
(V) & (Z) answer: Late breaking swings (say, in the last week or two of an election, as in the upper Midwest in 2016); big swings in a particular affinity group (e.g., Latinos in 2026, very possibly); the emergence of new and previously unaccounted for affinity groups (e.g., non-college men in 2016).
J.H. in Lodi, NY, asks: You wrote that, "[a polling lead of] +4 is not meaningfully different from the +2 and +1 from yesterday, in the world of poll math," but I thought you once said that even in a close race, being above 50% is significant, since it means that there's not enough wiggle room for the candidate who is behind to win over undecided voters. They have to persuade decided voters to change their minds, a more difficult task. How are both right?
(V) & (Z) answer: You are somewhat comparing apples and oranges. Because of how MoE works, leads of +1, +2 and +4 are all results that say: "We think we did everything right, but even if we're right, that does not mean the person we have in the lead will actually win. They are just slightly more likely to win than the person we have trailing."
If a candidate is above 50% in a poll, however, that works out to something like this: "The trailing candidate better hope that the poll underestimated their support. Because if the poll is accurate, or if it's understating the leader's support, then the trailing candidate has to win all the undecideds AND has to flip some people already committed to voting for the other candidate. Neither of these things is easy to do."
E.C.F. in Somerville, MA, asks: The thing I find most confounding is when one poll shows the Democrat ahead for the Senate and the Republican ahead for governor, and the next day a different poll of the same state shows the Republican ahead for Senate and the Democrat ahead for governor. I could understand if one model of the electorate favors one party, but I'm really curious how one poll can favor one kind of ticket splitter and another poll can favor the opposite kind of ticket splitter. Any insights?
(V) & (Z) answer: You can certainly imagine affinity groups that would disproportionately ticket-split in one direction or the other. For example, (male) Latino voters might like Gina Hinojosa (D) for governor in Texas because she's Latina, but Ken Paxton (R) for Senator because he's much more masculine/macho than James Talarico (D). Good-governance Republicans might be OK with Gov. Greg Abbott (R), but unable to abide the crooked Paxton. So, a pollster that assumes high Latino turnout might favor Hinojosa and Paxton. A pollster that assumes high Republican turnout might favor Abbott and Talarico.
D.V. in Columbus, OH, asks: With the polls tightening these past couple weeks, and the letter from M.S. in Portland, I'm definitely starting to get nervous, even though it's pretty normal for polling to tighten up at this time of year. I looked it up and, by a quick Google search, Sara Gideon led Sen. Susan Collins (R-ME) by an average of around 3.8 points going into the election. Collins actually ended up winning by 8.59 points (!), so how the heck did pollsters get things *that* horrifically wrong back then, and how can we be sure that their (ostensibly large) adjustments will be more effective this time around?
(V) & (Z) answer: The pollsters blew it, but not as badly as it looks. They undercounted Trump voters some, and also missed a small-but-meaningful late break for Collins. However, the biggest problem is that Maine was using ranked-choice voting back then, and 5% of voters cast their ballots for left-wing independent Lisa Savage. Nearly all of those folks had Gideon in second place. So, if Collins had gotten 1% fewer votes, the final tally would have been something like Collins 51.5%, Gideon 48.5%. But because Collins got just across the 50% line (50.98%), the ranked choice ballots didn't go into a second round.
Maine won't have ranked choice in this Senate election because there are only two candidates, Collins and Troy Jackson (D). And Donald Trump is not on the ballot, so missing a bunch of Trump voters who will also vote Collins is not an issue. That means that while Collins may certainly win, there is no reason to expect a repeat of what happened in the last election. She'll likely either win a squeaker or lose a squeaker.
S.P. in Harrisburg, PA, asks: There is a tendency among pollsters, over the last several elections, to understate Republican support compared to the actual election. Examples include Donald Trump in 2016 and 2024, Susan Collins in 2020, John Cornyn who outperformed polls in 2020 by 4 points, Roger Marshall in the Kansas Senate race in 2020, Chuck Grassley in Iowa in 2022 and J.D. Vance in Ohio in 2022. If there is a similar understatement of the Republican vote in these same states, then Maine, Iowa, Ohio, Texas, Kansas would all go Republican, and the Senate would stay Republican (not considering Montana and Alaska). There has been much talk of whether or not pollsters have learned from past mistakes this cycle, but at this point in the race, do you see some level of reason to be doubtful of the chances the Senate goes to the Democrats?
(V) & (Z) answer: The problem, which has gotten less pronounced over time, has been projecting what MAGA voters will do when Donald Trump is on the ballot. Nearly all of your examples are from elections where his name was there, and the two examples that are not were not big misses. In fact, the pollsters predicted a Vance win in 2022 by an average of 7.8 points, and he won by 6; they predicted a Grassley win by an average of 9.9 points, and he won by 12. That's one 2-point miss in one direction and one 2-point miss in the other.
The pollsters might indeed get it wrong this year; it's a strange year with a president who's done some very strange and impolitic things. But Trump is not actually on the ballot, and so the misses of 2020 and 2016 are not especially instructive.
D.B. in Mountain View, CA, asks: Has Ann Selzer (or anyone else) ever explained why she got that last poll wrong?
(V) & (Z) answer: She largely did not weight her samples, ever. Because Iowa is not a very diverse state, that was usually not a problem. However, in that last poll, she got a sample that proved to be way too Democratic. So, she got burned by the lack of weighting there.
W.K.D. in Fort Worth, TX, asks: Saw you have ActiVote polls in your database. Nate Silver says ActiVote is verboten.
(V) & (Z) answer: Smoking out "bad" or "fake" polls is a tricky business. What we are concerned about is eliminating pollsters who are actually campaign consultants and whose goal is to help the client win. If lying helps, they are all in for it. Usually their website is very open about this. Putting "We helped 18 Republicans win their elections" makes it pretty clear about what their business model is and is a fine way to attract new Republican candidates to hire them. ActiVote has a very extensive website and if they are campaign consultants, they are sure hiding it very well, which is not good for business.
Silver uses different criteria than we do. The tradeoff here is that being very picky about whom to include reduces the amount of data you have to work with. Also, when you start banning pollsters left and right for subjective reasons, you are introducing your own bias, so we try to have a fairly clean rule about who gets banned: campaign consultants. It is also noteworthy that some pollsters tend to work with leftish or rightish clients and are still pretty accurate. For example, Fox is using Beacon (D) and Shaw (R) working together this cycle and there is no reason to doubt these polls, even though Fox is the sponsor.
P.N. in Indianapolis, IN, asks: There always seem to be a few pollsters that are included by The New York Times and other aggregators that skew away from the pack when looking at Trump's approval ratings. Rasmussen, RMG and Morning Consult all come to mind. Why are their results so much more friendly to Trump? For example, Rasmussen has basically had his approval steady since April. Is there a chance that they are more accurate than everyone else? If not, why include them at all?
(V) & (Z) answer: Rasmussen and RMG are both the brainchild of Scott Rasmussen, who is right-leaning, and who assumes a more right-leaning electorate than most houses do. Morning Consult is not partisan in the way that Scott Rasmussen is, but they also assume a more right-leaning electorate than most houses. That is also true of Emerson and of Harvard/Harris. Sometimes they are right, and sometimes the other houses are right. There's no knowing for sure until the votes are counted, and so it is wise to include pollsters who lean in both directions, since any errors will likely cancel out.
J.Y. in Salem, OR, asks: I see from the October 2026 Times/Siena Poll of the Likely Electorate that 3% of Democrats either strongly or somewhat approve of how Trump is handling the job of president. So, my question is: Who are these people?
(V) & (Z) answer: There were people, mostly Southerners, who were Republicans in all but name for years, but who hung onto their Democratic identification for a very, very long time. Sometimes this was for political reasons, sometimes for what might be called sentimental reasons (e.g., "My grandpappy would be furious if I ever joined the party of the hated Abraham Lincoln.") Most politicians eventually had to bow to reality, at risk of losing reelection (for example, Alabama senator Richard Shelby finally flipped in 1994, even though he'd functionally been a Republican for 20 years). Some voters still haven't flipped.
There are also people who stay with the "wrong" party so they can vote in that party's primaries, or so they can respond to polls as a member of that party, or because they are too lazy to re-register, or because being registered with one party but voting for the other makes them feel like they are open-minded, and not "sheep."
Finally, there are some Democrats who think Trump is more "right" on the issues that matter than the Democrats are. For example, there are certainly staunchly pro-Israel Jewish Democrats for whom that issue is paramount, and who are happy that Trump is thick as thieves with Benjamin Netanyahu. Similarly, there are staunchly anti-Iran Persian Americans for whom that issue is paramount, and who are happy that a president finally bombed the evil Ayatollah.
Undoubtedly, there are some examples of each of these types in that 3%.
J.L.G. in Boston, MA, asks: When deciding on issues, do elected officials who disregard polls do so at their own peril? I'm curious how often leaders lead rather than be led by the wisdom of the masses. I imagine there's a mix, where they spend their political capital on the issues most important to them, and use polling on other issues to preserve their remaining capital?
(V) & (Z) answer: We are going to start with an anecdote. Bill Clinton was a brilliant politician. He was also a polling addict. And so, he would have his people poll the American people on everything, including questions like "Where should I vacation?"
Yes, it is true that Clinton was a centrist. But even presidents who were more partisan, and who served before the polling era, tended to be mindful of public sentiment. Franklin D. Roosevelt, for example, had numerous ways of keeping his ear to the ground.
The answer here, then, is this: Sometimes politicians will and should be out ahead of public opinion. But any politician who gets out too far ahead of public opinion does so at their own severe peril. So, they should always make sure they have a sense of where public opinion is. Most do this. Some, like the ones who attack Iran for spurious reasons, do not.
A.K. in Chicago, IL, asks: Has there ever been a politician, or an entire political party in the United States, or even somewhere else in the world, that saw that its policies, outside of a war, were hurting their constituents, or the country as a whole, and said, "Oops, our/my bad. This isn't working. We're going to go back and figure out a better way to do things?" I am not asking about bad poll numbers, but some kind of harm or unintended consequences caused by policies or actions, where the person/party took responsibility, apologized, said that they would think it through again, and changed course. If not, why not?
(V) & (Z) answer: While politicians have been known to apologize for one misstep (e.g., John F. Kennedy and Bay of Pigs), they do not often openly apologize for their entire political program, because that would send a very unhelpful message. But there are certainly politicians who aggressively recalibrate. The most obvious example that leaps to mind is FDR, who pursued fairly moderate policies in 1933-34, then saw all kinds of "FDR is not doing enough" blowback in 1935, and so pursued much more aggressive policies in 1936-37. He referred to all of this as the "New Deal," so as to obscure how very much he'd recalibrated, but historians refer to the "First New Deal" and the "Second New Deal," because the latter programs were so much more aggressive than the former.
F.S. in Cologne, Germany, asks: How did presidential candidates know where to campaign prior to the polling era? How did they know if a state was close or not? And was it generally anticipated how the outcome of the election would be? For example, was it clear that the elections of 1876 and 1916 would be close, and that the elections of 1912, 1920, 1924, 1928 and 1932 would result in a landslide? Were the results of any presidential election prior to the polling era surprising?
(V) & (Z) answer: The parties used to be much more aggressively structured, such that state-level leaders were in touch with county/parish-level leaders, and county/parish-level leaders were in touch with city/town-level leaders, and so forth, all the way down to people who were basically personally acquainted with every voter in their charge. In urban environments, these low-level, I-talk-to-everyone footsoldiers were called ward bosses or ward heelers; in rural environments the same task might be performed by some sort of party functionary, like a mayor, or maybe by a religious leader. So, it was possible to collect pretty good, pretty fine-grained information even without polls.
All of the Gilded Age elections were close, so there were no surprises there. The fundamentals of 1912 (Republicans effectively split into two parties), 1916 (Republicans are the majority party, but sitting Democratic president possibly on the cusp of a war), 1920/1924/1928 (Republicans dominant) and 1932 (Hoover blew it, clearing the path for the Democrats) were such that the nature of the election (close or blowout) was not a secret.
The election of 1856 was fairly dramatic, since the Democrats were the dominant party but were also flying apart at the seams. The election of 1864 was in serious doubt until September 2, 1864, when William T. Sherman captured Atlanta. The election of 1896 was a nail-biter in real time, even if it wasn't terribly close in the Electoral College. Probably the biggest "surprise" election was the one that took place at the dawn of the polling era, namely the 1948 "Dewey Defeats Truman" election. While they had polls, the polls were way wrong, which is why the outcome was a surprise.
J.P.R. in Westminster, CO, asks: Can you speak about internal polling? What secret sauce do the campaigns have that others do not, and does it make for more predictive models?
(V) & (Z) answer: The parties have extremely fine-grained information about individual voters, including things like how long they have been registered with [PARTY X], how often they have voted, what donations they have made, etc. So, the party pollsters can ask, and get pretty good answers, to questions like, "How do loyal Democratic donors feel about [2026 CANDIDATE X], as compared to [2024 CANDIDATE Y]." If [2024 CANDIDATE Y] lost their election by 1%, and the base is 5 points more favorable toward [2026 CANDIDATE X], that's a pretty good sign.
J.W. in Aston, PA, asks: How much do polls really sway an election, one way or another? The alleged fraudster Jack Sims was right-wing, but all of the polls leaned toward Democrats. Was the purpose to convince voters to stay home? There is something I am not getting here.
(V) & (Z) answer: It is hard to separate out one or two fake polls from all the rest of the background noise, but parties generally prefer that their candidate be moderate and that the other party's candidate be extreme. So, a fake pollster could try to elevate a more extreme candidate of the other party, or could try to prop up a more moderate candidate of their own party. There are certainly people out there who make decisions about how they will invest their time and their money based on whether the polls say a candidate is viable or not.
J.K. in Boston, MA, asks: Why would a Republican or Democratic pollster inflate numbers for their side? I get e-mail solicitations telling me how broke a candidate is even if they are flush with cash. I get others telling me how close the polls are, even if the candidate is ahead by double digits. I would assume it would be in the interest of the candidate to have the narrative be that they are neck and neck, or maybe just a little bit behind, to motivate their own side. But partisan polling operations inflate numbers to make them seem much more comfortable. Wouldn't that drive down donations and create complacency?
(V) & (Z) answer: The obvious answer is to weaken the opponent. If everyone believes that, say, the Democrat is up by 10 points with 2 weeks to go, then Republican donors might abandon the Republican candidate and might not bother to get to the polls. That could be more valuable to the Democrat than a few more donations or a few more door-knockers.
R.P. in Brooklyn, NY, asks: There has been a lot of discussion about state ballot initiatives allowing direct democracy and their positive and negative aspects. For example, ballot initiatives could be needed to pass laws where the interests of the lawmakers and citizens are not aligned. Could there ever be a binding nationwide ballot initiative, similar to a state initiative, and what would have to happen to make that possible?
(V) & (Z) answer: Congress could make this happen by passing a law. Then, Chief Justice John Roberts would get to decide whether Congress gets to delegate its authority in that particular way. It could also be done by a constitutional amendment, obviously. In fact, there is a group called National Initiative for Democracy that is trying to do just that. It was founded by former senator Mike Gravel (RIP).
However, speaking as California voters, this is a very, very bad idea. Once a population size gets to the point that you need 500,000, or 1 million, or 5 million signatures to cause an initiative to qualify for the ballot then the initiative process becomes the near-exclusive playground of billionaires and special interest groups. And someone like Elon Musk would think nothing of spending $500 million to secure passage of a law that says, for example, that electric vehicles will be eternally exempt from all taxation.
Presumably, your idea is to achieve certain kinds of reforms—say, overhauling the Supreme Court, or limiting presidential pardons—that Congress won't touch. But the solution here is to elect politicians who WILL touch those issues, not to open the Pandora's Box of federal ballot initiatives.
S.W. in New York City, NY, asks: Can you please clarify what "early vote" is, and is it the same in each state? Is it just voters who go to the polls early to vote? Does it include absentee ballots already received?
(V) & (Z) answer: The early votes include both the people who vote early in-person, and those who send in absentee ballots. Some states are pretty liberal in allowing such votes, some are more restrained, and some are positively draconian.
J.L. in Mountain View, CA, asks: There are quite a few states for which you have no Senate polls in your database: Colorado, Delaware, Idaho, Kentucky, New Jersey, Oklahoma, Oregon, Tennessee, West Virginia and Wyoming. And more with no polls in the last three months: Alabama, Louisiana, Mississippi and Virginia. With places like Kansas and Minnesota seemingly in contention, what are the odds that some of these result in surprises on November 3rd? Which states do you think are the most likely candidates for such a surprise, and why?
(V) & (Z) answer: There almost certainly will be a surprise or two on November 3, as there almost always is.
Our pick for a potential surprise is... South Carolina. The Democrats have a pretty strong candidate in Annie Andrews. The Republicans have a pretty weak candidate in Darline Graham. The state has a lot of Black voters, and they might show up in force if they think the blue team can actually win the election. And there's been an October Surprise, with the reports that former senator Lindsey Graham (R), brother of Darline, paid off three male escorts, including one who was underage (17) at the time of their alleged encounter. We do not know how that will play in culturally conservative South Carolina, but we do know that it's certainly not going to help Darline.
S.P. in Cranston, RI, asks: In my reading of Electoral-Vote.com, as well as some other reports on polling, I see a lot of senate and gubernatorial seats that are either shifting back and forth, or seem to have many results within the polls' margins of error. That has me wondering (and worrying) whether more elections should be rated as "too close to call" than you and other pundits are acknowledging.
(V) & (Z) answer: We don't know what you mean by "acknowledging." Our map is carefully color-coded to give a helpful overview at a glance. And any state that is not a solid color could most certainly go either way. That means that, at the moment, we have 12 states where the Senate race is too close to call.
T.S. in Seattle, WA, asks: We regularly see generic-ballot polls showing Democrats with a high single-digit/low double-digit lead. At the same time, Democratic senatorial candidates in light-blue states like Michigan, Minnesota and New Hampshire are struggling in what polling shows to be toss-up races. How can both of these polling results be true? If Democrats actually have a comfortable lead on the generic ballot, wouldn't these candidates be coasting to victory?
(V) & (Z) answer: The generic Congressional ballot is an assessment of how the country as a whole is feeling. It exists, by and large, as a way of projecting what is going to happen with the House of Representatives. A better way to project the House of Representatives would be to regularly poll all 435 districts—or, at very least, to regularly poll the 100 or so most competitive districts. But there is nobody who wants to invest the substantial resources needed to do that.
For Senate races, fundamentals matter. But candidate quality is also very important, and so too are local issues. In various ways, the Republicans have had good luck with candidate quality in those three states. They got the scion of a local political dynasty in New Hampshire, and they drew Democratic opponents with liabilities in the other two. Meanwhile, Donald Trump's policies have affected farm states particularly heavily, which means there are a large number of voters, including some former Trump voters, willing to express their pique through their ballots. The effects of those policies have been a little more muted in the less-farm-centered states of the northern climes.
E.S. in Miami Springs, FL, asks: I have a question regarding the Florida U.S. Senate polls, is it just me, or do I sense Angie Nixon's (D) support being rather underestimated, at least in polling? Granted, I could also be overly optimistic in favor of Nixon considering the Florida electorate, but at the same time she did pull off an upset win against her primary opponent Alexander Vindman, and nationwide the midterms are poised to be a blue wave, so who knows?
(V) & (Z) answer: The polls certainly did miss badly on the Democratic primary. And we have not seen a single explanation as to what went wrong. However, one possibility is that the pollsters underestimated Black turnout. Black people make up about 15% of the Florida population, so if that's what happened, and if it happens again, then Nixon might indeed be more competitive than the polls suggest.
N.P. in Ames, IA, asks: I was a bit surprised when you wrote that "Wedgewood at least helped derail the campaign of the centrist Mallory McMorrow." My question is that, since McMorrow was a serious candidate, how much influence would one really bad external poll have on a campaign? Wouldn't her team have internal polling and a ton of other data to spot this outlier, if it was an outlier?
(V) & (Z) answer: The thing that ends virtually all campaigns for political office is that the money runs out. The candidate herself said that the Wedgewood poll completely killed her fundraising, and thus left her unable to continue. She probably wouldn't have recovered even without that poll. But that poll seems to have taken away whatever sliver of a chance she had.
S.C. in Mountain View, CA, asks: You wrote: "A lot of people will be wondering if [Mallory] McMorrow [(D)] could have won had it not been for the fake Wedgewood poll."
I took a look at the individual (not aggregate) polling data listed by Wikipedia that was conducted before McMorrow suspended her campaign. The Wedgewood poll isn't listed, so I don't know when it was released, but beginning with the April 17-19, 2026, poll by the Glengariff Group, she was never above 20% and consistently in third place. Three pollsters (Mitchell Research & Communications, Susquehanna Polling & Research, and Quantus Insights) had her in single digits.
Hence my questions: (1) When was the fake Wedgewood poll released?; (2) Are the three pollsters listed above and their results legitimate, or were they cooking the books too?; (3) While it is egregious that Wedgewood published a fake poll, wasn't McMorrow cratering anyway, and wasn't likely to win the primary, or is there other evidence that, absent the fake poll, she could have turned things around?(V) & (Z) answer: The Wedgewood poll was released on June 29, and McMorrow suspended on July 5. She was already in deep trouble at that point, and it certainly appears the fake poll finished the job. It is not very likely that she would have recovered, but with that said, there are plenty of candidates who appeared to be doomed and yet came back to win. It's not common, but it does happen.
We will point out one other thing. Even if the Wedgewood poll was a minor part of what doomed McMorrow, that does not necessarily mean that some voters will not overstate its importance. Voters, as you may have heard, are sometimes willing to misrepresent the evidence.
D.M.G. in Valley Village, CA, asks: As a Mississippi expat, I am reading several stories about Mississippi being "in play" for the Democratic Party (something I never thought I would see in my lifetime). I also hear from old friends the dissatisfaction there. That said, I note that the last time Mississippi was polled for the Senate race, according to your site, was in April. I note the same for other states normally red but supposedly in play. I get it that these rural states have only a few representatives and in past years were considered hopeless for the blue team in the Senate, but without polls how can anyone know? Where are these stories coming from that these seats are in play?
(V) & (Z) answer: The publicly released polling in Mississippi has indeed been sparse, because nobody wants to pay for it. But those few polls that have been released show a very close race, with the gap between Sen. Cindy Hyde-Smith (R) and Scott Colom (D) around 4 points. It's certainly possible it's closer than that, since we have so little data. The very best evidence that Colom might have a chance is that the Democrats are now spending money there. They have internal polls they do not share with the general public, and they would not spend precious dollars if there was no point to it.
R.H.D. in Webster, NY, asks: How in God's name is Minnesota's Senate race so close?
This is the state that has the current longest streak of voting Democratic in Presidential elections (since 1976), gave us Hubert Humphrey and Walter Mondale, and was arguably the most affected by Trump's immigration reign of terror leading to two high profile deaths.
So why is it that while Sen. Amy Klobuchar (DFL) is expected to cruise to the Governor's mansion, at the same time Lt. Gov. Peggy Flanigan (DFL) is struggling against Michele Tafoya (R), who has ticked off hearty Minnesotans in her campaign, for the U.S. Senate seat?(V) & (Z) answer: Michele Tafoya started poorly, but has since bounced back, using her very polished TV persona in interviews, and carefully holding Donald Trump at arm's length. Meanwhile, Lt. Gov. Peggy Flanagan (D) has been damaged by the "welfare fraud" scandal that was manufactured by Republicans.
However, with that said, Flanagan has led in every single poll, including polls from Republican houses, except for one. And in that one, it was tied. Further, it is very possible that Democrats are using their poll responses to signal they are irritated with Flanagan ("I'm undecided!") even if they know they will ultimately vote for her.
V.G. in New York City, NY, asks: Your map has Seth Bodnar (I) ahead in the Montana U.S. Senate race. Most others (Nate Silver, FiftyPlusOne) have the Republican ahead by double digits. Can you explain this discrepancy?
(V) & (Z) answer: We are not secretive about our method. Click on the "Map algorithm" link in the legend box to the right of the map. We use the most recent poll, no matter how far back, and average it with all other polls within a week of it. And we do not use polls from partisan houses. There hasn't been much polling of this race, and most of the polling that has been done was done by Republican houses. The most recent poll that meets our requirements is the one that had the race as a tossup.
Do keep in mind that Alani Bankhead (D) is a wildcard that is very hard to account for. She is not going to win. And invariably, many people tell pollsters they will vote for a hopeless candidate, but then end up switching to a viable candidate when the time comes to cast their ballot. The number of Bankhead and/or undecided voters who end up as Bodnar voters is what will decide this election. And there are enough Bankhead and/or undecided voters that Bodnar certainly could pull this off if he can win most of them over.
M.J. in San Luis Obispo, CA, asks: Why does your Senate map still show South Carolina as "Strong Republican," when polls and even your own analysis show that race considerably tighter?
(V) & (Z) answer: Because we do not color-code the map based on vibes. We follow the rules we have laid out, as explained above. The more recent polls of the race, all but one of them from Republican houses, show that it's close (roughly, Darline Graham +2 points). However, the only poll that meets our requirements is the Abacus Data poll from August, and it has Graham with a huge lead. We hope there will be at least one or two nonpartisan polls of the state in the next week.
M.R. in San Diego, CA, asks: The most recent poll showed a 4-point spread in the Louisiana U.S. Senate race, but that was a month ago. Any idea why there's so little data from the Pelican State?
(V) & (Z) answer: The answer is pretty much always the same: Nobody is willing to pay for it. None of the universities in Louisiana has gotten into the polling business. The media in that state doesn't have a lot of spare cash lying around. And national outlets (e.g., The New York Times) clearly feel other states are of more interest to their readers/viewers. That said, in contrast to Mississippi, we haven't seen any coverage of the Democratic Party investing in the state. That's not a great sign for Democrat Jamie Davis.
J.I. in Hicksville, NY, asks: Louisiana and Mississippi appear close, but have not been polled in months. How would you handicap those races?
(V) & (Z) answer: If you made us bet, we would bet that the Republican hangs on in both places. That said, Mississippi (37%) is the Blackest state in the nation, Louisiana (33%) is second, and South Carolina (25%) is sixth. The Democrats' fates will be determined by Black turnout. And if that turnout is through the roof, well...
Put another way, if the Democrats manage to flip one of these seats, then there's a fair chance they flip all three.
A.B. in Eugene, OR, asks: Has no one looked at the South Dakota U.S. Senate race since the May 30 poll? The Democratic candidate dropped out in early August, leaving incumbent Mike Rounds (R) to face Brian Bengs (I). If Bengs picks up most of the Democratic votes, plus a few of the undecideds, it could be a nail-biter.
(V) & (Z) answer: Actually, since you wrote to us earlier in the week, there's been an Emerson poll. They have Rounds at 50%, Bengs at 42% and Undecided at 8%. That's not that close, but as we note above, Emerson tends to assume high Republican turnout. If they're off a bit on that, and if the majority of the undecideds break Democratic (and undecideds usually do break decisively in one direction or the other), then it could indeed be a nail-biter.
A.B. in Wendell, NC, asks: I used to live in Texas, in 2000-2005. I remember being very involved back then, so much so that I was a Deputy Voter registrar—you actually have to be deputized to do this in Texas. We were always talking about turning Texas blue... only to be disappointed every single time (well, except for Lloyd Doggett, he kept winning—and I remember how Lloyd would always tell me, as someone who is trans, "I support you, but that dog ain't gonna hunt in the House.")
Anyhow, do you think there is any chance the polling in Texas is more of the same wishful thinking, and we will be, once again, disappointed? I mean, I lived in Texas, and I know Texas. And there are a LOT of Texans who would vote for Satan Incarnate if he had an R next to his name on the ballot.
What's your take on the Texas polling and, if you really do think that James Talarico is going to win, based on the polls we have thus far, please detail why you think so. I know the polling looks good, but from my perspective, this is far from the first time Texas looked good, and we managed to snatch defeat from the laws of victory.(V) & (Z) answer: Texas elections tend to be determined, these days, by two things: guns and Jesus. James Talarico, in contrast to Beto O'Rourke, has not taken an impolitic position on the Second Amendment. Meanwhile, Talarico is a Christian pastor, while Paxton is very... un-Biblical, let's say. This may give just enough Republicans permission to cross the aisle, this one time. Indeed, if there is a shy voter effect (see above), we would actually guess it's Republicans embarrassed to admit they are voting Talarico, not Republicans embarrassed to admit they are voting Paxton.
We are not saying Talarico will win. But we are saying it's very plausible, more than it was for O'Rourke, Colin Allred or M.J. Hegar.
K.H. in Chicago, IL, asks: Can we get another picture or two of Tuco?
(Z) answers: This week, I was unable to finish grading essays because Tuco laid hands—well, paws—on my belt, chewing it up enough to damage it, but not enough to throw it away. So, I took a pair of scissors to remove the damaged end, but they slipped and I inadvertently sliced open the palm on my left hand. When I told the students, they wanted to see a picture of Tuco. So, I put one in the slide show for the Wednesday lectures.
While I was at school showing the picture to students, this is what he was up to at home:
He is a very naughty dog.