In 1936 a magazine counted nearly 2.4 million ballots and named the wrong winner. Politicians in five democracies misread the public in the same direction, and those who voice their opinions may create the bias.
11 October 2026
In the autumn of 1936 the Literary Digest mailed postcard ballots to about ten million Americans and asked how they would vote for president. Nearly 2.4 million sent them back. The magazine had called every presidential election since 1924, and this time its count gave the Republican, Alf Landon, a clear win over President Franklin Roosevelt.
Roosevelt won just over 60 percent of the vote and carried 46 of the 48 states. Two years later the Digest was out of business.
The story usually told in statistics classes is that the mailing list was to blame. It came mostly from telephone directories and car registrations, and in the Depression only the comfortable had either.
The people who studied it later found something else. In May 1937 Gallup's organization asked a national sample how they had voted, and whether they had received a Digest ballot and sent it back. Car and telephone owners turned out to be for Roosevelt too. The difference lay in who replied. Those who mailed their ballot back favored Landon. Those who did not favored Roosevelt, by about three to two. The political scientist Peverill Squire, who analyzed the survey in 1988, found that "if everyone who had received a [Digest] ballot had returned it the results would have, at least, correctly predicted Roosevelt a winner." He still judged the mailing list the larger fault. A 2012 reanalysis by Dominic Lusinchi put non-response first, and both agree that had everyone replied, the Digest would have named Roosevelt.
The Digest's problem was never the size of its sample. It was trusting that the people who chose to answer spoke for the people who did not.
Every politician runs a Digest poll, and it never closes. It is made of the people who call the office, write in, turn up at meetings and stop the candidate in the street. A random draw of voters would mirror the electorate. A channel that people choose to enter has no reason to.
In 2012 and 2014 the political scientists David Broockman and Christopher Skovron measured how far politicians' picture of their voters was off. They asked 3,765 American state legislators and candidates what share of the people in their own district supported a policy, and compared the answers with what the districts actually thought.
The largest error was on background checks for all gun sales. In the districts of the Republican politicians, 84 percent of residents supported them. The Republicans put support at 42 percent. In the Democrats' districts, support was 85 percent, and the Democrats put it at 53 percent.
Both parties guessed wrong in the same direction, on almost every issue. Democratic politicians overestimated support for the conservative position on nine of the ten questions (nine issues, with same-sex marriage asked in both years); Republicans on all ten. The Republicans' errors were larger every time, on average more than twice as large by my own arithmetic from the paper's table, and the authors call the politician's party "by far the most important determinant" of how wrong the estimate was.
When legislators ignore a majority, the first explanation people reach for is money. Money can explain a vote, but these were beliefs, and it is hard to imagine a donor paying Democrats to think their voters lukewarm about background checks. They thought it anyway.
Broockman and Skovron then looked at who reaches out to politicians. In two national surveys, Republican citizens were 28 to 39 percent more likely than Democrats to say they had contacted their member of Congress.
That tilt adds up. By the paper's count, half the residents of a congressional district held by a Republican call themselves Republicans, yet Republicans made up 62 percent of the people in those districts who had contacted their representative. Democratic members of Congress heard the same tilt, more faintly. These are figures for Congress: the authors know of no data on who contacts state legislators.
Guns are where both halves of the story meet: the politicians' largest errors, and a sharp imbalance in who speaks up. The paper's second gun question was a ban on assault rifles. In the districts of both parties, 57 to 58 percent of residents supported it. Republican politicians put support at 31 percent, Democratic politicians at 44, so on average both thought a majority measure was a minority one.
Then look at who contacts politicians about guns. Gun owners are about three adults in ten, and members of the National Rifle Association about six in a hundred. According to the Pew Research Center, gun owners are more likely than other Americans to have contacted a public official about gun policy, 21 percent against 12, and the Association's members three times as likely as other gun owners. A 46 percent contact rate among the Association's members sounds like a flood, but it comes from a small group. On Pew's figures, most people who have contacted an official about guns do not own one, and about half of all those who get in touch want stricter laws. Those who want looser laws are over-represented, but they make up about a quarter of the people who get in touch, not most of them.
So a national head count of who makes contact does not, by itself, explain why legislators thought a majority measure was a minority one. A single district can look different. Republican members of Congress hear mostly from Republicans, and among Republicans, those who want looser gun laws outnumber those who want stricter ones, 30 percent to 24. Something else may also tip the scale: how often the same people get in touch, how hard they press, how well they are remembered, or sources beyond the mailbag altogether. Pew offers one hint. Among gun owners, 19 percent of those who want looser laws had contacted an official in the past year, against 9 percent of those who want stricter laws. The study's authors stop short of a verdict: contact "could help explain" why perceptions on guns are so skewed. Nobody recorded what those contacts said, how often they came or how strongly they were put.
The partisan tilt in who makes contact also looks recent. In the national election surveys of 1980 to 1992, Republicans and Democrats did not differ in how often they contacted their representatives. Older studies of what politicians believed about their voters, from 1958, 1992 and 2000, show no conservative overestimate, though the authors warn that those studies "have significant limitations." They call contact "a potential mechanism" and say their evidence for it "must be regarded as tentative."
In 2023 a team led by the political scientist Jean-Benoit Pilet reported a similar survey of politicians in Belgium, Canada, Germany and Switzerland, four democracies with very different political systems. They interviewed 866 elected representatives in 2018 and 2019. Of those whose answers could be scored, 81 percent overestimated how conservative the public was, averaged across the questions. Depending on the country or region, the share ran from 64 to 92 percent.
One thing was different from America: right-wing politicians did not consistently misread the public worse than left-wing ones. The team also asked something the American study had not, and the bias held even when politicians estimated the views of their own party's voters, the people they might be expected to know best. The exception was immigration: in Canada and Germany, politicians thought the public more liberal on it than it was.
Pilet's team also tested the American explanation, and found that differences in who makes contact do "not seem to consistently explain" the bias there. In these four countries, citizens with conservative views were not consistently more likely than liberals to contact a politician, except on pensions. Whatever skews the picture there, it is not a simple head count. The authors suggest that conservatives who do speak up may express themselves in ways that are "more extreme, and therefore more memorable," or that conservatives tend to overrate how common their views are while liberals underrate theirs. They also list business interest groups and politicians' greater attention to affluent citizens. None of these ideas has been tested.
In all five countries, politicians on average thought the public more conservative than it was, and outside the United States the same held for their own party's voters.
"Politicians read accurately the voters who matter to them." Pilet's finding on politicians' own party voters answers this for the people who vote for a politician's party. Nobody has tested the narrower version, the voters in a party primary. Broockman and Skovron also found the same pattern when they compared politicians' estimates with the opinions of voters alone, and report that voters and non-voters hold "nearly identical average opinions on the issues we study."
"The 'actual opinion' comes from a model." The background-check figures above do not. They come straight from weighted survey answers in the politicians' districts, with no model in between. Pilet's team used separate surveys of citizens in each country.
"Then politicians should commission polls." This is the hard one. In Broockman and Skovron's data, politicians who took more polls "were not significantly less error prone," and most took none at all. Nobody yet knows why polling did not help.
The Digest's editors trusted their poll because it was enormous. Nearly 2.4 million ballots looked like a verdict, but size only measured how many people had chosen to reply.
A politician's mailbag feels like an answer for the same reason: the people in it are real, and there are a great many of them. In America the mailbag appears to lean conservative. Elsewhere nobody has yet found where the lean comes from, yet the picture politicians hold is still more conservative than the public.
A legislature that ignores what most voters want may have been bought. It may also be counting the replies. That second failure needs no villain, which makes it harder to fix: there is nobody to expose, only a sample to distrust. The Digest learned that from nearly 2.4 million postcards, too late. Politicians have not yet learned it from theirs.
The ideas, the argument and the conclusions in this essay are mine. I develop each essay in conversation with an AI assistant: I set the question and the direction, it turns my thoughts into prose, and I review and rework every draft until it has the shape and form I want. Every figure is sourced, and the sources are listed below.