What if the electoral survey polls were wrong?

Blog: JOSEP COLOMER'S BLOG

Abstract

The trend of recent elections does not support the widespread prediction of Donald Trump's victory in November. The bias in election forecasts may result from the over-polling of Trump's voters in electoral surveys.Let us look, first, at recent elections. The electoral cycle of special, state, and local elections from early 2023 has been seen as a success for the Democrats. Candidates from the Democratic party won in three of the four special elections for the federal House of Representatives. They also increased their representation in two state legislatures, made a net gain of four mayors, and improved positions in state supreme courts and other offices. If this is the current trend, the Democrats should be optimistic about the upcoming presidential and congressional elections.At the same time, the support for Donald Trump may be weaker than it may have appeared in the ongoing Republican party primaries. Let us focus on the fifteen states that held them on Super Tuesday. The total number of voters turning out was one-third of the votes for the Republican party in the general election four years before (9.1 million out of 27.2 million or 33.5%). Typically, the primary voters are more extreme in their preferences than their party's voters and even more than the whole electorate including independents. The Republican primary voters are older and much more conservative than the rest of the party voters. In eleven states with closed party primaries, nearly one-half of the participants were Evangelicals.Let us now turn to the inaccuracy or simply the failure of survey polls. The average of six well-reputed pollsters for Super Tuesday was that Trump would receive 79% of votes while Nikki Haley would get only 15%, so 64 points of advantage for the former president. The actual results were quite different: Trump got 71% and Haley 26%, "only" 45 points of advantage.The contrast between survey poll predictions and actual results cast serious doubts about the forecasts for the elections in November.One of the main suspects is the technique of running survey polls by cell phone. In the past, surveys were based on in-person interviews, in which the interviewers knocked on thousands of doors across the country. Starting in the 1980s, most interviews were developed over the phone. Currently, most surveys are run through cell phones.Historically, US telephones were "landline" devices and phone numbers were accessible from printed telephone directories, from which pollsters selected a sample assuming that it represented a large proportion of the population. Today, there are far fewer landline telephones, millions of cell phones, and scarce published directories.Many people with cell phones do not answer calls from unknown numbers, a precaution against telemarketing and scams. David Hill, president of a major research consultant, acknowledges that "whereas once I could extract one complete interview from five voters, it can now take calls to as many as 100 voters to complete a single interview." And "people who tend to pick up any and all phone calls tend to be older, conservative, less wealthy, and less educated", a profile that fits a typical Trump voter.As people with extreme political opinions tend to be more eager to participate in party primaries than most voters, they also tend to respond to more surveys on their cell phones. Some researchers hold that "there is currently no way to select survey samples in a way that covers the entire population." Many surveys are not based on random samples, but on "polls of the willing." A few hundred people who were willing to answer a random phone call make a survey that is misinterpreted as an expression of the national electorate.In 2016 it was said that there were "secret" or "hidden" Trump voters who did not respond to the polls, so the surprise of the election result. Now some analysts hypothesize, analogously, that there are "hidden" anti-Trump voters who elude to respond to the polls. But it may be more reasonable to suspect that today there are too many Trump voters who are over-polled on landline and cell phone surveys.The bias of survey polls in favor of Trump may be both good news and bad news. Good news because it may mean that Joe Biden is ahead in the voters' actual preferences. Bad news because if the survey polls bias continued and Biden won, Trump might launch an even stronger denial than in 2020. This time, however, the administration and the police would still be in the hands of the Democratic incumbent. COMMENTSI hope you are right.Rein TaageperaUniversity of California, Irvine, and University of Tartu, EstoniaI think you're onto something.  I have a piece coming out in Sabato's Crystal Ball on Tuesday raising doubts about these polling numbers, especially among Black voters. Alan Abramowitz        Emory University Excellent piece. Thank you.Gianfranco PasquinoUniversita di Bologna & Johns Hopkins UniversityEstoy muy de acuerdo -- llevo tiempo defendiendo esta idea en mis conversaciones y correos con amigos.  Es muy difícil saber cual es exactamente la magnitud de este fenómeno pero parece quer sí que es muy real.  Confio en que Biden va a ganar en noviembre.  Probablemente por un margen mayor que en 2020.Una abraçada,Robert FishmanUniversity of Notre Dame