According to Rambam, it is natural for people to follow others when it comes to thoughts and wishes. To put it differently: people, on average, tend to be led. Rather than lead, let us just say. If one takes Rambam into account, one knows everything one needs to know about why pollsters, on the occasion of the 2024 US election, look pale to some (not everybody). Seemingly, they could not predict Trump’s victory..? Is that so?
Let us what the best forecasters said. Nate Silver is known, hardly requires an introduction. His magic? He aggregated all Trump vs Biden and Trump vs Harris polls he could lay his hands on, but, unlike other aggregators, like ‘Real Clear Politics’, he weighted them differentially. In weighting them he took into account not just the sample sizes but other characteristics that, in his eyes, could impact on the trend. Timing of surveys was taken into account, for example. This is not something that anyone else have done: conventions, assassination attempts, drop-outs of competitors all impact in short term but not in the long term. Etc etc, His final model was a DT/KH tie. A perfect tie.
What does one deduct from that? This is where Rambam comes in handy, you see. But not just him. In 2015 the United Kingdom had an election with an ‘unexpected result’, which was a victory of the Conservative Party. I say ‘unexpected’ because anyone I know actually expected the Conservatives to win and many of my friends in polling industry did (I will talk about them later). The permanently surprized British establishment went as far as a to launch an inquiry-‘why pollsters were all wrong?’. Leading statisticians were appointed to lead the Inquiry.
Results? You can read them here for yourself, but in a nutshell…. What the Inquiry chose to highlight as main factors accounting for the polling industry failure to predict the election results were: (1) ‘unrepresentative samples’ and (2) a degree of the ‘last minute swing’ towards the Conservatives. And there was something else, which was not among the listed ‘main contributors’ to the failure yet mentioned and well-investigated and not ruled out, called ‘herding’. I will tell you about this one later. The first thing, the unrepresentative samples, sounds severe but actually is not that damaging except when the results of a poll are expected to be very very precise. I wrote about it in the past, in the context of elections in Israel. Most polls today are based on panels rather than simple random samples. Panels are databases of people who agreed to take part in polls in return for some rewards or due to curiosity or as a way ‘to vent’, you name it. There is quite a bit of selectivity there: panel participants are, first of all, educated, opinionated, engaged. Their social opposites are left outside the panels. Classic polls based on simple random sample could have picked them up, but these polls are VERY expensive and nobody does them any longer, or almost nobody. The bottom line: the better educated and the opinionated may be better represented in the databases used by the pollsters, and they may be more Left-wing.
Also, they may way be more vocal in expressing their opinion. It has been documented many times, by Pew among others, if I remember well, that the Right wing voters (let us call them that) are less vocal, passionate, and (!) ready to share their views. Maybe because their lesser than average passion makes them less expressive, on average, or maybe they are a little unsure to voice an opinion that may be taken as unpopular in the ‘current climate’. This is our good old friend ‘social desirability’, meaning: not all opinions are equal and expressing some leads to punishment. The current climate in the UK is that being on the Left is OK and being on the Right is fascist. In my entire time in Cambridge, I have not seen one ‘Vote Conservatives’ sign in the pre-election run. Not one. I saw plenty ‘Vote Labour’ and a lot of ‘Vote Liberal Democrats’. (For Americans, Liberal Democrats is a third party in a two party system- a party that calls no real shots). Why are Conservative supporters silent? As per above- either they are the quieter, on average, bunch, or, just a thought, they are afraid of a brick thrown into the window. Some political commentators in my local pub are convinced it is the latter reason. If some, a small proportion of the Conservative voters, do not self-identify as Conservative voters in surveys, surveys will underestimate the amount of Conservative popularity. Simples. By the way, the anti-Conservative bias in British polls outlived all inquiries: it was still well-visible in the 2024 general election, the British Polling Council has made it crystal-clear.
Then there is ‘herding’ of pollsters. Which mean that pollsters know the truth, they see the real results (ties, or a small Conservative advantage) but they do not *report* it or report the opposite of what they find. How is this done? It is very simple. The political elite, on average, is a tad more to the Left than the Rest, at least rhetorically. And survey companies are run not by statisticians mostly, but by ambitious public figures, businessmen. To be sure, at a lower, working-horse ranks, survey companies are full of statisticians but statisticians do not matter for reporting, or , to be precisely, they are not the only ones who matter. Who does? The companies’ director friends and donors. Those heads of consultancies in this or that aspect or law and even the proverbial Lord Lilienblum of Foghorn whose wife is very displeased by the prospect of the Conservatives’ win. May be if the Conservative lean is not disclosed, that will help the anti-Conservatives whoever they are -she thinks naively. And so, donors and friends gently direct the companies directors who in turn direct statistically educated working-horses Cynthia and Tom. The donors’ desires are a law, or close to it. And so, some survey companies, a sufficient number, will say things in a way that they are not. Is it that stupid? Not 100%. In telling themselves a story why they should not report the Conservative lean the companies directors take into account the fact that some pollsters, a small number, genuinely find a small anti-Conservative lean, so the possibility exists. That strengthens them further. Hence the word ‘herding’, behaving like a herd, saying things that others are saying just because others are saying it. *The more important* others.
Cynthia and Tom wonder, in their private moments, whether all this ultimately works against them. After all, if they are ‘wrong again’, their reputation suffers. But Cynthia and Tom also know that elections happen once per 4-5 years whereas food is bought weekly and mortgages are repaid monthly. So they think of their table and mortgages first and doctor the findings. That amplifies the existing problem (shyness of the Right wing voter) further.
Who says all that? The contrarian demographer Daniel Staetsky? Hardly. Well, the British inquiry earlier and Nate Silver himself just recently. There’s more herding in swing state polls than at a sheep farm in the Scottish Highlands-as he humorously put it. Herding can be diagnosed rather easily by statistical methods. Cynthias and Toms warned about that too.
So, were pollsters wrong? No. They collected good data. Looking at the data that Nate Silver made available I, personally, saw a tie. And having seen a tie, in a climate where the Right wingers are shy and herding takes place, I decided that Trump will take it. And so did -personally known to me and appropriately anonymised-Cynthia and Tom. I will tell you more than that. Having considered the situation in all its gravity 2 days before the election Cynthia decided to buy Tesla shares. Tesla is Musk’s thing and all things Musk were expected to go up if Trump wins. She bought them at 242 and sold them a little later at 278. She needed a new sofa for a while. Several days after the election she counted her till and decided she could have it in leather.
Pollsters are not as useless as they seem. They win when allowed to. Be like Cynthia.