Original article (in Slovenian) was published on 25/10/2024; Author: Pia Zala Meden
Jaka Smrekar, professor of statistics at the Faculty of Mathematics and Physics, explains that the study mentioned in the N1 article has several shortcomings, including a small sample size of monkeys, and that its effect is modest because the monkeys’ predictions were correct 54% of the time.
N1 published an article on 7 October entitled Could monkeys predict the winner of the US election? Scientists say: Yes, which claimed that a team of scientists had concluded monkeys could predict the winners of US elections.
According to the news portal, the findings of a new scientific paper show that people follow their primal instincts when voting, preferring to pick leaders with more masculine features that reflect their position of dominance in society.
They were referring to a paper published on 19 September on arXiv, a server of scientific papers that have not yet been peer reviewed which is operated by Cornell University in the US. N1 did point out that the paper was not peer-reviewed.
The authors of the paper, from the fields of biomedicine, psychology and neuroscience, analysed whether monkeys can predict the winner of an election by observing the facial features of candidates. They defined a manly face as one with a more pronounced jawline and less prominent cheekbones. They studied the behaviour of macaque monkeys and found that, like humans, they respond spontaneously to evolutionary cues of masculinity in the faces of political candidates.
Their study involved six male monkeys. Three of them were presented with 124 pairs of candidates from past US gubernatorial elections and 149 pairs from elections to the US Senate, and the other three observed 13 pairs of former presidential candidates over five sessions. The scientists assumed that the monkeys voted for the candidate they had watched the longest.
The monkeys correctly predicted the outcome of the gubernatorial and Senate races 54.6% of the time, with a 2% margin of error. They picked the actual winner of gubernatorial and Senate races 58% of the time, and the winner of the presidential election 50.4% of the time, with a maximum margin of error of 5.7%.
Jaka Smrekar, professor of statistics at the Faculty of Mathematics and Physics, University of Ljubljana, explained to Razkrinkavanje.si that the statistical results in the N1 article refer not to the sample of monkeys, but to pairs of candidates.
He pointed out that the sample of candidates was large enough for ‘normal’ statistical inference, but the sample of monkeys was very small, which is “one of the several shortcomings of the study”. He also pointed to the importance of statistical significance and impact in assessing the credibility of the results. Statistical significance is the probability that a difference between two groups is due to chance, while effect size refers to the differences between groups and helps to understand them.
“Assuming that the study was correctly conducted and the statistical statements are correctly generalised, we could say that the monkeys predicted the winner with a probability of more than 50%. If it is actually about 54%, the effect of the study is modest.”
As for the survey’s correlation coefficients, which reflect the degree of correlation between variables, he explained that “although they are significantly non-zero, they are so close to zero in terms of effect that we cannot even begin to talk about a linear correlation”.
N1 explained that the article had been updated accordingly after they received questions from us, and that this had been marked in the text. They added that they had reproduced an article published by IFL Science, a science news website, authored by a woman who holds a PhD in mathematics.
According to IFL Science, that author specializes in dynamic systems and number theory.
N1 corrected the statement about the proportion of outcomes correctly predicted by monkeys and the number of monkeys participating in the survey, but not the title of the article, which provides an affirmative answer to the question whether monkeys can predict election outcomes.
The claim that scientists are saying that “monkeys may be able to predict the winner of the US election” is unsubstantiated. The data from the scientific paper, summarized in the N1 article, is not scientifically credible. The result of 54% correct predictions is too close to chance, according to the statistician, to claim that monkeys can predict the outcome of elections, and the sample of monkeys was too small.
