That Conservatives Smell Different Than Progressives Study Stinks

Several readers asked me to examine the peer-reviewed study “Assortative Mating on Ideology Could Operate Through Olfactory Cues” by Rose McDermott, Dustin Tingley, and Peter K. Hatemi appearing in the American Journal of Political Science.

It’s showing up everywhere. Like the Powerline blog, also in Pacific Standard, and The Week, and of course The Washington Post.

The paper opens with a proposition which is surely false, though it is asserted by many authorities, “Similarity between spouses is common across domains, but in humans, long-term mates correlate more highly (between 0.60 and 0.75) on social and political attitudes than almost any other trait, with the exception of religion”. I would have guessed race, followed closely by geographical and age. But skip it.

“Olfactory mechanisms have proven important in mate seeking and reproduction in both humans and animals because smell may signal mate immunocompetence, social compatibility, or other characteristics associated with mate quality and optimal reproduction.” Cue hippy joke #1.

While you’re at it, cue the amygdala, too, a pea-sized pair of brain “organs” which, as near as I can tell, account for every single human behavior that can be studied by academics. Our authors call to it here, too.

Here’s the question: “Why and how might smell signals be linked to political ideology?” Well, smell helps “maximize prospects for disease avoidance, cheater detection, defense against out-groups, and social cohesion”. Cue hippy joke #2.

Hey, did you know that “greater disgust sensitivity, which is intimately interconnected with the neural substrates of smell, predicts more conservative positions, particularly around issues involving morality and sexual reproduction”? If this is true—and it is peer reviewed—it must imply that conservatives excel at sniffing out the cheesy arguments of their unwashed opponents.

Whatever it is is in the genes, too. “Suggestively, Hatemi et al. (2011) identified several genomic regions that account for variation in ideological orientation, one of which contained a large number of olfactory receptors.” That “suggestively” actually means, “Oooh, we hope it’s true!”

Tying it all together:

If social attitudes are linked to odor, as the literature suggests, then one mechanism that odor preferences transfer from parents to children may operate through their mother’s choice of mate. In this way, social processes may drive some of the pathways by which individuals come to prefer those whose ideological “smell” matches their own.

The Just-So stories having ended, we proceed to the experiment itself, in which “participants rated the attractiveness of the body odor of unknown strong liberals and strong conservatives”. Ten lefties and 11 righties “provided body odor samples” (“menstruating or pregnant women were excluded”). One other sample was excluded: I couldn’t discover whether it was a leftie or rightie.

Affiliation was decided by a questionnaire (7 point scale), and so was smell (5 point). On an 11-point scale, dear reader, rate your belief in the validity of these “instruments”.

Anyway, the noses went to work. Data was collected. People were thanked. Now it is here is where you would expect a simple summary, maybe some pictures, showing the distribution of smelliness broken down by the evaluators’ and targets’ political affiliations. Did the raw data reveal that leftie evaluators prefer the smell of target lefties? And did conservative evaluators dislike the smell of leftie targets?

Alas, we shall never know. For why use actual data when you can have a statistical model instead? Actually three models. Regression of course (one logistic, two Gaussian). As if the uncertainty in a 5-point smelliness scale is well approximated by a normal distribution. And what’s with shoehorning in strange terms like “absolute value of the difference between the target’s and evaluator’s ideology, multiplied by negative 1” and “Avg. Target Attract” and “Avg. Eval. Attract”?

No wee p-values for the logistic model, but why worry when the results are “consistent with our theoretical expectations.” Sadly, consistent with is now our highest standard of evidence. Skip it.

Oh, wait, now I get it. The “absolute value of the difference between the target’s and evaluator’s ideology, multiplied by negative 1” is what had the wee p-values, and an effect size of about 0.02. That’s for a change on a smelly scale of 1 to 5. And don’t forget the maximum political difference can only be 6, so that the maximum effect can only be 0.12. At best: if the model is good.

That’s it. That’s the study. That wee, which is to say, trivial effect confirmed by a wee p-value, all wrapped up in an inappropriate model. But the authors still say “individuals find the smell of those who are more ideologically similar to themselves more attractive than those endorsing opposing ideologies”.

The authors added a few hundred more words in an attempt to escape the obvious. They never make it. But the press believed they did.