Friday, 15 August 2008

EP debunking Friday - sniffing out the perfect partner

As people keep moaning about the length of these debunking posts, I'll keep this one short. Browsing Feminist Philosophers I was reminded of this story from a few days ago:

"They say that opposites attract. But it seems the Pill may be preventing women sniffing out men who are opposite enough...But researchers found that the Pill disrupts a woman's power to recognise the aroma of a suitable partner."

This story is based on this study, "MHC-correlated odour preferences in humans and the use of oral contraceptives" an epub in Proc. R. Soc. B. The reasoning behind the study is that the major histocompatibility complex (MHC) of our genome shows great variation and this is thought to be related to immune function, so it would be genetically better to have offspring with individuals with genetically dissimilar MHCs from you.

So what does the study do? It looks at a bunch of women, none of whom were on the pill, but some of whom were thinking about taking it. So they got these women to smell the shredded sweaty shirts of men with either genetically similar or dissimilar MHCs:

"Almost 100 women then sniffed the shirts and gave their opinions on the 'pleasantness' and ' desirability' of the odour twice over a three-month period. Many started taking the Pill during the experiment - and their opinions of the smell of the T-shirts changed...Researcher Craig Roberts said: 'The results showed that the preferences of women who began using the contraceptive Pill shifted towards men with genetically similar odours.'"

So this is supposed to be evidence that:

"Saddled with the wrong man - someone who in scientific terms has similar genes - she may find it hard to become pregnant and any children she does have may have a lower resistance to infection.

What is more, when she stops taking the Pill and her sense of smell returns to normal, she's more likely to fall out of love, the Liverpool and Newcastle universities research suggests."

So is this what the study actually shows? Of course not. Without even getting into the question of whether woman actually choose partners based on the smell of their sweat, in actual fact the study found that women didn't rate the pleasantness, desirability, or intensity of the odour of men with dissimilar MHCs as any higher than men with similar MHCs.

The only significant finding was an interaction on an ANOVA where women on the pill showed a reduction in the amount they favoured MHC dissimilar odours, but there was no statistically significant difference between women on the pill and women not on the pill in the degree to which they favoured genetically dissimilar men. The interesting finding is that women not yet on the pill, but who intended to go on the pill seemed to favour MHC dissimilar men more than other women (presumably this wasn't statistically significant) and this is what is driving the interaction (the preference for dissimilar men normalises back to being like other women after they are taking the pill (see the figure, average odour 'desirability' for MHC dissimilar men minus MHC similar men, white bar represents first rating {before women started the pill} and the grey bar the second rating {with women in the pill group having started the pill}).

So it is pretty unlikely that taking the pill will cause problems becoming pregnant or children with a lower resistance to infection, since women not on the pill don't seem to favour MHC dissimilar odours (and thus men?) anyway, and women taking the pill don't show any difference in their preference for MHC dissimilar men compared to women not on the pill anyway.

The Mail does carry a contrary opinion from Professor Bill Ledger (Sheffield University) who wonders whether the smell of sweat is likely to overide the 'intellectual and emotional feeling' of a relationship but curiously doesn't attack the methodology of the paper. I wonder whether this is another example of embargoed press releases where those commenting on the paper don't actually get to see it and point out the obvious flaws.


LemmusLemmus said...

I've only read the Daily Mail article and the abstract of the study. It's a close call, but I'll say the abstract is an even worse piece of prose:

"Women using oral hormonal contraceptives have been reported to have the opposite preference, raising the possibility that oral contraceptives alter female preference towards MHC similarity, with possible fertility costs."

Um, didn't we have reason to believe there are "fertility costs" to taking the pill anyway?

"However, single women preferred odours of MHC-similar men, while women in relationships preferred odours of MHC-dissimilar men, a result consistent with studies in other species, suggesting that paired females may seek to improve offspring quality through extra-pair partnerships."

That looks like jumping to conclusions to me. The bit about single women also seems to contradict the hypothesis they outline in their first sentence.

I'm not against EP at all, but as in any field, quality varies.

pj said...

Oh yeah, I didn't talk about the data-dredging to find differences between single and married women, but, as you say, if it is only married women that favour dissimilar MHCs (for extra-pair copulations, of course) and these women already have partners, it is hard to see what deleterious effects on the quality of offspring are going to ensue from taking the pill, over and above the infertility that they're after!

I'm not against EP in theory, its just the practice that seems to be so poor.

LemmusLemmus said...

For some good EP, see David Buss's work on jealousy and Daly and Wilson on stepparents and child homicides.

If you want to make "EP debunking Friday" a regular feature, don't miss the work of Satoshi Kanazawa - although I'm afraid Andrew Gelman has already reaped the low-hanging fruit.

pj said...

I can only vaguely rememebr the Daly and Wilson stuff with step-parents but seem to recall being unimpressed. From what I recall, the argument that killing step-children is adaptive behaviour is pretty weak given the incrdibly low homicide rate in step-parents.

Woobegone said...

Clearly this study is hardly EP's finest hour. But there's a lot of excellent work being done on the attractiveness of human faces, and the results cry out for an EP interpretation (although they could be explained away in other terms, it would be a real stretch).

Basically, facial features correlated with being an 18 year old girl are attractive in female faces, although youth is not particularly attractive in men. Meanwhile any kind of atypical features, or too much asymmetry, are unattractive in both men and women. Given that female fertility declines with age while male fertility doesn't so much, and given that unusual features are generally a sign of ill-health or mutation, this is an EP fan's dream. Even better, these preferences have been seen in all cultures studied so far, and newborn babies prefer to stare at attractive faces just as adults do.

LemmusLemmus said...


if I remember correctly, the argument is not that killing stepchildren is adaptive. Rather, it is:

1. Violence is a normal strategy in the human toolbox.

2. All other things equal, violence against people who are genetically related makes less sense in evolutionary terms.

3. Stepparents who live with their stepchildren are an ideal test case for this.

Of course, most stepparents don't kill their children (and the theory doesn't predict this), but data presented by D&W and others show that they do so at rates that are orders of magnitude higher than the rates for biological parents.

pj said...

Woobegone, I think you're falling into the trap that EPers like to set - just because evolutionary explanations can be valid, it doesn't mean that any given one is valid.

I'm sure there are good evolutionary explanations for attractiveness and sexual drive - there has to be - but most of the time the EP evidence is piss poor and the willingness to speculate far from the evidence overwhelming.

For instance, in that article you posted, the evidence that attractiveness has anything to do with mate quality is pretty limited:

"Facial attractiveness and some of its components may have modest associations with health, although the evidence is far from overwhelming."

pj said...


of course, but the explanation that "research concerning animal social behaviour provide a rationale for expecting parents to be discriminative in their care and affection, and more specifically, to discriminate in favour of their own young" (D&W from wikipedia) is not particularly well supported by endless studies and speculations that a higher rate of homicide (or violence) by step-parents is due to discriminating in favour of their genetic offspring rather than a myriad of other potential explanations (e.g. relationship strife, decreased attachment due to shorter exposure).

The difference between EP and other sciences is the relative paucity of data, particularly good quality data, but the relative freeness with which the EPers make sweeping claims about what is 'adaptive'.

LemmusLemmus said...


well, assuming they're not lying to us, D&W developed their hypothesis on the basis of EP reasoning, no one had ever looked into this factor, and their hypothesis was confirmed. In principle at least, one could test for alternative hypotheses such as shorter exposure. I certainly would like to read such a paper, although I'm afraid the data would be hard to come by.

Woobegone said...


OK, the evidence that attractive people have better genes is shaky, and I wouldn't want to push that claim too far although I'd be surprised if there's no correlation.

But there has to be some reason why people all over the world find certain facial features attractive and others unpleasant. As far as I can see the EP explanation is the only convincing one. Perhaps a general symmetry-preference could explain why we like symmetrical faces, but I can't see how else you can explain why men prefer faces with features associated with being a young woman.

Like you I've got no time for the kind of EP which consists in coming up with explanations for various features of modern life e.g. I find the attempts to provide an evolutionary rationale for homosexuality, such as the "gay uncle theory", to be rather embarrassing). But equally I'm not one to immediately cry "Just so story!" whenever I hear an evolutionary explanation.

Judge every case on its individual merits and all that, and I think that facial attractiveness is a very good example of the right kind of evolutionary psychology.

Paul said...

"but most of the time the EP evidence is piss poor and the willingness to speculate far from the evidence overwhelming"

I'm really glad you said that. I've just completed my doctoral thesis which involved reading a lot of EP stuff. I've had this suspicion myself a lot of the time.

What's more, the tendency to argue from theories of adaptivity to what we ought to be doing or how we ought to live our lives is also alive and well. This is perhaps justified sometimes but the is-ought problem rarely seems to trouble EP investigators.

pj said...

woobegone, I agree, but EP has told us little. If you'd asked people before EP they'd have posited some interaction between culture and biology underlying concepts of attractiveness. And the biological factors would be thought to have at least a partial adaptive explanation.

Without further evidence, and experimental or detailed epidemiological evidence is something that the field of EP conspicuously lacks, it is just a more plausible just-so story than many of the other EP just-so stories. Being a priori plausible doesn't make something scientific. Plausible adaptive just-so stories are two-a-penny in any field, the point of science is to provide convincing scientific evidence.

pj said...

lemmuslemmus - actually I think that the relationship had been posited before D&W. My problem with their work, like that of many other EPers is that it may be hard finding evidence to test the theory and its alternatives, but that is why science is hard. People in every other field seem to be able to do it. But EPers often seem to concentrate on finding fairly poor database type population level evidence to support a usually fairly widely accepted phenomenon, which they then explain in adaptive terms (often fairly crude adaptive terms too). The amount of effort to actually get into the details of the data, perhaps conduct some case control studies even, doesn't seem to happen and they skip off to the next phenomenon to explain in adaptive terms. I'm afraid that just isn't science, that is just-so story telling.

These hypotheses are also very very weak from an explanatory pespective, with, for example social modifiers being deployed to explain away contrary evidence (such as the lack of a 'Cinderella effect' is Sweden), while other social modifiers that affect the putative adaptive phenomenon are ignored (e.g. the protective effect of marriage).

Finally, the EPers are always very vocal about their rather unconvincing results and far too keen to see their conclusions put into practice in influencing public policy given how damn weak the evidence that underlies them is.

pj said...

Another great study I've commented on is here at "Shake that booty". It highlights the way that EPers seem to play very fast and loose with evidence for and against Ep positions, with hypotheses framed so loosly that essentially any finding is interpreted as supporting the theory, while negative results are brushed aside.

LemmusLemmus said...


your mission, should you choose to accept it, is to prove D&W wrong. That's how science is supposed to progress, no?

I've never heard of an EPer trying to influence public policy, but then, of course, I don't see and hear everything.

I never read the "sexy walk" study, but if your representation of it is any accurate, that's indeed shoddy, especially the explanation (if I remember correctly) that fertile women walk "less sexy" in order to avoid getting raped. I mean cooome on...

pj said...


apart from the fact that I do not do that sort of research, I don't see that D&W have enough evidence for their hypothesis to make it worthwhile trying to test it. Currently all there is is an observation and a putative explanation. That is not a scientific theory, that is, as I've said, a just-so story.

To take a single example, Temrin et al produced data from Sweden that contradicted D&W's findings from the US, Canada and Britain (undermining the evolutionary hypothesis and suggesting that cultural factors are important).

D&W rightly queried the methods used by Temrin et al, and reanalysed the data for 1-4 year olds only (which is where they found there was a step-parent homicide effect in their other studies) with:

2% in genetic + step households
11% single genetic parent
87% two genetic parents
1% neither genetic parent

The homicide numbers were 53 by genetic parents and 4 by stepparents.

They then go on to analyse homicide rate by 'dyads' (that is pairs of relationships) concluding that 99% of dyads are genetic, and 1% are step relationships - and doing a binomial test to suggest that 4/57 is significantly greater than chance.

However, I would disagree rather strongly with their analysis method which essentially represents double counting every two parent genetic family (remember their hypothesis here is that the step-father is doing the killing) and fails to identify the perpetrator (this has been a criticism of much of D&W's work).

Temrin et al expanded their Swedish dataset and found that:

"We therefore use two categories of stepfamilies
in our analyses. In the first (stepfamily1), we
include only victims that were actually killed by the stepparent.
In the second category (stepfamily2), we include
all victims who lived with a step-parent and a genetic
parent irrespective of who was the offender.
For children aged 0–15 years old there was no overrepresentation
in the relative frequency of stepchildren as
victims compared with victims in families with two genetic
parents in either of the two stepfamily categories (table 1).
For the youngest age class (less than 5 years old), there
was a significant over-representation for victims in the category
stepfamily2 compared with victims in families with
two genetic parents (table 1). However, when we compare
victims in the category stepfamily1, i.e. where the stepparent
was the actual offender, with victims in families
with two genetic parents there was only a tendency in the
same direction (table 1)."

i.e. while there was an increased rate of infanticide in step-parent families for the under 5s there was not a significantly increased rate of murder by the step-parent (stepfamily1), but there was a significantly increased rate of infanticide in general (stepfamily2). They also found children living with a single genetic parent were much more likely to be killed too (but this wasn't part of the main analysis). There was also a (non-significant) trend for the step-parent to kill their own genetic children in preference to their step-children (again contrary to D&W).

So 94/100 murders were genetic parents from a two-genetic parent family, 3/100 genetic parents from a genetic/step family, and 4/100 step-parents from a genetic/step family (including one killing involving the genetic parent and the step-parent together) and in one of these killings the killer was probably more like the mother's boyfriend than a step-parent (which could get the rates to be 3 killings a piece!) If we choose to analyse by dyads like D&W do then we get a significant increase in killings by step-parents (4/100 when they make up 1% of dyads) but then this is also seen for genetic parents in genetic/step families (3/100, also 1% of dyads) (this dyad analysis is incorrect because data is not available from single parents - which represented a third of killings in the dataset as a whole, when only 13% of children in this living arrangement). This is further evidence that it is something about step-parent families, not about genetic relationship that underlies the increased infanticide rate.

So what do D&W have to say about the Swedish finding? Well they say:

"Because these estimates include all parental and stepparental killings, many of which have different
typologies and different risk factors than fatal batterings, they are not strictly comparable to the
Canadian, British and Australian numbers discussed above, but they certainly suggest that the magnitude
of Cinderella effects may vary considerably across countries (see also Temrin et al. 2004).
it may well be the case that the modern Swedish welfare
state provides a social climate in which stepparents do not experience, and thus do not resent, heavy
pseudoparental obligation"

i.e. the classic EP get-out, using vague social factors to explain away their allegedly cross-cultural evolutionary imperative.

LemmusLemmus said...

Holy Moly, someone's put in some research. Will respond later when I've done some reading and thinking.

Paul said...

Don't know if you've come across this? It's the podcast of the Maudsley debate on antidepressant effectiveness.

pj said...

Cheers, a colleague told me about it but for some reason I couldn't track it down.

Listening now.