No, that’s not polygyny… comments on Ross et al. (2018)

Royal Society recently put out a massive paper by Ross et al. with over 9000 authors (you know, one of those ones) on polygyny and wealth inequality. The title, “Greater wealth inequality, less polygyny: rethinking the polygyny threshold model” would have you thinking that the authors were able to refute or quantitatively disprove the polygyny threshold model with some sophisticated mathematics, but unfortunately this is not the case. Instead, the paper uses a strange mixed sample of hunter gatherer and highly developed industrial populations to argue that the transition to agriculture increases socioeconomic inequality, and additionally results in conditions of subsistence living that for most make polygyny effectively impossible.

Don’t you love it when the author and affiliation list is so big you can’t even screencap it? Maybe it’s deliberate!

Firstly, we should realize that this doesn’t amount to either a refutation or even the titular ‘rethinking’ of the polygyny threshold model. While results from their quant analysis are basically legit, it doesn’t change the fact that the authors have effectively based their study on a tautological proposition; subsistence living results in no surplus wealth (also tautological) which means that it is exceedingly rare for polygyny to be mutually beneficial. Alright. So where’s the challenge to the polygyny threshold model?

I have read a lot about polygyny, but I have never encountered any claim that polygyny ipso facto increases linearly with socioeconomic inequality per se. Rather, claims are made that conditions of high socioeconomic inequality will guarantee polygyny, as male reproductive success is subject to greater resource-dependent elasticity than female fitness due to inherent biological features (i.e. 9 months of pregnancy). This great presentation has more details, but for those with little time:

I had to screenshot this in word since I don’t have LaTeX on my WordPress acc ;_;

Or if you prefer (from the presentation linked above; this contains an error, as the 1948 paper cited is by A.J. Bateman, not Bateson):

To be fair, the authors recognize this by stating their intention to merely “extend” the polygyny threshold model, but I’d argue they haven’t done so in a way that’s significant enough to merit the “rethinking” boast. But this is not to suggest the paper has no value. Instead, what the authors have actually done is modeled the conditions for polygyny to take place in a largely monogamous society at subsistence-level conditions – unironically a notable achievement. This is a far more interesting result, and one that would merit wider recognition than the paper has currently received.

There are still some problems though. For instance, the paper notes:

“Sequential marriage can be considered a form of polygyny insofar as men typically replace divorced wives with younger women, allowing a subset of males in the population to increase their lifetime reproductive success relative to less wealthy males in the population, as has been shown in many of the populations sampled…”

Now this actually is a problem, since the definition of polygyny that the authors are using is not actually “polygyny” but “effective polygyny” so defined. I hate it when researchers redefine constructs in this ad-hoc fashion (especially when it’s not highlighted in the abstract) because it can mislead people who don’t read the full paper, and most of the postdocs I know don’t. Luckily, I did.

I think the problem with including sequential marriage into a working definition for polygyny is that there are substantial qualitative differences that distinguish these behaviors. For instance, technical polygyny (one man, multiple women at the same time, in a sexually exclusive [typically marital] arrangement) actually alters the operational sex ratio, among other things. Sequential marriage, by contrast, only means that the available pool of females includes larger numbers of women who already have children – that is, single mothers. Of course this may change the calculus for male satisfaction or some other outcomes, but these are not equivalent to the social effects we expect from a normatively polygynous mating equilibrium. For example, they completely negate some of the reported correlates for polygynous mating, such as female suffering and self-reported detriment to well-being noted amongst women in polygynous marriages (source from observational study in Cameroon). I understand that well-being wasn’t strictly a feature of relevance to Ross et al.’s analysis, but it DOES have implications for the precise ‘leveling’ of the polygyny threshold. A situation where a woman is going to be a second or third-order co-wife is very different to one in which she’s merely a second or third-order sequential wife. These differences matter quite a lot if we’re paying any attention to the implications for (1) the OSR (2) female well-being and gender inequality (3) male violence and intrasexual competition, among many other things.

Sourcing locations for data used in the paper. Does this look representative to you?

Now look, I’m not trying to be an ass and negate all the hard work the forty-thousand authors of this study did. But I do find it somewhat annoying when people publish work under “GOTCHA!” titles like “rethinking XYZ” despite nothing comparable to this having actually taken place. Far from rethinking, the authors actually RELIED ON the polygyny threshold model for their analysis, and came to the result that the agricultural transition killed incentive for polygyny amongst most normal people living in subsistence-level conditions. Fair enough. But why not just say so?

IMHO, the far more interesting result we get from the paper is this: we know that transition to monogamy occurred around the transition to agriculture in some societies, and this paper provides some really awesome and useful analysis to explain why that might have happened. But what it DOESN’T do is explain why monogamy actually became a social institution to the exclusion of plural marriage. Just because it isn’t worth it to have 2+ wives doesn’t mean that your society will necessarily ban having 2+ wives. We still don’t have an answer for why polygyny becomes legally and socially prohibited in these agricultural societies. However, I think that primate inequality aversion (as exhibited by this outraged capuchin monkey) might be a good place to start.

I don’t have the data to hand, but I do have a hypothesis. Agricultural transition makes polygyny functionally impossible for overwhelming majority of people, who are living at subsistence. But it DOES NOT affect the ability of men with sufficient social standing and resources to obtain and retain multiple wives. Historically, such men were stratified into classes or castes – merchants or Japanese 商 etc. It seems plausible to suggest that the impoverished majority of monogamous males (and perhaps their wives!) would have expressed strong opposition to their rich rulers taking multiple wives, and rallied to condemn this behavior. Others have articulated this hypothesis before (e.g. Henrich, Boyd, and Richerdson, 2012) but this study provides some useful background evidence for its plausibility. If you’re a man farming away in a Neolithic village under fairly awful living conditions, you might be able to tolerate paying taxes to your overlord despite his nice villa on top of the hill. But what if he has 6 wives and your daughter is one of them? Perhaps there might be an ‘outrage threshold’ we need to think about alongside the polygyny threshold model.

God, why does Gurlockk get to have the biggest rocks, the shiniest gems, and 12 wives when I can’t even count past ten?

The meta-analysis that wasn’t: assessing Flynn Effects through diachronic change in ICV

A while ago I was conducting a meta-analysis on diachronic variation in cranial volume measurements for different East Asian populations. I got into the project after being inspired by Lynn’s suggestion that the Flynn Effect is primarily nutritional, as presumably this would also show an effect on height, head size, and thus ICV. It could even be possible to control for ICV changes to show only the direct change to IQ over time, which would be a more rigorous way to compare Flynn Effect magnitudes. What a great idea, I thought.

Unfortunately as I got deeper into the project, I realized that virtually all of my data couldn’t be sourced back to any obtainable research. A lot of numbers were sourced from papers that were only available in physical archives in Japan or Korea, so I did what any good researcher does and promptly gave up.

However, even with the limit to my data integrity acknowledged, I was still able to get some plots of the cranial volume measurements that show some interesting results. Here’s a scatterplot I built using values from all studies I could find; the x-axis shows the year of the study’s publication and not the date of subject collection, expiry, or measurement. Note that all datapoints are n-unweighted (raw averages).

While initially there doesn’t seem to be much to say about this graph, it’s noteworthy that there don’t appear to be (when plotted in this form) any outliers, but two relatively clear clusters by gender. Clearly the female cluster is far less grouped than the male one (the source of its statistical insignificance that can be clearly seen in the following trendlines) but both clusters have mutually exclusive ranges, which are interesting. Since all of the studies here had sample sizes that were above 20 we’d expect distributions to be roughly normal, meaning that group-level sex differences in ICV are likely very reliable. This of course is no surprise seeing as ICV is effectively a function of body size and height, which are also variant with respect to sex in the same way. This becomes more obvious when we ignore the insignificant ethnic variation and go purely to the sex differences:

Overall it is fairly clear that ICV as reported in studies is increasing overtime. However, note the lack of any female reports going back beyond the early 1920s? This is a sign that our data is shitty. For many of the earlier studies no sex information was present, and although in some cases I was able to make informed estimates based on the averagees (i.e. a mean ICV value close to 1500 is almost certainly male or mostly male in content) it was never really certain how accurate or representative the values might be. However, if we are correct in inferring that earlier studies likely did not disambiguate male and female samples (as opposed to exclusively using male samples) then it could be suggested that the actual observed gap would be even greater, seeing as the female values would go down and the male ones would go up. This of course would throw the trendline for diachronic variation into doubt, which is currently strong with p value < 0.05 when excluding females.

 

Email me if you want the full dataset to work on.