Skip to content
 

“Breakfast skipping, extreme commutes, and the sex composition at birth”

cat_sniffing_breakfast_cereal3

Bhash Mazumder sends along a paper (coauthored with Zachary Seeskin) which begins:

A growing body of literature has shown that environmental exposures in the period around conception can affect the sex ratio at birth through selective attrition that favors the survival of female conceptuses. Glucose availability is considered a key indicator of the fetal environment, and its absence as a result of meal skipping may inhibit male survival. We hypothesize that breakfast skipping during pregnancy may lead to a reduction in the fraction of male births. Using time use data from the United States we show that women with commute times of 90 minutes or longer are 20 percentage points more likely to skip breakfast. Using U.S. census data we show that women with commute times of 90 minutes or longer are 1.2 percentage points less likely to have a male child under the age of 2. Under some assumptions, this implies that routinely skipping breakfast around the time of conception leads to a 6 percentage point reduction in the probability of a male child.

Here are the key graphs. First, showing that people with long commute times are more likely to be skipping breakfast:

Screen Shot 2016-04-03 at 2.39.10 PM

I have no idea how 110% of people are supposed to be skipping breakfast, but whatever.

And, second, showing that people with long commute times are less likely to have boy babies:

Screen Shot 2016-04-03 at 2.41.05 PM

I have no idea what’s going on with these bars that start at 49.8%, but whatever. Maybe someone can tell these people that it’s ok to plot points, you don’t need big gray bars attached?

Anyway, what can I say . . . I don’t buy it. This second graph, in particular: everything looks too noisy to be useful.

Or, to put it another way: The general hypothesis seems reasonable, when the fetus gets less nourishment, it’s more likely the boy fetus doesn’t survive. But this all looks really really noisy. Also, the statistical significance filter. So the estimates they report, are overestimates.

To put it another way: Get a new data set, and I don’t expect to see the pattern repeat.

That said, there are papers in this literature that are a lot worse. For example, Mazumder and Seeskin cite a Mathews, Johnson, and Neil paper on correlation between maternal diet and sex ratio that had a sample size of only 740, which makes it absolutely useless for learning anything at all, given actual effect sizes on sex ratios. They could’ve just as well been publishing random numbers. But that was 2008, back before people know about these problems. We can only hope that the editors of “Proceedings of the Royal Society B: Biological Sciences” know better today.

23 Comments

  1. Tor says:

    In the second graph, it’s really only the 90+ minutes group that makes any correlation. At least to my eye, the other points look uncorrelated.

    I am also suspicious of using a weak measure, getting a weak effect size, then assuming the effect is actually strong.

    • Andrew says:

      Tor:

      Yes, it’s just so tempting for people to study sex ratio, but there’s so little signal to work with and so much noise, it’s really really tough to find anything and really really easy for researchers to fool themselves.

    • elin says:

      That’s what jumped out at me, the magic number 90. Of course that is a truncated distribution who knows about what happens with 90 to 94 versus 120 to 134! But aren’t those 15 minute bands either a bit narrow (15 minutes late is something I plan for on my commute) or a bit wide (since those groupings are artificial)?

  2. Dale Lehman says:

    Also, it seems that they only “document” that longer commute times correspond to missing more breakfasts. What about longer commute times for pregnant women – do they miss breakfast more? And, do they continue to have these long commute times while pregnant? And, how long do they continue to commute for such long times while pregnant? Of course, the sex should be determined at conception, not during pregnancy, but how many of the pregnancies result in births? Also, the measure they use is the fraction women with long commutes with male children under the age of 2 – it does not say that they had these long commutes while pregnant. Enough noise that even my aging hearing is swamped.

    • Carlos Ungil says:

      > And, do they continue to have these long commute times while pregnant? And, how long do they continue to commute for such long times while pregnant?

      Whether they continue to have these long commute times after learning about their pregnancy, and for how long, is irrelevant. But you know that already, because in the very next sentence you say that

      > Of course, the sex should be determined at conception, not during pregnancy

      I think this detail is not lost on the authors of the study either, because in the quoted fragment they talk about “environmental exposures in the period around conception”.

      • Dale Lehman says:

        But it might influence the miscarriage rate. And we don’t know how large the sample of 90+ minute commutes is. And, since the births are measured by “male child under the age of 2,” we don’t know if these women had these commute times while they were pregnant or after their babies were born.

        • Carlos Ungil says:

          I’m sorry, I think I misread your comment. I thought you were mixing some legitimate concerns and some other not so legitimate ones, but now I see how all your comments fit together.

  3. Slugger says:

    Obviously, to maximize male offspring one should arrange to commute 60-74 minutes. This information will be very useful in cultures that value sons.

  4. Another problem here is that we don’t know the actual distribution of mothers’ commute times. The breakdown (by 15-minute intervals) could distort things. The sample sizes for some of these intervals could be much larger than others. My guess is that the 90+ group is fairly small.

  5. Llewelyn Richards-Ward says:

    It’s a general finding that intra-day glucose variability is high and that, during pregnancy, this natural variation becomes higher. I infer that this is the mechanism that the article alludes to (as this is what skipping breakfast does). So why not make these links clearer by actually measuring the actual variable of interest (along with the more likely endocrinological and prenatal environmental factors that might vary gender)? And how exactly is gender ‘selected’?Statistically, the high intra-day variability and increased changes during pregnancy would contribute a lot of noise to any correlations split into a dichotomous outcome frequency. Just hard to see how these studies might show anything useful when I think through the possible mechanisms of action, starting with not eating breakfast, to changes in the body, to gender selection. Surely all this comes before statistical dark arts are applied?!

    • Andrew says:

      Llewelyn:

      Actually, it makes to apply the “statistical dark art” of design analysis sooner rather than later, thus learning early on that the signal is going to be much smaller than the noise, so as to know not to waste time constructing biological explanations of patterns that are essentially random numbers.

      • Llewelyn Richards-Ward says:

        It is so obvious when you say it like this — but why is design not happening? This genuinely puzzles me, even if only as a feature of human behaviour. Surely, seeing the variances in glucose availability through the day would warn a researcher that choosing an indrect measure of breakfast consumption would be nothing more than sampling noise. I guess that this underscores the value of prior information, yet again. :)

        PS: Please forgive the Harry Potter metaphor.

        • Andrew says:

          Llewelyn:

          I think the problem is this: What makes a project a success, in the view of most researchers, is attainment of a “p less than .05” result along with a corresponding story. The story isn’t hard to come by so what’s really precious is the statistically significant result. A typical researcher doesn’t care about Type M and Type S errors because both of those are conditional on statistical significance, and once a researcher finds statistical significance it’s time to declare victory.

  6. Jonathan says:

    Their key line is really this, isn’t it: “Glucose availability is considered a key indicator of the fetal environment, and its absence as a result of meal skipping may inhibit male survival.” There are three huge if’s in that statement: “is considered”, “absence as a result of meal skipping” and “may inhibit”. They are then leveraging from some rather inspecific activity data about drive times, that they then try to relate to meal skipping and then to blah, blah, blah. And that bluntly would only be of value if the correlations were dead on obvious OMG. Consider a sort of pposite case: 2nd hand smoke. They collected data about birth rates and other stuff, related that to the presence of smokers in the house and that made a case that this caused that. And yet that case is arguable, at least with regard to cancer and some other effects. To me, this kind of study fails because they’re taking this chain and presenting it as real: “may inhibit” up to “meal skipping” up to “is considered” using lousy looking data about commutes as the lever to test this entire chain. That’s sloppy thinking. It isn’t to me that the statistical results are minimal or noisy but that any results would be utterly meaningless up to some very high measure of certitude and then I’d be thinking not that they’re right but that something has gone wrong in their data or analysis. Why? Because this isn’t a logical approach to the problem. Example: lots of women skip lunch, so why the bleep is drive time/maybe skipping breakfast meaningful? I can’t see a choice made in this work that isn’t like sticking a hat on a cat and thinking that means cats wear hats.

    • Simon Gates says:

      The thing that bugs me is: could these researchers really not see that they were constructing a ridiculously tenuous argument, essentially just pulling explanations out of the air, like politicians and (some) journalists do? Or were they well aware and just didn’t care, because they got significance and therefore publication? Either way, science loses.

      For what it’s worth, and from my previous exposure to this sort of research, I’d guess that there are several factors at play here – but one of them is the persuasiveness of “significance”. It’s hard to let go of the idea that “something must be going on” when you see that, especially when the effect is unexpected. As Feynman said: the easiest person to fool is yourself.

      • Martha (Smith) says:

        “could these researchers really not see that they were constructing a ridiculously tenuous argument, essentially just pulling explanations out of the air, like politicians and (some) journalists do?”

        My impression is that many people (including many, but not all, social scientists) indeed do see scientific argument as indistinguishable from what “politicians and (some) journalists do”.

  7. Slugger says:

    Now that I am fully caffeinated and not yet stupified from gluttony, I’d like to share some thoughts.
    First, Dr. Gelman, I am thankful for this site which continues to teach me something almost every day.
    It might be a little strange to study maternal impacts on sex ratios since women only produce ‘x’ haplotype gametes. Yet when I look at PubMed and type in “sex ratio maternal” I get 4080 citations but “sex ratio paternal” only gets 773. Many of the citations are chaff, but the discrepancy is real. I bring this up because I have been reading a book by Nadia Hashimi, The Pearl that Broke Its Shell. This novel describes Afghan society where women who birth sons have much more status than mothers of daughters. Basic biology is unknown to Afghan tribal people, of course. If one were to study parental effects on offspring gender ratios, looking at maternal factors might doom one to producing low quality studies.

    • Andrew says:

      Slugger:

      I think one problem here is “folk biology,” that is, the vague notions about biology held by Afghan tribal people, Readers in Management at the LSE, etc etc. To these people, the difference between men and women is so large that it just seems natural that it should seep into all aspects of human biology. Hence the intuition that the sex ratio of babies is under the mother’s control, or affected by beauty, or social status, or whatever. Empirical variation in sex ratio is tiny (for example, the difference in proportion girls, comparing blacks and whites in the U.S, is something like .005, and that’s one of the largest differences out there), but it’s hard for empirical data to beat out intuition—especially when intuition can be supported by “p less than .05” noise.

      • Ben Bolker says:

        It’s not just folk biology and groupthink. Females (especially internally fertilized, live-bearing females as in mammals) have lots of physiological control over sex of offspring, whether pre-fertilization (filtering which sperm get to the egg) or post-fertilization (selectively terminating zygotes of one sex or the other). From a theoretical point of view, females may also have a stronger evolutionary pressure to manipulate sex ratio because they (typically) invest more in individual offspring.

        There are plenty of good biological examples out there (e.g. red deer); if you’re interested in sex ratios more generally, sticking to non-human animals might be one way to stay away from the more sensational/fishy results.

        • Andrew says:

          Ben:

          Yup, it’s a different story with other animals. For whatever reason, humans have an extremely stable sex ratio, so we’re pretty much the worst animal to study in this regard.

          • Ben Bolker says:

            “Humans have an extremely stable sex ratio”: relative to what reference population? It might be the opposite, i.e. biologists who are interested in sex ratio evolution tend to focus on organisms with mutable sex ratios. Humans might not be much worse than a typical (randomly sampled) mammal …

  8. jrkrideau says:

    Is there any mention of commute by type of transportation? As a cycle commuter and occasional bus user I tend to think of this. If there is any effect here I’d like to see the effect of mode of commute

    I (male) tend to arrive at work by bike invigorated–think the equivalent of a good work-out at the gym, or when taking a bus during heavy rain or a snow storm, much more relaxed than my stressed-out motor vehicle driving colleges.

    Those are about the ugliest and hardest to interpret dynamite plots I have seen in year.

Leave a Reply