A new study by Colen and Ramey has been getting a lot of buzz: they argue that the long-term effects of breastfeeding have been exaggerated. In their study they looked at discordant siblings, pairs in which one child was breastfed and another was not. They found that the effects of breastfeeding were attenuated and that statistical significance evaporated.
As I've been writing this post in my head, it's been getting longer and longer. I am going to compel myself to be brief, and I also need to be clear that I don't yet have the full text of the paper. I'll report back when I get it. A few considerations:
1. Breastfeeding is not important because it is The Magic Elixir That Prevents All Ills. Breastfeeding is important because it has modest but statistically robust effects across a surprisingly large number of domains. Don't go looking for big effect sizes in long-term breastfeeding studies.
2. Especially don't go looking for big effect sizes in long-term breastfeeding studies of siblings. Of course you're going to see attenuated effects in discordant sibling pairs. Many traits are predicted by a mix of genetic and environmental factors. Siblings share, on average, 50% of their genes and a variable but important proportion of their environment. Breastfeeding doesn't override those predictors with its Magic Elixir-ness; it influences them to a modest but meaningful degree in population-based samples.
Two parents who each have a BMI of 25 are less likely to produce offspring with BMIs of 17 than two parents who each have a BMI at the low end of the normal range. Even if one of their kids is breastfed for 5 years, the mix of other genetic and environmental predictors is really important. It doesn't mean that breastfeeding can't influence BMI across populations just because you don't see big differences in their particular children.
3. "Statistical significance" is a moving target. It is not unusual at all to see attenuated effects and loss of statistical significance when you substantially reduce your sample size, as they did for the portion of the study that looked explicitly at siblings vs. the portion of the study that didn't. Something I can't tell from the abstract: how did they correct for multiple comparisons? When you look at a bunch of different outcomes in a single study, you have to account for the fact that you're running a lot of statistical tests. You don't want to say, "Hey, look! This is significant!" when it's really just random noise in your data. If, hypothetically, I wanted to publish a null finding on breastfeeding, I would look at several outcomes known to have modest effect sizes and use a Bonferroni correction for multiple comparisons. This would mean that instead of a p < .05, I'd need a much smaller (and thus less likely) p value to report statistically significant findings. Highly unlikely that a reviewer would complain; highly unlikely that I'd find statistically meaningful results (Again, I'll report back about what they actually did.)
4. If you must look at breastfeeding outcomes in discordant sibling pairs, then please don't look at them in the 1979 NLSY cohort. Remember when Der et al. looked at IQ in discordant sibling pairs and were roundly criticized for their methodology? Like Colen and Ramey, they used -- wait for it! -- data from the 1979 NLSY cohort. Again, I can't say for sure what's going on without the full text. But it seems implausible that the issues raised in this comment by Melissa Bartick have been addressed adequately in the present analysis. A brief snip for those who don't want to click over: "All we can conclude from this study is that breastfeeding for a very short duration, with unspecified doses of breastmilk, in non-low birthweight children, had no effect." (P.S. about that quote: No idea if excluding LBW kids was a Der et al. decision that might be different in the Colen/Ramey study or a 1979 NLSY decision that affects both sets of results.)
To sum up: no one should breastfeed her baby because she thinks breastfeeding will cause him to stand head and shoulders above his peers, impervious to pain, suffering, and B-minuses on his report cards. At the very same time, no one should pay a whole lot of attention to a single study examining outcomes in a cohort where the median breastfeeding duration was 3 months. As I said at the beginning of this post, the long-term effects of breastfeeding are modest. Women need to do what works for their families and not worry that they are costing their child 15 IQ points by not breastfeeding him. But the effects of breastfeeding are also statistically robust, even in studies that address problematic covariates. (Half a standard deviation in verbal IQ! Best available study and it still blows my mind.) We owe it to ourselves, and to our kids, to build a culture in which breastfeeding can work for most families.