by Emily Oster
Benefits for Mom
Free birth control
More weight loss
Better bonding with your baby
Save money
More stress resistant
More sleep
Form better friendships
Lower risk of cancer
Lower risk of osteoporosis
Lower risk of postpartum depression
Benefits for the World
Lower methane production from cows
You will note that one of these benefits is “better friendships.” Really? Don’t get me wrong—it can be lonely and isolating to be a new mom, and meeting other moms is a great idea. That’s what stroller yoga is for. But I’m hard-pressed to figure out which of my friendships were enhanced by my attempts to feed a screaming baby in a hot closet.
And it is true that I can find no peer-reviewed evidence—reliable or otherwise—to suggest that friendships are enhanced by breastfeeding. Many of the benefits cited here do, however, have some basis in evidence, just not always especially good evidence.
In particular, as I mentioned in the introduction, most studies of breastfeeding are biased by the fact that women who breastfeed are typically different from those who do not. In the US, and most developed countries, more educated and richer women are more likely to nurse their babies.
This wasn’t always the case. Breastfeeding has come in and out of fashion over the years, including over the past century. In the early part of the twentieth century, nearly all women breastfed, if they were physically able to, but the introduction of more “modern” formula starting around the 1930s led to a rapid decline in breastfeeding. This is likely, at least in part, because breastfeeding has always been hard. By the 1970s, the majority of women fed their babies with formula. But public health campaigns beginning at that time promoted the benefits of breastfeeding, pushing back against the trend of using formula. In response to this changed climate, formula manufacturers themselves did some breastfeeding promotion. Breastfeeding rates have increased since then. This increase has been greater in some groups than others, notably among more educated and richer women.2
The relationship between breastfeeding and education, income, and other variables is a problem for research. Having more education and more resources is linked to better outcomes for infants and children, even independent of breastfeeding. This makes it very difficult to infer the causal effect of breastfeeding. Sure, there is a correlation between nursing and various good outcomes—but that doesn’t mean that for an individual woman, nursing her baby will make the child better off.
To give a concrete example, take one study, conducted in the late 1980s, of 345 Scandinavian children that compared IQ scores at age five for children who were breastfed for less than three months versus more than six months.3 The authors found that the children who nursed longer had higher IQ scores—about a seven-point difference. But the mothers who breastfed longer were also richer, had more education, and had higher IQ scores themselves. Once the authors adjusted for even a few of these variables, the effects of nursing were much, much smaller.
The authors of this and other studies claim that once they adjust for the differences they see across women, the effects persist. But this assumes that the adjustments they make are able to remove all the differences across women, and this is extremely unlikely.
For example, in most studies of breastfeeding, researchers do not have access to the mother’s IQ. More commonly, they’ll see a measure of the mother’s education, which is related to IQ. On average, a woman with a college degree will perform better on an IQ test than a woman with less than a high school degree. But these education categories are not a fully accurate measure of IQ.
When we look at breastfeeding, we find that mothers with higher IQ scores are more likely to nurse their babies, even within groups of mothers of the same education level.4 Those mothers with higher IQs, again among peers of the same education level, also have (on average) children with higher IQs.5 Even if researchers are able to adjust for a mother’s education, they are still left with a situation in which breastfeeding behavior is associated with other characteristics (in this example, maternal IQ) that may drive infant and child outcomes.
How do we get around this issue? Some studies are better than others, and we should look to those for answers. When I looked at the data for the effects of breastfeeding, I tried to tease out the good studies from the less-good ones, and I’ve based my conclusions only on the better studies. To link most obviously to the example above, a study that is able to adjust for maternal IQ is going to give more believable results than one that isn’t.
As you know by now, this book is focused on evidence in the form of data and what we can learn from that data. But there is another type of evidence, one that you see a lot on the internet. I’d refer to this as “things people said” or “it happened once to my friend” evidence. You know: “My friend didn’t breastfeed, and her kid went to Harvard.” “My friend didn’t vaccinate, and her kid is super healthy!”
Here is what we learn from this: nothing.
Heed the statistics mantra: anecdote is not data. (I might put that on a T-shirt.)
Now, as breastfeeding will take us more deeply into questions of data, a word on the types of studies I’ll use throughout the book.
AN ASIDE ON RESEARCH METHODS
When researchers study breastfeeding—or any of the other things I talk about in this book—they are looking to learn about the effect of whatever they are studying while holding everything else constant. Our “ideal” experimental setup would be to see a child first after being breastfed, then the same child after not being breastfed, but with everything else exactly the same—same timeline, same parents, same parenting style, same home environment. If we could see that, we would just need to compare the child’s later outcomes to know the effects of breastfeeding.
Of course, this is not possible. But when researchers conduct an analysis, this is what they are aiming for. How close they come depends a lot on how good their research methods are.
Randomized Controlled Trial
The “gold standard” for research methods is the randomized controlled trial. To run this kind of study, you recruit some people (ideally a lot of them) and then choose randomly which people will be “treated” as part of your study and which will be the “controls.” For a randomized trial of breastfeeding, you’d want to have the “treatment group” breastfeed, and the “control group” not. Since you have chosen randomly who will be in which group, the groups are, on average, the same, other than the breastfeeding. You can then compare what happens for the breastfeeding group with what happens for the control.
A practical challenge with this type of study is that you typically cannot force people to do things, especially with their children. Instead, most studies I’ll report on use an “encouragement design”: One group is encouraged to do the behavior—breastfeed, or sleep train their child, or engage in some discipline program—and the other group is not. This encouragement could, for example, take the form of telling the group about the benefits of that behavior, or giving them some training or guidance about how to accomplish the behavior successfully. Assuming that the encouragement changes how many people do the thing you are studying, you can draw causal conclusions.
Randomized trials are expensive to run, especially if they are big, and they can, of course, have problems with implementation. But they are the closest we’re able to come to our ideal treat-the-same-kid-in-two-ways setup, so when I find them, I give them a lot of weight.
Observational Studies
A second, very large group of studies will fall under the “observational study” category. These studies compar
e, for example, children who are breastfed with those who are not, or those who are sleep trained with those who are not, without having randomly assigned people to groups.
The basic structure of these studies is similar. Researchers access (or collect) some data on children, either short- or long-term outcomes, along with some information on parental behaviors. They then analyze the differences between kids in different groups—comparing, say, the kids who are breastfed with the kids who are not.
This type of study will make up the vast majority of the data we have to work with, and they vary widely in quality. One source of variation is study size—some of these are bigger than others, and bigger is typically better. But more important, there will be a lot of variation in how close they can get to the ideal of comparing the same child across one variable in two otherwise identical scenarios.
When they do their comparisons, researchers have to adjust for inherent differences across families that make different parenting choices. Most studies do this by adjusting for some aspects of the parents, or of the child, but their ability to do this well depends on the quality of the data.
On one end, you have sibling studies, which compare two children within the same family who were treated differently on the variable you care about. For example, one of the kids was breastfed, and one was not. Since these children have the same parents and grew up together, there is a strong argument that, other than the breastfeeding, they are similar. These sibling studies are not perfect—you have to ask, why nurse one kid and not the other?—but they have a lot of value in eliminating some of the most important problems in observational studies. There is likely some randomness in the choice to nurse, perhaps related to how much each baby takes to it (I’m thinking of my own experience here).
Many other studies do not compare siblings, but they do see a lot of information about parents: education, maybe IQ tests, income, race, other aspects of the home environment, characteristics of the child at birth, etc. Once the authors adjust for these variables, they can get closer to comparing two identical children. I’ll often call these variables controls. The more things we control for—meaning, the more variables we can hold constant across children and families—the more confident we can be that we are really learning the effects of breastfeeding.
On the other end there are studies that have just one or two controls—that, say, adjust for differences in birth weight across children, but nothing else. These are more suspect.
Case-Control Studies
There is a final class of results that come from what are called case-control studies. These studies tend to be used when there is a rare outcome. Let’s say you want to look at the relationship between reading to your child and your child learning to read very early (say, before the age of three). Learning to read before three is a very rare outcome. Even in a very large dataset, you might have only a few cases. This isn’t enough data to learn about what determines this outcome.
With a case-control approach, researchers start by identifying a set of “cases”—people who had the rare outcome. In our example, that means they go out and actually look for children who could read fluently before age three, and they collect a bunch of data about them. They then look for a set of controls—children who are similar on some dimensions but didn’t read until later—and compare them. They ask whether some behavior—in this example, parents reading to the kids—are more common in the children who were early readers.
In general, these studies are worse than the other types. They have, first off, all the same problems as observational studies: the people who are in the case group may be different in many ways from those in the control group, and it is hard to control for those differences. This problem is often more extreme since the control group is typically recruited to the study in a different way from the treatment group.
There are other problems, too. These studies usually rely on asking parents about aspects of their behavior far in the past—parents may struggle to remember, and their memories may be affected by what has happened with their child in the intervening years.
Finally, these studies tend to be small, and the authors are often looking at many possible variables that might be associated with what they are studying. This can lead to spurious conclusions.
There will be times when these are the only studies we have to go on, and we do want to try to learn something from the data they contain. But I tend to approach these with caution.
BACK TO BREASTFEEDING
In the particular case of breastfeeding, we’ll see all the kinds of studies described above. There is one large randomized controlled trial of breastfeeding, which was run in Belarus in the 1990s.6 This study encouraged some women to breastfeed and not others, and there were differences across groups in breastfeeding rates. This study will be relevant for looking at some short-term health outcomes, and some longer-term things like child height and IQ.
There are also some very nice observational studies. There are a few that compare siblings, which is great, and others that were not able to use siblings but do have a large sample size and observe a lot of data about kids and their parents.
Finally, for a few rare and tragic outcomes—childhood cancer, SIDS—we will have to look at some case-control studies, and try to learn what we can from them.
In the rest of this chapter, I’ll go through the short- and long-term benefits of breastfeeding to kids and to moms in detail. I will leave aside the issue of methane and say only that it is true that cows produce methane, and it is also true that formula usually contains milk products, so in that sense this benefit is valid.
Oh, and I should say that even if you’ve decided to breastfeed, making it work is not always easy. To tackle that (stay out of hot closets!), check out the next chapter.
The Benefits
BREASTFEEDING AND EARLY-LIFE HEALTH
Breastfeeding and early-life health is the most well-studied set of relationships. It was the initial focus of the large randomized trial I mentioned earlier, and these are also the relationships with the most compelling set of mechanisms. We know breast milk contains antibodies, so it is therefore more plausible that it is protective against some illnesses.
We’ll start with the randomized trial. This study, called PROBIT, was run in Belarus in the 1990s. It followed 17,000 mother-infant pairs across a number of sites in Belarus. The authors started with a sample of women who intended to breastfeed; half of these women were randomly chosen to receive breastfeeding assistance and encouragement. The rest were not discouraged, but they were not provided with support.
The encouragement had a big effect on breastfeeding. At three months, 43 percent of children of moms who were encouraged were exclusively breastfed, versus just 6 percent of children whose mothers were not. There were also differences in whether the babies got any breast milk at this point. At a year, the any-breastfeeding rates were 20 percent and 11 percent, suggesting that the effects of the encouragement persisted.7
You’ll notice that the encouragement didn’t mean all the moms who were encouraged to breastfeed did, or that all the moms who were not encouraged didn’t. The results, then, may be smaller than they would be if there were a larger difference in breastfeeding between the two groups.8
The study found two significant impacts: In the first year, breastfed babies had fewer gastrointestinal infections (i.e., diarrhea) and lower rates of eczema and other rashes. To put some numbers to it, 13 percent of the children of mothers in the group that wasn’t encouraged to breastfeed had at least one diarrhea episode, versus only 9 percent of those whose mothers were encouraged. The rate of rashes and eczema was also lower in the group whose mothers were encouraged to breastfeed: 3 percent versus 6 percent.
These effects are significant, and as a share of the overall rates of these illnesses, they are reasonably big. For example, rashes and eczema were reduced by half. Having said that, the overall rates are
worth keeping in perspective: even in the group that breastfed less, only 6 percent of children were reported to have this complication. It is also important to note that these are typically fairly minor illnesses.
There is one very serious early-life illness—also linked to digestion—that seems to be affected by breast milk. Necrotizing enterocolitis (NEC) is a serious intestinal complication that is a risk for very preterm babies (it is most common for babies weighing less than three and a half pounds at birth). Breast milk (from either the mother or a donor) has been shown to lower the risk of this condition in randomized trials.9 This may bolster our confidence in the general links with digestion, although for full-term (or even nearly full-term) babies, NEC is vanishingly rare.
In the PROBIT trial, there were also many illness measures that didn’t seem to be affected by breastfeeding, including respiratory infections, ear infections, croup, and wheezing. Indeed, the share of kids in each group who had these problems was virtually identical. It is important to be clear on what this means. It does not mean we are sure breastfeeding has absolutely no effect on respiratory problems. These estimates come with statistical errors, what we call “confidence intervals,” which give us a sense of how sure we are about the estimate we observe. In this particular study, we cannot reject the possibility that breastfeeding could matter in either direction—that it could decrease or increase respiratory infections.
What we can say is that the data doesn’t support the claim of a reduction in respiratory infections as a result of breastfeeding.
