Tuesday, September 22, 2009

Questioning assumptions

The hotel at which I am staying for a 1 1/2 day NIH meeting provided either USA Today or the Wall Street Journal as a complementary paper. I chose the WSJ. Probably no surprise there. So, today I had the opportunity to see not only the news feed from the Johns Hopkins Bloomberg School of Public Health, but all the articles in the paper. One caught my attention (as I am a "winter baby"): "New Light on the Plight of Winter Babies" by Justin Lahart.

What is the issue here? Past studies have consistently found that, on average, babies born in January have lower educational attainment and earnings than babies born in other months. This puzzled economists for a while. Almost two decades ago, a pair of economists put forward the hypothesis that because of when it is legal for kids to drop out of school, being born early in the year made it possible to drop out at an earlier grade. As a result, the average January baby would earn less, although many January babies do not drop out. Other economists have suggested other reasons, but the best known is the early drop out presumption.

The reason that this blog entry is titled "questioning assumptions" is that the assumption throughout is that the babies born in different months are otherwise similar. In other words, birth month is relatively random and all mothers are just as likely to have a baby in any month. A recent working paper has suggested that the mothers of babies born in different months are quite different. Each year, mothers of children born in January were less likely to be married, more likely to be teenagers, and had a lower average level of education. In contrast, mothers of babies born in May had the opposite characteristics. What this suggests is that the babies (or at least the families of the babies who may have a large role in determining educational attainment) are not all the same. As a result, we can't say whether the effect of month is actually a time effect or is an effect of family background.

I imagine that economists will debate for some time how important this alternative explanation is, but the fact is that a basic assumption driving the old analysis is now questioned.  Economists will also have to explain why there are differences in mothers associated with the month of their child's birth.

How could this be applied to analyses in health care? There are two examples I thin of immediately:

(1) Not all babies who are breastfed have mothers who are the same. Thus, any findings about the effects of breastfeeding are shaped by the differences in the babies' families.

(2) Not all infants are expected to be as healthy by their mothers when they are fetuses. The expected health of the baby may affect the mother's choice of when to initiate prenatal care and how much to get.

The key is that when we are not using a randomized study design, we have to carefully consider all our assumptions. Sometimes we have no way of assessing the validity of our assumptions. However, when we are given an opportunity to do so, we should take it and reassess conclusions as necessary. That way we can make more informed (and hopefully better) policy.

No comments:

Post a Comment