Wednesday, May 27, 2009

A New Dimension to the Debate--And We Debunk It Too!

Just as more and more people are questioning the wisdom of the 21 drinking age, especially its purported lifesaving effects, a new dimension to the debate was added by a recently published study. Watson and Fertig (2009) did a study on the effects of the drinking age on three measures of infant health outcomes: premature birth, birth weight, and congenital anomalies (birth defects). All data was from 1978-1988. The results were hailed as proof that a lower drinking age leads to worse infant health outcomes, presumably through unwanted pregnancies that result from drunk sex. But after reading the study, there are several apparent flaws and caveats that the media glossed over or misrepresented:

  • Only the first two measures of health outcomes (prematurity and birth weight) were statistically significant, while congenital anomalies were insignificant and even had the "wrong" sign. The latter is difficult to explain away, since alcohol is a known teratogen.
  • Effect sizes were generally very small (10% or less). Such effects are very difficult to interpret in epidemiological studies, and causality virtually impossible to determine.
  • R-squared values were very low despite controlling for numerous variables, indicating that the regressions had a bad fit. This can lead to spurious effects.
  • Missing data can lead to noisy data.
  • Curiously, the three worst years for teen pregnancy since the 1950s (1989, 1990, 1991) were not included in the data for some reason.  All of which occurred after all 50 states raised the drinking age.
  • For survey data concerning drinking before or during pregnancy, underreporting (due to changes in alcohol laws or social acceptability) could be a confounding factor.
  • While much ado has been made of the fact that the apparent effects were stronger for black women, the authors do not seem to offer an adequate explanation for why. Although it could be that white women are more likely to abort unwanted pregnancies, or even simply have better access to birth control than black women. Another possibility is differential access to good prenatal care for those who carry to term.
  • There was no effect for immigrant women.
  • Controlling for beer taxes reduced the effects by half for the 14-17 year old black women, and made the effect on 18-20 year olds less significant as well. This implies that beer taxes may have an effect. But this variable was excluded from the authors' preferred specification.
  • When 14-24 year olds as a whole were looked at, the drinking age effects interestingly had the opposite sign than when women under 21 were examined separately. That means that for 21-24 year olds, a lower drinking age was associated with better health outcomes that outweighed the purported adverse effects in 14-20 year olds, especially for black women. We speculate that, like with some studies of traffic fatalities, the pregnancies (and their poorer outcomes) were simply shifted a few years into the future. Or it could simply be that the regressions had a bad fit, putting a giant question mark over the entire study. Either way, the net effect for 14-24 year olds is the opposite of what was predicted. These possiblities were completely ignored by both the authors and the media.
  • Effects were stronger for women under 18 (the spillover group) than for women aged 18-20 (the target group). While some spillover effect is plausible, alcohol was just as illegal for women under 18 either way, so the effect should still be the strongest for 18-20 year olds. Otherwise, it is very likely a proxy for something else.
  • While the apparent effects were robust to state-specific time trends, they were not robust to age-specific time trends. In fact, the "race effect" was reversed in the case of birth weight. Even the authors concede the possibility that secular trends may affect the youngest women first, and let the reader decide.
  • As far as race goes, one potential confounder is "liquorlining," or the practice of concentrating alcohol outlet density (chiefly liquor stores) in poor minority communities. Some states probably do this more than others. This was not addressed.
  • The authors found that controlling for paternal information weakens the relationship between drinking age and infant health outcomes. And missing paternal age (thought to be a proxy for unwanted pregnancy) was significantly correlated with a lower drinking age for black women under 21 (but not for white women), leading to the assumption that drunk sex was the causal factor behind it all. But the effect was statistically insignificant in states that did not have parental notification laws for abortion. That counterfactual is telling, especially since that was true for 18-20 year olds was well (notification laws only affect those under 18). Why the latter was the case is not explained.
  • And again, for 14-24 year olds as a whole the sign was "wrong," and both "wrong" and significant for black women in notification states. This supports the idea of merely shifting unwanted pregnancies, or just a bad fit of data, but again that possibility went unaddressed.
In other words, the study makes at most a very weak case for keeping the drinking age at 21. No scratch that, it is downright anemic. Effects could not be disentangled from age-specific secular trends, so the results are inconclusive at best. Even if the purported effects were real, which is still not very clear, the effect sizes are of little practical significance since so many other things affect both teen pregnancy and infant health outcomes much more strongly. And for 14-24 year olds as a whole, a drinking age of 18 actually appears to be beneficial in terms of net effects when the data are taken at face value.

The greatest irony of all is that America tends to have some of the worst infant health outcomes relative to other industrialized countries, especially infant mortality, yet our drinking age is the highest. This apparent paradox is mainly due to less prenatal care, among other things. In addition, our teen pregnancy rate is also the highest, and other countries had generally seen more progress in reducing its rate than we did in the years following the drinking age hike. In other words, those who put too much stock into this study clearly fail to see the forest for the trees.

Furthermore, it does not follow that the results are still relevant today since only 1978-1988 were examined. First of all, the effects may just be short-term. Secondly, a lot has changed since 1988. Although they rose in the late 1980s (after raising the drinking age, oddly enough), teen pregnancies (and abortions) have plummeted since 1991, reaching lows that had not been seen for decades. Birth control (including condoms) is more available now and used more frequently than before. And as hard as it is to believe, teenagers are also less likely to have sex now than in 1990. In other words, today's younger generation is significantly more responsible (at least in terms of sex and reproduction) than the Boomers and Gen-X were when they were teenagers. But you would never hear the media say that.

This is pure mission-creep. It's time to quit punishing today's young people for the sins of their parents.

UPDATE:  It appears that in 2005, for unknown reasons, birth weights from healthy, white mothers (of all ages) were 3 ounces lower than in 1990, and the length of gestation has shortened slightly.  This is rather surprising since for the past 50-70 years, birth weights have been rising.  And trend began reversing in 1990, which was a few years after raising the drinking age.  If you recall, the Watson and Feritg study only included 1978-1988, a relatively short timeframe during which both birth weights and drinking ages rose.  This latest finding casts further doubt on a study that is already on shaky ground.

No comments:

Post a Comment