Regarding the authors’ incomplete transition analysis, we know that while more than half of the children in their sample (total sample 548) increased their consumption of sugar-sweetened beverages (312) during the period of study, almost as many children went from being obese to being non-obese (35) as went from being non-obese to obese (37). Also, more than 65 percent of the children were never obese during the course of the study.
Looking at the body mass index results, the authors’ analysis suggests that while average BMI went up by 1.5 points (rising from 22.23 to 20.73, as shown in their report’s Table 1), even the least conservative model presented in their report’s Table 2 suggests that the average change in sugar-sweetened beverage consumption (average consumption goes from 1.22 servings to 1.44) explains a BMI increase of only 0.05 points (0.22 servings change × 0.24 effect of one-serving increase on BMI). That is, even using the largest estimated effect in Ludwig and Gortmaker’s paper, the effect of the observed increases in average sugar-sweetened beverage consumption explains only about 3 percent of the observed change in BMI in their sample. If beverage consumption is any part of the increasing BMI story, it is not more than a bit part.
It would have been valuable to examine the movement out of obesity too, since almost as many children transitioned out of obesity during their study as became obese.
It is possible that any effects of beverage consumption operate more intensively on some individuals than on others Thus, it is appropriate to analyze obesity transitions in both directions. While the authors analyze transitions into obesity, they report no results concerning transitions in the opposite direction. To provide a complete and accurate picture of what is going on here, we suggested in our article that it would have been valuable to examine the movement out of obesity too, since almost as many children transitioned out of obesity during their study as became obese. Recall that the majority of children increased their sugar-sweetened beverage consumption in this time period. If the conclusions drawn by the authors are robust, we should observe that those transitioning out of obesity were systematically more likely to be in the minority of children who decreased their consumption of sugar-sweetened beverages. The authors did not perform (or at least present) this analysis even though it would have been easy to do so and would have served as valuable evidence one way or the other regarding the relationship between sugar-sweetened beverage consumption and obesity.
With respect to the authors’ choice of what unit change they used to frame the policy implications of their results, a one-standard-deviation change is often a useful benchmark. However, it may be less useful in a distribution such as the one they were analyzing. Specifically, if one-quarter of their sample increased consumption by more than four times the average change and there were not similarly extreme reductions in consumption among other subjects, this suggests a highly right-skewed distribution. In such a skewed distribution, a standard deviation is a less useful unit of reference. With linear models, even very small effects can be made to appear large if authors posit a large enough change in a given explanatory variable. In our view, this is what Ludwig and Gortmaker have done. This was not an ad hominem attack. It is a difference in judgment and style that we stand by.
On a parting note, we did disclose in our article that we have indirectly received a nominal amount of funding from the American Beverage Association. We do not believe this creates a conflict any more than does, say, taking grant funding from the Robert Wood Johnson Foundation, which has a stated commitment to promoting taxes on items such as sugar-sweetened beverages and which funds a large number of studies done in this area. In any event, all of our analysis of the literature in our Regulation article was descriptive and is easily evaluated for accuracy.
