Bleich et al. Respond

We appreciate Sainani’s careful review of our research. She highlights two interesting issues regarding the interpretation of our results but does not uncover limitations in our article not already noted.¹ Because we cannot be responsible for the media’s interpretation of our results, characterizing our article as “misleading” is unfair. She correctly notes, however, that our analysis did not account for the potential correlation of customers within stores.

Sainani criticizes our article for “a needless emphasis on odds ratios” but we present the change in sugar-sweetened beverage rates (Table 2 in original article), odds ratios (ORs; Table 3 in original), and the predicted probability of beverage purchases pre- and postintervention (Figure 1 in original). Because we acknowledge that the use of odds ratios in certain contexts may be undesirable, we deliberately reported the unadjusted frequencies and predicted probabilities to give an undistorted sense of the effect size. She provides an example of a newspaper misinterpreting what an odds ratio actually means, but we certainly cannot be held responsible for how the media presents our results.

Sainani also criticizes our interpretation of the differential effect of the three caloric conditions. Her critique that “the data provide no evidence that the three caloric conditions differ” confused us, because we pointed out that “3 ORs for the different types of information were not significantly different from each other in a formal test.”^(p332) Nevertheless, the three conditions do actually have different statistical relationships relative to the baseline. Again, how the media elected to report the differences by caloric condition is outside of our control.

Sainani’s concern about clustering is valid. Possibly, adolescents were correlated by store—either because some individuals visited stores multiple times, or because friends in the store at the same time may be similar. We acknowledged the former in our limitations section but could not make an adjustment because we did not have individual-level identifiers in the data. However, we ran our analysis again to account for possible clustering at the store level, and the significance of our results remained unchanged.