Are Sugar-Sweetened Beverage Information Labels Well-Targeted: Evidence and Welfare Implications

We will use online ads to recruit people who are shopping online for soda, and direct them to a two-stage online Qualtrics experiment. In the first stage (Part I), we will elicit subjects' nutritional literacy and self-control for resisting sugary drinks, following the methodology of Allcott et al. (2019). Allcott et al. (2019) show that these two measures explain 31% of sugary drinks consumption in a nationally-representative sample of approximately 20,000 households, after controlling for detailed demographic information and elicitations of "tastes" for various types of drinks. Additionally, Allcott et al. (2019) demonstrate how these measures can be used to provide an estimate of over-consumption of sugary drinks at the individual level. Consequently, the first stage allows us to obtain estimates of the internality.

A week later, we will recruit participants in the first stage to participate in the second stage of the experiment (Part II). We create a time delay to minimize the potential effect of the first-stage questions on the second-stage choices. In the second stage, we will first ask participants to select their three favorite sugary drinks from a list. We will then ask participants to make an initial set of choices between those drinks and sugar-free alternatives (e.g., Seagram's Ginger Ale versus 365 Ginger) at various relative prices. Specifically, we will elicit their willingness to pay using a multiple price list, as is standard in the behavioral economics literature, where consumers answer a series of questions about whether they would prefer to buy a sugar-sweetened beverage at price X or a sugar-free alternative at price Y. Choices will be incentive-compatible: for each consumer we will honor their choice in a randomly selected decision (i.e., we will purchase the chosen package of soft drinks for them plus award them any additional money associated with that choice).

Following an initial set of choices, we will ask subjects to choose again. In this second set of choices, we will randomize whether participants see the soft drink options paired with an information label. We will also randomize the types of labels, including: (i) an inflated nutrition label that draws attention to the differences in calories and sugar content, (ii) the stoplight warning label developed by Grummon, Hall, et al. (2019), and (iii) a graphic warning label developed by Donnelly et al. (2018). The control group will instead make the same choices without seeing any label. Leveraging the within-subject nature of design, we will then estimate how well-targeted each information label is by estimating how the subjects' internalities covary with their changes in the prices at which they are willing to buy soft drinks after seeing the information label.