Joshua Petimar and Jason P. Block

Most nutrition policies that are enacted to enhance diet quality (e.g. nutrition labelling, or beverage taxes) aim to change consumer behaviour in the food retail environment. While their ultimate success is measured by improvements to diet and health, initial success should be captured through sales. Surveying customers in retail settings is perhaps the most common method of collecting sales information, but this strategy has many limitations. Participant enrollment can be resource-intensive, and food retail managers and owners often use blanket anti-solicitation policies to inhibit research in their establishments. Data collection therefore often occurs in busy parking lots or sidewalks, leading to low response rates that could damage both a study’s generalizability and internal validity. Sample sizes in these studies are often too small to detect modest, but potentially meaningful effects of nutrition policies.

Many of the pitfalls of collecting customer-level receipt data can be avoided by acquiring sales data through strategic retail partnerships. Our study in The BMJ highlights the benefits of these partnerships. We collected three years of sales data (49 million transactions) from a large franchisee of a national fast food company to evaluate calorie labelling, a policy that has been implemented in major food establishment chains in the US. The small effects we estimated (a 4% immediate decrease in calories purchased per transaction followed by a <1% post-implementation weekly increase) could never be detected even with receipts from tens of thousands of participants. 

The sales data we used in this study come with benefits beyond statistical power. Primarily, they included all sales made in the franchise over the three-year period and are, by design, generalizable to all customers of those restaurants during this period. We effectively estimated the precise effect in the entire population, though we included confidence intervals because we viewed the franchise as a sample of the company’s restaurants. Lastly, the data in our study were recorded by retail software and did not require direct participation by customers, who may not accurately record or describe their purchases in a survey-based study. 

The benefits that come from sales data acquired through retail partnerships are not without limitations. Retailers are often reluctant to share these data; they have very little to gain by doing so, especially with clear data use agreements that allow freedom to publish results. Using sales data can also be a messy enterprise because the data are not collected for the purposes of research. Without an adequate data dictionary or knowledge of the franchise’s internal collection procedures (some of which changed at various points over the study period), we spent significantly more time exploring and cleaning the data than if we had collected it ourselves. This issue can be exacerbated if the partnership is not collaborative; some retailers may agree to share their data, but not otherwise be open to answering questions about specific variables and general data collection. Sales data also often have no specific individual-level information on customers; in our study, we inferred some of this information from ecological US census data.

Retail partnerships are nevertheless invaluable for answering nutrition policy-related questions with sufficient power and accuracy. In this era of “big data,” which makes it easier and less expensive than ever to acquire, manage, and analyze large datasets, there are few drawbacks to collecting retail sales data to answer questions concerning the broad food environment. Regular collection of these data by academic institutions and governments would provide a gold mine of resources to evaluate large-scale policies. Such an initiative may foster important discoveries not only in population health but also for retailers interested in learning more about how policies change consumer behaviour.