Individuals with obesity and related conditions are often reluctant to change their diet. Evaluating the details of this reluctance is hampered by limited data. I use household scanner data to estimate food purchase response to a diagnosis of diabetes. I use a machine learning approach to infer diagnosis from purchases of diabetes–related products. On average, households show significant, but relatively small, calorie reductions. These reductions are concentrated in unhealthy foods, suggesting they reflect real efforts to improve diet. There is some heterogeneity in calorie changes across households, although this heterogeneity is not well predicted by demographics or baseline diet, despite large correlations between these factors and diagnosis. I suggest a theory of behavior change that may explain the limited overall change and the fact that heterogeneity is not predictable. (JEL D12, D83, D91, I12, M31)
OPAM: Online Purchasing-behavior Analysis using Machine learning
