Sample exclusion criteria and number and percent of observations per exclusion criterion at the purchase and household-month levels.

<p>Notes: 1. In February 2017, Kantar Worldpanel (KWP) switched from calendar months to custom production periods running from the last week of a month to the last week of the following month to compile purchase records. As a result, purchase records were allocated to both the first and second production periods of 2017. To identify the duplicated records, KWP modified their purchase date year to 1997. Since we aggregated data at the household-calendar month level, these records were removed so as not to double-count them. 2. Bulk products were excluded due to lack of product-specific information (e.g., weight, barcode) allowing for linkage to nutrition facts panel data. 3. Across data years, between 93.4% and 95.7% of all zero-price non-bulk purchases were reported as gifts. As a proportion of total non-bulk purchases, the prevalence of these purchases has been relatively constant, ranging from 1.5% to 3.4% by data year. 4. Including purchases of the only product that was recategorized as a snack upon the introduction of this category, based on a search of snack barcodes in other categories. 5. 5,252,524 after aggregating at the household-day-product level. After all exclusions were implemented, a further 453 household-day-product observations were dropped because the calendar month of the purchase was the month preceding the household’s first production period, as households newly added to the panel after the production period change mentioned in Note 1 begin contributing data at the end of their first calendar month rather than at the start. 6. Another 24 household-month observations were excluded from the estimation sample because of missing administrative region, which meant that the regional unemployment rate, which is included in the models, was missing for these observations.</p> <p>(DOCX)</p>