Table 3.
Difference-in-difference Poisson regression models estimating the effects of SNAP issuance periods on SSB ad counts, adjusting for state and calendar time
| Short issuance (CT, 3 d; NE and NJ, 5 d) | Long issuance (CA, 10 d; TX, 15 d) | All states combined | ||||
|---|---|---|---|---|---|---|
| Mean | sd | Mean | sd | Mean | sd | |
| Summary statistics (all stores) | ||||||
| % Florida SSB ads in circulars during comparison states’ issuance days, mean (sd) | 46·7 | 13·4 | 47·2 | 13·9 | 47·2 | 13·9 |
| % Florida SSB ads in circulars during comparison states’ non-issuance days, mean (sd) | 48·2 | 12·7 | 48·5 | 11·6 | 48·5 | 11·6 |
| % Comparison states’ SSB ads in circulars during issuance days, mean (sd) | 50·0 | 15·9 | 52·4 | 18·3 | 51·5 | 17·5 |
| % Comparison states’ SSB ads in circulars during non-issuance days, mean (sd) | 47·2 | 15·5 | 50·7 | 19·0 | 48·3 | 16·7 |
| b | 95 % CI | b | 95 % CI | b | 95 % CI | |
| Coefficient estimates | ||||||
| State effect: Relative difference from FL, b (95 % CI) * | 0·987 | 0·945, 1·031 | 1·032 | 0·978, 1·088 | 1·002 | 0·961, 1·045 |
| State effect: Adjusted absolute difference from FL, b (95 % CI) § | −0·143 | –0·629, 0·343 | 0·260 | –0·183, 0·702 | 0·019 | –0·401, 0·439 |
| Time effect: Issuance relative difference, b (95 % CI) † | 0·983 | 0·976, 0·991 | 0·985 | 0·977, 0·993 | 0·984 | 0·976, 0·993 |
| Time effect: Absolute difference from issuance period, b (95 % CI) § | −0·186 | –0·274, –0·099 | −0·129 | –0·197, –0·061 | −0·160 | –0·245, –0·076 |
| State × Time effect: Relative difference in differences, b (95 % CI) ‡ | 1·031 | 1·022, 1·040 | 1·025 | 1·011, 1·039 | 1·029 | 1·019, 1·039 |
| State × Time effect: Absolute difference in differences, b (95 % CI) § | 0·339 | 0·239, 0·440 | 0·206 | 0·095, 0·317 | 0·284 | 0·187, 0·381 |
| Observations | 24 171 | 18 004 | 36 064 | |||
SNAP, Supplemental Nutrition Assistance Program; SSB, sugar-sweetened beverages.
Author’s analysis of daily ads derived from the weekly circulars of 563 stores in six states collected from August to September 2019. For each issuance duration in each column, the issuance period for Florida was reassigned to coincide with the longest issuance period of the comparison states before estimation. For example, when compared to short-issuance states (CT, NE and NJ), the issuance period in FL was recoded to occur in the first 5 d of the month. When compared to CA and TX, and to all states, the issuance days in FL were recoded to occur in the first 15 d of the month. All difference estimates are derived from Poisson difference-in-difference regressions, and they indicate the effect of issuance, state and their interaction on the number of SSB ads. Total number of beverage ads is included in each regression as a covariate with its coefficient constrained to 1. All models include state-level fixed effects, store location- and week-level random effects. Robust standard errors are adjusted for clustering within store.
Boldface indicates statistically significant difference at P < 0·05.
From the coefficient on the indicator variable for state that is equal to one for the comparison state or states and zero for Florida.
From the coefficient on the indicator variable for SNAP issuance day.
From the coefficient on the interaction of SNAP issuance day and the indicator for the comparison state or states.
From the margins command in Stata, version 15.