Table 3.
Per capita annual beverage and tax spending as proportion of household income: Estimates from ordinary least squared regressions of log-transformed spending on income categories.
| Philadelphia | Seattle | San Francisco | ||||
|---|---|---|---|---|---|---|
| Log of proportion of HH income paid in taxes per capita (exponentiated) | Proportion of HH income paid in taxes per capita | Log of Proportion of HH income paid in taxes per capita (exponentiated) | Proportion of HH income paid in taxes per capita | Log of Proportion of HH income paid in taxes per capita (exponentiated) | Proportion of HH income paid in taxes per capita | |
| Panel A: Beta (95% Confidence Interval) | ||||||
| Lowest income (constant), < 200 %FPL | -Ref- | -Ref- | -Ref- | -Ref- | -Ref- | -Ref- |
| Middle income, 200−400 %FPL | 0.33 (0.22, 0.52) | − 0.37 (−0.55, − 0.18) | 0.27 (0.13, 0.51) | − 0.15 (−0.3, 0.00073) | 0.46 (0.22, 0.91) | − 0.035 (−0.061, −0.0091) |
| Highest income, > 400 %FPL | 0.15 (0.10, 0.22) | − 0.44 (−0.62, − 0.26) | 0.08 (0.041, 0.17) | − 0.17 (−0.32, − 0.023) | 0.25 (0.14, 0.46) | − 0.049 (−0.073, −0.024) |
| p−value for test of highest income=medium income | <0.01 | <0.01 | <0.01 | <0.01 | 0.011 | 0.02 |
| Panel B: Predicted percent of income paid in beverage tax (US dollars per year) based on regression estimates from Panel A | ||||||
| Lowest income | 0.17% (0.12, 0.22) | 0.50% (0.32, 0.68) | 0.07% (0.041, 0.12) | 0.20% (0.045, 0.35) | 0.02% (0.01, 0.04) | 0.06% (0.039, 0.086) |
| Middle income | 0.06% (0.036, 0.083) | 0.13% (−0.055, 0.32) | 0.02% (0.010, 0.040) | 0.05% (−0.10, 0.20) | 0.01% (0.0056, 0.022) | 0.03% (0.0020, 0.054) |
| Highest income | 0.02% (0.017, 0.036) | 0.06% (−0.12, 0.24) | 0.01% (0.003, 0.012) | 0.03% (−0.12, 0.18) | 0.01% (0.0033, 0.011) | 0.01% (−0.010, 0.038) |
| Observations | 585 | 585 | 212 | 212 | 344 | 344 |
Source/Notes: Author’s calculations based on Homescan and OmniPanel data. Panel A displays exponentiated regression coefficients for the log of dollars spent or dollars spent on the tax as a proportion of household income, with income categories modeled as indicator variables. The low-income group is the referent category, which is the constant term in these models. The coefficients for middle- and high-income groups are the difference in spending as a proportion of income between each income group and the low-income group. Confidence intervals that include the null value (0) indicate that the estimates are not statistically significantly different from estimate for the low-income group. Since the outcomes are log-transformed, exponentiating the coefficients for middle and high income gives the relative difference between low-income and that income group. For instance, in Philadelphia, the coefficient for the middle income group was –1.1; exponentiation of this coefficient equals 0.33, indicating that mean spending in the middle income group is 33% of the value for the lowest income group (or stated another way, 67% lower than the lowest income group (1–0.33 = 0.67)).
Panel B displays the estimates back transformed to predicted proportion of income for each income group. These are obtained through exponentiating the linear combination of relevant coefficients.