Sugar-Sweetened and Diet Beverage Consumption in Philadelphia One Year after the Beverage Tax

In January 2017, Philadelphia (Pennsylvania) implemented an excise tax ($ 0.015/ounce) on sugar-sweetened and diet beverages. This study is a general population-based study to report on the longer-term impacts of the tax on within-person changes in consumption 12 months after implementation. A quasi-experimental difference-in-difference design was used to contrast Philadelphia vs. nearby comparison cities (Trenton, New Jersey; Camden, New Jersey; and Wilmington, Delaware) at baseline (December 2016–January 2017) vs. 12-month follow-up (December 2017–February 2018). A random-digit-dialing phone survey was administered to a population-based cohort. Analyses assessed changes in 30-day consumption frequency and ounces of sugar-sweetened and diet beverages (and a substitution beverage, bottled water) in the analytic sample (N = 515). After 12 months, relative to the comparison group, Philadelphians were slightly more likely to decrease their frequency of sugar-sweetened beverage consumption (39.2% vs. 33.5%), and slightly less likely to increase their frequency of sugar-sweetened beverage consumption (38.9% vs. 43.0%). The effects of the tax estimated in the adjusted difference-in-difference analysis were very small (for example, changes in monthly sugar-sweetened beverage consumption in Philadelphia relative to comparison cities was −3.03 times or −51.65 ounces) and confidence intervals were very wide. Results suggested that, one year after implementation, there was no major overall impact of the tax on general population-level consumption of sugar-sweetened or diet beverages, or bottled water. Future studies should test whether the tax’s effect differs in vulnerable sub-populations.


List of tables and figures
. Number (%) of survey participants by time interval during follow-up survey. Table S2. Baseline characteristics of participants retained in the analytic sample (N = 515) and dropped out due to loss-to- follow-up (N = 1904) . Table S3. Description of the difference-in-differences models. Table S4. Daily consumption of bottled water, sugar-sweetened and diet beverage during baseline and follow-up (N = 515). Table S5. Within-person change in daily consumption of bottled water, sugar-sweetened and diet beverage during baseline and follow-up (N=515). Table S6. Difference-in-differences analysis (adjusted for covariates): relative risk of being in the specified category 12 months after the tax in Philadelphia relative to that of the comparison cities (N = 515).    1 Note that the time intervals are not equally spaced. Survey participants did not join the study at a constant rate and, thus, the time intervals were chosen in order to have sufficient numbers of participants in each time interval. 2 First, missing baseline characteristics were inputted using data collected during follow-up. Second, missing education and income per capita data were inputted using census data. Missing race were inputted by race consisted at least 50% of the census tract population. Missing per capita income were inputted by the per capita income in the census tract. Participants aged 18-20 at baseline were assigned high school/GED education if they lived in a census tract with more than 50% finished high school, or some high school if otherwise. Participants older than 20 at baseline were assigned college education if they lived in a census tract with more than 50% finished college, or high school/GED if otherwise. Finally, the rest of the missing baseline characteristics were inputted by sample means by age group, sex and race. Participants missing beverage consumption data during either baseline or follow-up were excluded. 3 Participants who reported consuming more than 10,000 ounces of sweetened or diet beverages during the past 30 days were identified as outliers and excluded from the main analysis. Only two participants were excluded due to unreasonably extreme values. Table S2. Baseline characteristics of participants retained in the analytic sample (N = 515) and dropped out due to loss-to-follow-up (N = 1904) 1.

Baseline characteristics Philly Retained
Philly Dropped-out  Table S2. Footnote 1 Missing baseline characteristics were inputted using follow-up data, census data (<5% of participants) and sample mean by age group, sex and race (<0.5% of participants). 2 There were very few underweight (<2%), thus, they were grouped with normal weight. 3 Presence of a chronic condition was assessed by asking whether the participant was ever told by a doctor, nurse, or other health professional that they had at least one of the following: high blood pressure, high cholesterol, diabetes, or history of heart disease. 4 Only participants aged >=21 were asked about alcohol consumption. Alcohol use "higher" was defined as more than seven drinks for female or 14 drinks for male.
[37] 5 Lived in ZIP code that is on the Philadelphia border. Abbreviation: GED, General Education Development test. Table S3. Description of the difference-in-differences models.
Linear regression for continuous outcomes ∆ = 1 + 2 ℎ + ∆ is the within-person change in the volume or frequency of beverage consumption. ℎ is the treatment group ("Philly" vs. "non-Philly"). The coefficient of interest is 2 , which can be interpreted as the change in beverage consumption volume or frequency in 12 months after the tax in Philadelphia relative to that of the comparison cities. Multinomial logistic regression for binary or categorical outcomes log Pr( =m) Pr( = ℎ ) = 1 + 2 ℎ + is the categorical variable for within-person change beverage consumption behavior. ℎ is the treatment group ("Philly" vs. "non-Philly"). The coefficient of interest is ( 2 ), which can be interpreted as the difference in the odds of being in a specific category vs. no change 12 months after the tax in Philadelphia relative to that of the comparison cities. Logistic regression for binary outcome log Pr(Y i =1) Pr(Y i =0) = β 1 + β 2 Period i + β 3 Philly i + β 4 Period i × Philly i + β j Cov j is the variable for daily consumption (1or 0). is the time during which the outcome is measured ("baseline" vs. "12-month follow-up"). ℎ is the treatment group ("Philly" vs. "non-Philly"). denotes the individual participants. The coefficient of interest is ( 4 ), which can be interpreted as the change in the odds of daily consumption 12 months after the tax in Philadelphia relative to that of the comparison cities.  Table S6. Difference-in-differences analysis (adjusted for covariates 1 ): relative risk of being in the specified category 12 months after the tax in Philadelphia relative to that of the comparison cities (N = 515).
1 Results were adjusted for baseline age, sex, race, education, income, body mass index, health status, smoking, alcohol use, survey method (cellphone vs. landline), if they lived in ZIP code that is on the Philadelphia border (to control for potential cross-border shopping), and week of baseline survey (to control for seasonal trend). With beverage consumption responses at baseline and 12-month follow-up (N=372)

Change in
After exclusion of participants due to unreasonably extreme consumption of SSDB (N=371) Completed the baseline survey on or before Jan 15, 2017 (N=158) Figure S2. Power analysis for sugar-sweetened beverage consumption.
A. Detectable effect size vs. power in a two-sample t-test for a normally distributed outcome with a standard error of 45 (approximately the standard error for within-person change in monthly sugarsweetened beverage consumption frequency in the Philly group) and sample size of 515 (357 in one group, 158 in another group).
B. Detectable effect size vs. power in a two-sample t-test for a normally distributed outcome with a standard error of 780 (approximately the standard error for within-person change in monthly sugarsweetened beverage consumption ounces in the Philly group) and sample size of 515 (357 in one group, 158 in another group). Figure S3. Sensitivity analysis. Difference-in-differences analysis adjusted for covariates.