Sugar-sweetened beverages (SSB) are significant sources of added sugar and calories in the American diet [1
]. High consumption of SSB is associated with increased risk of obesity, type 2 diabetes, cardiovascular disease and other health problems [3
]. Reasons for high consumption include the low price of SSB relative to healthier beverages, high availability, and aggressive marketing [9
]. A systematic literature review of food price elasticity found that the demand for SSB is responsive to price changes [12
], which forms the basis for implementing a beverage tax to reduce consumption.
Several municipalities (i.e., Berkeley, Boulder, SF, Seattle, Oakland, Albany) have enacted sugary beverage taxes as a way to generate needed revenue from sales of luxury/discretionary items [13
], and as part of a multi-pronged approach to improve population health [14
]. The Philadelphia beverage tax proposal was introduced with the explicit goal of financing universal pre-kindergarten programs and improvements to parks and recreation facilities, but did not mention potential benefits to health due to reduced SSB consumption [15
]. The Philadelphia beverage tax was the largest beverage tax at the time it passed ($
0.015/ounce) and has been in effect since 1 January 2017. Unlike other city-level beverage taxes that affect only SSB, the Philadelphia beverage tax covers both sugar-sweetened and diet beverages (SSDB). Recent reports have documented that, after one year, the Philadelphia beverage tax has been fully passed along to customers (97% pass-through rate) leading to an average of 34% price increase for taxed beverages [16
Prior work describing the impacts of beverage taxes, have largely focused on changes in beverage purchases/sales, with most studies finding declines compared with predicted sales in the absence of the tax. One year following the tax, SSB sales declined by 9.6% in Berkeley, California, and by 22% in Seattle, Washington [17
]. In Mexico, where cross-border purchases were minimal because the tax was implemented nationwide, a 7.6% decline was reported after two years of the tax [19
]. Similar beverage tax in Catalonia reduced the prevalence of regular taxed beverage consumers by 39% [20
]. A recent systematic review and meta-analysis of the real-world SSB tax impact found that a 10% price increase caused by the tax was associated with an average of 10% reduction in beverage purchases [21
]. With more than a 30% price increase due to the Philadelphia beverage tax, two recent studies reported large declines in sales from 46% to 51% in Philadelphia, and from 22% to 24% after adjusting for cross-border shopping [16
]. Another study reported that consumers decreased SSB purchase by 8.5 ounces per shopping trip in Philadelphia relative to its comparison communities [23
]. However, a decline in sales or purchases within the taxed area does not necessarily translate into a reduction in consumption, which is a more public health-relevant outcome, because consumers may increase purchases of taxed beverages in nearby jurisdictions where beverages are not taxed (i.e., cross-border shopping). [16
] In addition, the sales data do not capture out-of-home consumption and purchases from stores not participating in the large market research networks.
There is limited empirical evidence of beverage tax impacts on consumption in US contexts. Previously, our team examined the impact of the Philadelphia SSDB tax on consumption using a repeat cross-sectional design, and found a decrease in frequent sugary soda consumption and an increase in bottled water consumption shortly after tax implementation [24
]. Similar short-term results were reported in another study conducted in Berkeley [25
]. The results of longer-term impacts are mixed. Some studies have found declines in consumption [26
] while other studies suggest that the impact on total SSB demand or consumption at the population level may be small [16
]. In the present study, we examined the within-person change in bottled water and SSDB consumption in Philadelphia, relative to nearby comparison cities, one year after Philadelphia’s SSDB tax. Our study adds to the small, but growing, literature on the longer-term impact of beverage taxes on consumption, utilizing a quasi-experimental difference-in-differences design in a population-based sample.
Participants’ baseline characteristics are described in Table 1
. The final analytic sample included 515 participants (357 in Philly and 158 in non-Philly). Forty-nine percent were male, 52.2% were white, 42.5% were older than 50 years, 19.8% had income per capita lower than $
15,000, 28.0% had educational attainment at high school level or lower, 28.7% were obese, 13.8% reported fair or poor self-rated health status, 40.4% reported having been told by a health professional that they had chronic health condition, 15.5% were currently smoking and 23.7% had higher alcohol use. About two-thirds of the participants were reached by cell phone and one-third lived near the Philly border at baseline. Table 1
shows that Philly and non-Philly participants mostly shared demographic and health characteristics. Philly was less advantaged (according to socioeconomic status and health variables) but, across most variables, the differences were small.
Bottled water and SSDB consumption during baseline and follow-up are summarized in Table 2
and one-year within-person changes are summarized in Table 3
. Due to wide distribution of the consumption data, the SDs for the within-person changes in consumption frequency and ounces are extremely large. The bivariate Spearman correlation between the baseline and one-year consumption of SSBs was approximately 0.60 (not shown in tables). We observed declines in SSB consumption in Philly and non-Philly. The Philly group had a relatively higher percentage decrease and lower percentage increase in SSB consumption compared with the non-Philly group. Consumption of diet beverages remained almost the same for both groups. One year after the tax, in Philly and non-Philly, similar proportions of participants had an increase or a decrease in bottled water consumption. The majority of participants in Philly and non-Philly had no change in the category of being a daily drinker of bottled water or SSDB. Very few participants changed their daily consumption behavior, and the results for this outcome were unstable (Tables S4 and S5
The adjusted difference-in-differences analysis results are shown in Table 4
. No significant changes were found in consumption over time between Philly and non-Philly. For example, the estimated difference-in-difference in 30-day SSB consumption frequency and ounces in this study was small (−3.03 times per month or −51.65 ounces) and confidence intervals were very wide and p
-values were approximately 0.5 (Table 4
-A). A small sample size and high variability led to high uncertainty in within-person adjusted results (all p
-values ≥ 0.11). Results were insensitive to models without adjustment for covariates (difference-in-differences results without adjustment for covariates were largely the same) and results were insensitive to using log-binomial models (multinomial results were similar and inference was unchanged, Table S6
Sensitivity to different definitions of the pre-tax period are presented in Figure S3
; the point estimates were unstable and results were not statistically significant. Furthermore, we tested sensitivity to treatment contamination and/or lower-dose tax exposure in a few ways. First, we examined sensitivity of results to excluding participants who lived on the Philly border (43.4% of the Philly sample and 6.3% of the non-Philly sample resided in a ZIP code that touched the Philly city border, excluded N = 165); results remained null (Figure S3
). Next, we excluded Philly residents who regularly consumed SSB and reported only purchasing those beverages outside of Philly at follow-up (excluded N = 65); results also remained null (Figure S3
We conducted a population-based longitudinal study using a random-digit-dialing phone survey of 515 participants in Philly and outside of Philly to estimate the impact of the Philadelphia beverage tax on consumption one year after implementation. In Philly and outside of Philly, there was a very small decline in monthly SSB consumption (for example, −3.0 times or −50 ounces). In Philly, a slightly higher proportion of participants decreased their monthly SSB consumption (39.2% in Philly vs. 33.5% non-Philly) and a slightly lower proportion increased their monthly SSB consumption (38.9% in Philly vs. 43.0% non-Philly). However, the majority of the participants (around 80% in both groups) did not change daily consumption of SSBs. After covariates adjustment, there were no differences between Philly and non-Philly in within-person change in consumption regardless of the type of beverage or operationalization (daily, continuous ounces, continuous frequency of consumption). Finally, the small effect sizes found in this study suggest that the tax may not have a substantive impact on population level health. We were unable to test whether the tax’s effect differed in sub-populations.
Notable strengths of our study were that we assessed changes in residents’ consumption, utilized a cohort and contrasted within-person changes in the tax site vs. multiple comparison areas in neighboring states, and that our study was population-based and thus designed to assess the impact of the tax on the general population. Compared to the census, our analytic sample was slightly older and had higher median income, nevertheless it was roughly comparable on other census characteristics [30
]. Furthermore, the prevalence of SSB consumption in our analytic sample (e.g., 24% daily SSB consumption and 11% daily soda consumption) roughly aligned with the prevalence of SSBs reported in other population-based samples within Philadelphia [38
To date, only one Philadelphia tax study has evaluated the tax’s impact on consumption. Cawley et al. enrolled shoppers via exit interviews and collected beverage consumption data before and one year after the tax [23
]. They over-sampled disadvantaged residents (sample was over 60% African-American, majority low income and eligible for food assistance programs, most households had children). Their analytic sample of approximately 400 participants had high consumption of soda (approximately one-half of the sample consumed soda daily). One year after the tax, there was a small decrease in adult soda consumption in Philadelphia (−7.2 times per month) whereas consumption increased at comparison sites (+10.3 times per month). This resulted in a −10 to −11 times per month adjusted difference-in-difference (DID) decline in adult soda consumption frequency, but not for other beverages [23
]. To summarize, differences in recent results from Cawley et al. and our study were likely due to Cawley et al. utilizing a population sub-sample (majority disadvantaged, adults living with children, and adults with a high consumption of SSB at baseline).
Experience in Berkeley also suggested that the beverage tax has little effect on the general population, but does have an impact within disadvantaged subgroups and among persons with high SSB consumption. In Berkeley, a population-based study by Silver et al. (2017) conducted one year following the tax (which was also limited by small sample sizes, large standard errors and low baseline consumption) found very small non-significant declines in taxed beverage consumption (compared to the non-Berkeley area, −25 ounces per month or −0.85 ounces per day, consumption frequency was not reported in this study) [17
]. However, another Berkeley study conducted by Lee et al. (2019) three years after the tax enrolled a more demographically diverse sample of participants who drank more SSBs. They found SSB declined in Berkeley but not at the comparison site, resulting in modest difference-in-difference (DID) reductions in SSB consumption (declined −0.55 times per day, roughly −16.5 times per month) and increased consumption of water (+1.02 times per day, roughly 30.6 times per month) [27
Our current study suggests that the tax’s impact on general population consumption of SSB may have been more modest than was originally anticipated. Early prediction/simulation studies, supported by theories of price-elasticity, forecasted a fairly large impact of the beverage tax on SSB consumption [12
]. However, most of the simulations did not account for many negative pressures on health decision making, such as persistent consumer preferences, intensification of marketing by SSB manufacturers or retailers, or availability of cross-border shopping. Thus, early work may have over-estimated the simulated impact of the tax [39
]. Furthermore, recent studies have reported large and significant reductions in sales of/demand for taxed beverages in Philadelphia vs. comparison areas. [16
]. Results from consumption studies like ours may be discordant with sales/demand studies for a number of reasons. Sales data do not capture out-of-home consumption (e.g., restaurants, vending machines, friends’ homes) and typically represent only larger-format retail establishments. Additionally, sales data studies have difficulty accounting for cross-border purchases; consumers may increase purchases of taxed beverages in nearby jurisdictions where beverages are not taxed [16
]. Indeed, the two studies that examined Philadelphians’ sales/demand for taxed beverages estimated that the impact of the tax on SSB demand was attenuated due to cross-border shopping [16
]. The exit interview conducted by Cawley et al. found that Philadelphians were not more likely to travel outside of the city to shop due to the tax, but when they did shop outside, they purchased more taxed beverages [23
]. Because our analyses focused on consumption, the results will not be biased by cross-border shopping. Nevertheless, for descriptive purposes, we collected the taxed beverage shopping location (i.e., only in Philly, only outside of Philly, or both) among a subset of the sample; and those results also suggest increases in cross-border shopping. Among the subsample of Philadelphians asked the question, the proportion who only purchased SSDB outside Philly increased from 15.1% (37/245) at baseline to 25.6% (65/254) after one year, and shopping in and outside of Philly increased slightly from 25.7% (63/245) to 32% (82/254).
This study has several limitations. First, our sample size was relatively small, and due to the large variance in consumption data, this study was only powered to detect large changes relative to the baseline consumption. The Cawley study had a smaller sample than ours, but had sufficient power to detect statistically significant changes due to observing a larger average effect (DID −17.5 SSB per month vs. our very small DID −3.03 SSB per month) and due to enrolling a relatively homogeneous sample. Studies conducted in population-based samples reported similar variability in beverage consumption as observed in our study [25
]. Interestingly, Cawley et al. (2019) conducted a similar exit interview in Oakland, California in a small sample with similarly low baseline beverage consumption (22.63 times per month) as in our study, and they reported no substantial changes in overall consumption of SSB for either adults or children one year after the tax [28
]. Second, our results may be generalizable to other large US cities with demographically diverse populations, but are not likely generalizable to subpopulations, such as high consumers of soda and lower income residents. Third, attrition in the follow-up survey was higher among Hispanic, lower income, lower education participants who may be more likely to consume more SSB and be price-sensitive [41
]. This pattern was similar for Philly and non-Philly, except that a larger share of non-Whites were loss to follow-up in non-Philly. Hispanic and Black participants had higher SSB consumption at baseline and follow-up (data not shown). If participants with high SSB consumption would react to the tax more than others, the impact of the tax might be overestimated. Fourth, with only one pre-tax time period, we were unable to test the parallel trends assumption of the difference-in-differences method (which assumes the average outcomes for treated and control groups would have followed parallel paths over time). Fifth, we adjusted for differences in baseline characteristics in the Philly group and the non-Philly group by including covariates in the difference-in-differences model and time invariant characteristics were controlled for via within-person change analyses. Nevertheless, unobserved time-varying confounding remains a possibility. Finally, self-reported consumption of beverages likely has measurement error. We adapted a validated beverage questionnaire and results represent within-person changes in consumption, thus accounting for correlated errors within an individual’s self-report.