Department of Agricultural and Resource Economics, UCB UC Berkeley

Peer Reviewed Title: Soda Wars: Effect of a Soda Tax Election on Soda Purchases Author: Taylor, Rebecca Kaplan, Scott Villas-Boas, Sofia B, UC Berkeley Jung, Kevin Publication Date: 12-02-2016 Series: CUDARE Working Papers Permalink: http://escholarship.org/uc/item/0q18s7b7 Copyright Information: All rights reserved unless otherwise indicated. Contact the author or original publisher for any necessary permissions. eScholarship is not the copyright owner for deposited works. Learn more at http://www.escholarship.org/help_copyright.html#reuse

eScholarship provides open access, scholarly publishing services to the University of California and delivers a dynamic research platform to scholars worldwide.

Soda Wars: Effect of a Soda Tax Election on University Soda Purchases Rebecca Taylor, Scott Kaplan, Sofia B. Villas-Boas, and Kevin Jung∗ space space space December 1, 2016

Abstract This paper examines how consumers alter their behavior due to a local excise tax aimed at dealing with the potential health hazards of sugar consumption in sugar-sweetened beverages (SSB). Using panel data of product purchases from university retailers in Berkeley, California, we measure the consumption effects of a soda tax campaign and election. We implement difference-in-difference and event study empirical strategies to estimate the change in soda consumed relative to the change in consumption of control product categories. Our results show that soda consumption significantly drops immediately after the election, months before the tax is implemented in the City of Berkeley or on campus. We also estimate that beverages are inelastic goods, which implies that an actual price increase due to a tax would lead to a smaller than proportional drop in demand, making the tax a potential revenue booster. Our findings have interesting policy implications, suggesting the effects of media coverage and election outcomes on attitudes and behaviors may be larger than the effects of the soda tax itself. Keywords: Taxes, soda, media effects, differences-in-differences, event study. JEL Classification: C23, C25, D12, H20

∗

Author Affiliations: Villas-Boas is a Professor of Agricultural and Resource Economics (ARE) at the University of California, Berkeley. Kaplan and Taylor are PhD students in ARE at the University of California, Berkeley. Kevin Jung is an undergraduate student alumn from U. C Berkeley. Corresponding author email: [email protected]. We thank Jayson Lusk for suggestions and participants at the 2016 Agricultural & Applied Economics Association Annual Meeting in Boston, and seminar participants at UC Berkeley for helpful comments. We thank the Cal Dining staff in granting us access to the data. We especially thank Tracy Ann Stack, Samantha Lubow, and Jaylene Tang for their institutional knowledge and data support. We are grateful to the Giannini Foundation for financial support.

1

Soda Wars

1

Introduction

With the current trend of sugar consumption, exercise, and dietary habits, it is estimated that 40% of Americans born from 2000 to 2011 will get diabetes in their lifetimes, with the percentages for African-American women and Hispanics placed even higher at 50% (Gregg et al. 2014). While researchers and industry participants agree on the health dangers of sugar, and in particular sugar-sweetened beverages (SSB) such as soda and energy drinks, there is disagreement on how to design laws and policies to change behavior. Policy proposals include soda bans (James et al. 2004; Fernandes 2008; Huang and Kiesel 2012), soda taxes (Brownell and Frieden 2009), school nutrition education programs (James et al. 2004; Fernandes 2008), and warning labels on sugary drinks advising the dangers of obesity, diabetes, and tooth decay (Roberto et al. 2016). This raises the empirical questions: (1) how do consumers react to such policies and (2) are there differences between direct regulations and informational campaigns? This paper examines how consumers alter their purchasing behavior due to the campaign attention and election of a local excise tax aimed at curbing SSB consumption. We take advantage of a tax policy change—referred to as Measure D—in the city of Berkeley, California. Measure D imposes a penny-per-fluid-ounce tax to be paid by distributors of SSBs. The aim of the policy is to lower the consumption of SSBs, or if demand is deemed to be unresponsive,1 to raise tax revenues which could fund nutritional programs and education. On November 4, 2014, Measure D was put to a vote and passed with 75% in favor. An aggressive campaign war preceded this vote, dubbed “Berkeley vs. Big Soda.” This campaign cost $3.4 million, with roughly $1 million spent in favor of Measure D and $2.4 million spent against it.2 The specific objective of this paper is to examine how consumers reacted to the pre-tax media campaign and to the election outcome. There is evidence suggesting that highlighted news coverage can lead to sharp information updates (Huberman and Regev 2001) and investigating whether this also leads to behavioral changes has important policy implications, especially if behavioral changes happen before the policy change. Our study uses a detailed dataset from university retailers in Berkeley, consisting of monthly purchases by Universal Product Code (UPC). We use a difference-in-differences (DID) strategy to measure the

2

Soda Wars

change in SSB consumption against untreated products (in comparable control product categories), and untreated months (the pre-campaign period). Using monthly panel data allows us to control for seasonality in sales, and using a log specification of the DID model allows us to obtain an estimate of the price elasticity of beverages. Additionally, we estimate an event study model to test the identifying assumption of parallel trends in the pre-campaign period. We verify that soda would have evolved in the same trend as other products had there not been a soda tax campaign or affirmative election outcome. In the DID approach, we find a small and insignificant drop in soda sales during the campaign period and a large and significant drop following the election. In particular, after the election yet before the SSB tax implementation, the quantity of soda sold relative to other beverage and candy products declines by 30%. From the event study analysis, we find that the control categories exhibit similar pre-campaign trends as the treated soda category, yet after the election (and before the actual pass-through of the tax) soda sales drop significantly compared to the control categories. We also estimate that beverages are inelastic products. Therefore, we predict that when the tax is eventually passed through into the prices that campus retail customers see, the price increase will lead to a smaller than proportional drop in demand. This implies that a soda tax will not result in elastic soda quantity reductions, but rather can serve as a category suitable for increasing tax revenues. It is important to note that the university retailers in our analysis are not representative of the average U.S. retail outlet, especially in terms of clientele, and this could have large implications for whether our results will generalize to other locations. However, there are several advantages of using this empirical setting for our experimental design. First, the layout and products offered, as well as the promotional effort and posted prices, are uniform across campus locations. Second, we know exactly when and by how much the soda tax is passed on to consumer, and do not have to infer the pass-through from the data. Since SSB taxes are often levied on the distributors of SSBs—who have a choice on how much of the tax they will pass on to consumers3 —there is an empirical literature asking who bears the SSB tax incidence. Cawley and Frisvold (2015) and Falbe et al. (2015) examine the incidence effects of the Berkeley soda tax, both finding that roughly half of the tax was passed on to consumers in Berkeley three to four months after the tax implementation within the city of 3

Soda Wars

Berkeley in March 2015. Aguilar et al. (2016) estimate the consumer incidence effects of a sugary drink tax in Mexico and find more than full price pass-through for sugary drinks in the year after tax implementation. In our setting, we know exactly when and by how much the campus retailer changes prices. In particular, due to the costs of changing prices, our campus retailers chose not to pass-through the tax to consumers for a year after receiving the tax invoices, and thus we are able to look at how soda demand changes on-campus when the prices off-campus start to increase yet the prices on-campus remain the same. Our paper relates to the literature on whether consumers update their beliefs and change their behavior based on information provided by the media and by advisory campaigns. Several studies show that new information about food-related health problems and foodsafety can alter preferences and consumer demand (Chavas 1983; Brown and Schrader 1990; Van Ravenswaay and Hoehn 1991; Yen and Jensen 1996). For example, Smith et al. (1988) analyze the impact of an incident involving contamination of milk with heptachlor in Hawaii in 1982 and find that negative media coverage has a larger impact than positive coverage. The approach of our analysis is particularly close to Schlenker and Villas-Boas (2009), who compare an actual food scare in the beef market with a media-discussed related event. The authors find that the media covered event had almost 50% of the effect of the actual food scare event.4 Our paper also fits into a growing literature informing policymakers about the potential impacts of soda taxes. Using quasi-experimental methods and purchase data, we are able to contrast the quantity effects of the SSB campaign and election with the predicted quantity effects of a price increase from the soda tax. A closely related study (Falbe et al. 2016) uses survey-based evidence on SSB consumption—comparing the responses of survey participants in Berkeley to survey participants in Oakland and San Francisco, which act as control cities where SSB taxes were not voted into law. Falbe et al. (2016) estimate that the quantity of SSBs purchased in Berkeley dropped by 21%. We extend their analysis by using actual purchase data, rather than stated consumption levels which could be biased. Furthermore, Falbe et al. (2016) conduct the surveys in two separate blocks of time—before the campaign and after the tax implementation—and thus they cannot distinguish between the election’s effect and the tax’s effect on SSB demand. 4

Soda Wars

Other studies have found smaller consumption responses to SSB taxes than those found in Berkeley. Aguilar et al. (2016), who use a quasi-experimental approach similar to ours and weekly scanner panel data, examine the effects of a countrywide sugary drink tax in Mexico on the consumption of SSBs and the substitution toward other non-taxed goods. They find a 6% decrease in sugary drink calories purchased; however, they do not find a change in the total calories purchased across all products in their dataset. Moreover, while they document a decrease in the consumption of sugar, they also document an increase in the consumption of fat, sodium, and cholesterol, indicating a change in the composition of the food basket. In the context of the U.S., Fletcher et al. (2010) use the variation in soda taxes across states and estimate that a 1% increase in the tax implies a reduction of only 6 soda-calories per day, but that those calories are shifted to calorie consumption from other foods. Our results suggest that the larger reduction in consumption in Berkeley reflects the effects of the tax campaign and election, which may have shifted social norms. The Berkeley soda tax differs from other soda taxes in several ways. In particular, it was voted on and passed by the people of Berkeley, and there was an extensive campaign to inform voters about the tax. Conversely, when the Mexican government announced the soda tax in September 2013, it took the soda industry and the public by surprise according to media reports.5 In light of these differences, an important policy implication of our study is that the effects of media coverage and election outcomes on attitudes and behaviors may be larger than the effects of the policy itself. The rest of the paper proceeds as follows. Section 2 describes the empirical setting and summarizes the data, while Section 3 outlines the research design (i.e. the DID and event study strategies). Section 4 presents the results and Section 5 concludes. 2

Empirical Setting and Data

Since 2009, the soda industry has spent more than $117 million nationally to stop soda tax initiatives, such as those considered by the U.S. Congress and in states such as Maine, Texas, and New York.6 For Berkeley’s Measure D in particular, the American Beverage Association of California contributed almost $2.5 million to defeat the tax, while supporters

5

Soda Wars of Measure D spent just under $1 million.7 One of the strongest supporters of Measure D—“Berkeley vs. Big Soda”—gathered industry, individual and lawmakers’ support and funded an aggressive advertising campaign promoting “Yes on D”, emphasizing the need to fight “Big Soda”. While, as previously mentioned, the SSB tax in Measure D affects all beverages containing sugar, at a rate of $0.01 per ounce, our survey of the media and advertising campaign concluded that the media paid particular attention to soda, rather than SSB products in general (see appendix figure A.1 for one such example). Thus, we will look at the effects of the campaign war on soda separately from other SSB products. Given that time series data on campaign expenditures are not available, we investigate the intensity of the campaign over time by examining media article search data for the term “soda tax”. Figure 1 depicts monthly Google trends data for news searches of the term “soda tax” in the San Francisco-Oakland-San Jose area, from 2010 though 2015. This figure indicates that news searches spiked in November 2014, when Measure D was voted on and passed into law. We use a unique data source to estimate the effect of media coverage and campaigning on consumer purchasing decisions: a retail dataset from dining locations at a large university. This dataset includes monthly data on quantities sold and revenue sales at the product level— i.e., campus retailers sold x units of product i in month m, where a product is represented by a unique bar-code (UPC). The dataset includes all beverage products, as well as all chocolate and candy products, for the period November 2013 through December 2015.8 We categorize products into nine product groups: 1) soda, 2) water, 3) juice, 4) energy drinks, 5) milk, 6) coffee, 7) tea, 8) diet drinks, and 9) candy. The university retailers where we perform the empirical analysis may not be representative of average U.S. purchase outlets, but there are several advantages of using this empirical setting for our experimental design. First, the layout and products offered are uniform across campus locations. Second, the promotional effort and posted prices are common across campus. Third, we know exactly when and by how much the soda tax is passed on to consumer. For the campaign effect analysis, we define regular soda as our treated product category, which we will compare to the eight other beverage and candy product groups. However, it is important to note that regular soda is not the only product that falls under regulation. 6

Soda Wars

Given the wording of Measure D—“The City hereby levies a tax of one cent ($0.01) per fluid ounce on the privilege of distributing sugar-sweetened beverage products in the city”— any drinks with added sweeteners are taxed. So for example, 100% juices are not taxed, but juices with sugar or corn syrup added are taxed. The following beverage products are taxed: regular soda (i.e. Pepsi), sport and energy drinks (i.e. Gatorade), sweetened tea, and lemonade. Exempted are the following: water, diet soda, diet energy drinks, beverages containing only natural fruit and vegetable juice, beverages in which milk is the primary ingredient, beverages or liquids sold for purposes of weight reduction as a meal replacement, medical beverages beverages used as oral nutritional therapy or oral rehydration electrolyte solutions for infants and children), and alcoholic beverages, although the last two categories are not sold on campus. First we use the pre-campaign period data to investigate whether the pre-period is balanced in terms of pre-existing trends in demand for the treated (soda) and control (other beverages and candy) product categories. Figure 2 presents the quantities sold of each product group per month in the pre-campaign period. The largest category in terms of quantity sold in the pre-campaign period is juice, followed by energy drinks and water, and then soda, milk, tea, coffee, and candy see lower levels of sales.9 While the various products differ in levels, their trends are quite similar, with sales peaking in April—the weeks leading up to final exams—and plummeting in June—after the Spring semester ends.10 Thus, while soda has different quantities sold than the other products, to the extent that these differences are constant over time, product group fixed effects will control for all possible time invariant determinants of drink and candy demand. In evaluating the effects of the soda tax campaign, we will compare the pre-campaign period to three separate post-campaign periods: (1) the pre-election campaign period—July 2014-October 2014, (2) the post-election and pre-implementation period—November 2014February 2015, and (3) the post-implementation period—March 2015-December 2015. It is important to note here that while the City of Berkeley implemented the SSB tax in March 2015, campus retailers did not start receiving the SSB tax on invoices from their vendors until August 2015, and did not pass the tax on to consumers in any form until August 2016, which is after our sample period.11 The tax is set up such that it is paid by the distributor, 7

Soda Wars

who may or may not pass the cost on to their consumers. Falbe et al. (2015) and Cawley and Frisvold (2015) both find incomplete pass through of Berkeley’s SSB tax on to consumers three months after the policy implementation, with roughly half of the tax passed on. In our setting, campus retail food and beverage prices are sticky and only change once per year, occurring during the summer months of June, July, or August. Since the tax was not passed through to consumers on campus during our sample period, this paper examines how the soda tax campaign, election, and increase in prices off-campus affect the sales of soda on-campus. 3

Empirical Strategy

Our approach has two parts. In the first, we use a difference-in-differences (DID) strategy to measure the change in soda consumption due to the soda tax campaign and election, as well as the implied price elasticity of beverages given actual purchases. In a complementary second part, we estimate an event study model to test the identifying assumption of the DID model, namely that soda sales would have continued on the same trend as the other products had it not have been for the campaign and election. 3.1 Difference-in-Difference Model The DID model compares purchase behavior for soda (i.e., the treated category) with purchase behavior for the eight other beverage and candy product groups (i.e., the control categories), in the pre-campaign, campaign, and post-campaign periods. In total, products are categorized into nine groups: 1) soda, 2) water, 3) juice, 4) energy drinks, 5) milk, 6) coffee, 7) tea, 8) diet drinks, and 9) candy. Using data from November 2013 through December 2015, we compare the pre-campaign period (Nov 2013-Jun 2014) to three subsequent periods: 1) Pre-Election Campaign (Jul 2014-Oct 2014), 2) Post-Election/Pre-Implementation (Nov 2014-Feb 2015), and 3) Post-Implementation in the city of Berkeley (Mar 2015-December 2015). By comparing the soda purchase behavior in the pre-period to each of these postcampaign periods, we attempt to distinguish the effects of the campaign from the effects of the election and the effects of prices increasing off-campus.

8

Soda Wars

The DID model specification is as follows: Qigm = β1 Priceigm + β2 Sodaig + β3 Campaignm + β4 PostElectionm + β5 PostPolicym + β6 (Soda*Campaign)igm + β7 (Soda*PostElection)igm (1)

+ β8 (Soda*PostPolicy)igm + igm

where Qigm is the quantity sold of product i in product group g and month m, P riceigm is the unit price of product i in product group g and month m, Sodaig is an indicator for product i being in the treated soda product group. Three time indicators—Campaignm , P ostElectionm , and P ostP olicym —define four time periods. The pre-campaign period is when all four are zero, the pre-election campaign period is when Campaignm = 1 and the others equal zero, the post-election/pre-policy period is when P ostElectionm = 1 and the others equal zero, and the post-policy period in the city of Berkeley but not on campus is when P ostP olicym = 1 and the others equal zero. For shorthand, we refer to these periods as: pre-campaign, campaign, post-election, and post-policy. The coefficients of interest are those on the interactions of Sodaig and the campaign periods. The coefficient for Soda ∗ Campaignigm is the effect of the campaign on soda sales relative to the control product categories, the coefficient on Soda ∗ P ostElectionigm is the effect of the election, and the coefficient on Soda ∗ P ostP olicyigm is the effect of the SSB tax change in the city of Berkeley. Although useful for examining the average treatment effect of the tax change on the treated soda categories, specification (1) does not control for potentially important covariates that, if omitted, could lead to a biased estimate of the treatment effect. For example, consumer demand may differ by product group, as well as over time, and over seasons. To reduce the likelihood that the estimated treatment effects are biased, we next include fixed effects for the nine product groups αg and for the month-of-sample αm in the following regression: Qigm = β1 Priceigm + β6 (Soda*Campaign)igm + β7 (Soda*PostElection)igm (2)

+ β8 (Soda*PostPolicy)igm + αg + αm + igm . 9

Soda Wars

We also estimate specifications of equation (2) with product fixed effects αi instead of product group fixed effects and with logged quantities and prices (i.e., lnQigm and lnP riceigm ) in order to directly estimate price elasticities. 3.2 Event Study Model The identifying assumption of the DID model is that of parallel trends, where soda sales would have continued on the same trend as the other products had it not have been for the campaign and election. To directly test this assumption, we complement the DID model with the following event study model:

(3)

Qigm = β1 Priceigm +

24 X

βm (Sodaig ∗ Dm ) + αg + αm + igm

m=1

where (Sodaig ∗ Dm ) is a set of dummies equaling one for soda products in month-of-sample m. We use 24 months (from January 2014-December 2015), and the tenth month-of-sample (m = 10, which is October 2014) as the omitted dummy. Thus, equation (3) is the same as equation (2), except instead of splitting the sample periods into four periods, we compare soda sales to the untreated products in every month of the sample. The βm vector are the parameters of interest. We will plot the βm coefficients over time to trace out the adjustment path from before the campaign to the election and policy implementation. Importantly, if the soda tax campaign is unassociated with underlying trends, there should be no trend in the βm in the pre-campaign period. 4

Results

4.1 Average Treatment Effect of the Soda Tax Campaign on Soda Purchases We present the results from the reduced form specification of equations (1) and (2) in Table 1, where the dependent variable in all columns is the quantity sold of product i in product group g and month-of-sample m. The columns in Table 1 are organized as follows: column (1) reports the results from the specification of equation (1); column (2) adds product group fixed effects; column (3) adds month-of-sample fixed effects; column (4) replaced product 10

Soda Wars

group fixed effects with product fixed effects; and column (5) replicates column (4) with quantities and prices in logs. The parameters of interest are the three interactions of the soda indicator and the campaign period indicators. The price coefficient is statistically different from zero and negative in almost all specifications. Thus as one would expect, as price goes up demand goes down. Given the log-log specification in column (5), we can interpret the coefficient on Price per Item in column (5) as a price elasticity. This coefficient equals -0.343, which would mean that demand for beverage and candy categories is inelastic; however, the estimate is not statistically different from zero. While not precisely estimated, our elasticity coefficient is at the lower bound of what the literature finds. Andreyeva et al. (2010) survey more than 180 studies and find price elasticities of sodas ranging from -0.33 to -1.24 and price elasticities of foods between -0.27 and -0.81. Given our inelastic estimate, we do not predict there to be a significant demand drop once the tax is passed through into university retail prices. In all five columns of table 1, the coefficients on the interactions Soda ∗ P ostElection and Soda ∗ P ostP olicy are negative, large in magnitude, and statistically different from zero, suggesting that soda sales decreased more than the sales of the other product groups during the post-election and post-policy periods. While the coefficients on Soda ∗ Campaign are all negative, they are much smaller in magnitude and are not statistically different from zero, except in column (4). This suggests the campaign period did not alter soda sales relative to the control product groups. In summary, even though no price change occurred on-campus during the post-election period, we find consumers purchase less soda relative to the other product groups. When the SSB tax is implemented off-campus (i.e., during post-policy period), soda sales on-campus remain below those of the other product groups. Next, we show heterogeneity in campaign effects on soda sales when compared to individual control product groups. 4.2 Heterogeneity in Treatment Effect of Soda Tax Campaign on Soda Purchases, by Comparison Product Group To further understand whether and how consumers switch away from soda, we estimate equation (2) comparing soda to each of the other product groups individually. Table 2 presents 11

Soda Wars

the results. Each column includes the products in the soda category and the products in one other category. In Table 2 we find that the soda tax campaign leads to fewer purchases of soda compared to all other categories, except for milk (column 4). For the campaign period, soda sales decrease significantly relative to juice (column 2), coffee (column 5), tea (column 6) and diet drinks (column 7). For the post-election and post-policy period, the quantity of soda sold declines compared to the majority of other products (i.e., the coefficients in the rows Soda ∗ P ostElection and Soda ∗ P ostP olicy are negative and statistically different from zero in almost every column). In terms of magnitude, the coefficients on products in the rows Soda ∗ P ostElection and Soda ∗ P ostP olicy continue to be larger compared to the coefficients on products in the Soda ∗ Campaign row. 4.3 Event Study Results Given the interesting patterns we find in the DID results, with the treatment effects being largest in the post-election and post-policy periods, we next explore the parallel trends assumption and the dynamics of the treatment effects over time using our event study model. Figure 3 plots the estimates we obtain from equation (3), with the βm plotted in black and the 95 percent confidence intervals plotted in gray. Vertical red lines separate the sample into the four treatment periods. The omitted dummy is D10 , which corresponds to October 2014. In the periods before the election, we find roughly parallel trends, with the majority of the βm not statistically different from zero. However, after the election in November 2014, the βm estimates begin to decline, indicating that soda sales dropped relative to the control product groups. By the time the city implements the policy in March 2015 (in the city of Berkeley but not yet on campus), the βm are no longer declining, but are at a constant level significantly lower than the pre-campaign period. These event study results suggest that the election period drove the decline in soda sales on-campus, and that consumption of soda relative to the other product categories remained depressed after the policy implementation in the city of Berkeley.

12

Soda Wars

5

Conclusion

This paper uses a detailed, product-level dataset of quantities sold over time to measure the quantity response to a soda tax campaign. We estimate a significant drop in soda consumption in response to the election outcome relative to other beverage and candy products. These findings have implications for measuring the actual policy effect, given that the election win of the soda tax measure itself altered consumption behavior even before a price change occurs due to the tax. We find a 30% drop in soda sales relative to other product groups during the post-election period. In a related and contemporaneous study using survey data, Falbe et al. (2016) find an average 21% drop in SSB quantity sold. However, because the surveys were completed only before the election and after the tax implementation, Falbe et al. (2016) are unable to distinguish whether this effect was from the campaign and election or from the tax itself. Our results show that soda sales fell on-campus after the soda tax election yet before prices changed due to the tax. This suggests that comparing pre-campaign to post-implementation consumption may confound a tax effect with an information effect. This has important implications for external validity. If the Berkeley SSB tax is replicated elsewhere without a proceeding campaign war and affirmative election outcome, its effects on consumption may differ substantially. In other words, given the amount of money spent in the Berkeley campaign on each side of the battle, it is important to understand how much behavioral change was due to the election and how much was due to the tax itself. The policy implication is that information dissemination is an important avenue to move demand. Moreover our findings are consistent with findings in other settings. For instance, in the context of standards in egg production, Lusk (2010) finds that the publicity surrounding a vote to pass a proposition pertaining to animal welfare in itself had a significant impact on consumer behavior, beyond the effect the policy had once implemented. We provide an analysis of actual consumer responses, allowing direct estimation of revealed preferences for beverage product characteristics captured by price, product type, and time varying events in the form of the soda campaign and election. In so doing, we provide policymakers with an average price elasticity estimate for beverages in a university setting.

13

Soda Wars

This estimate is an important first step towards understanding the efficacy of tax-induced price changes as well as an understanding of consumer preferences for soda and substitute beverages, when subject to negative advertising campaigns for soda. Given the elasticity estimates in our setting, we predict that quantity will drop in an inelastic fashion when the tax is passed through into consumer prices. Therefore significant revenue can be raised from taxing soda without causing large deadweight losses. However, the low price elasticity means that a very high levy would be necessary to significantly change the behavior of buying soda in order to reduce sugar consumption and impact health. In future work, we will use post-tax implementation data from the university to estimate the actual quantity responses when the tax is indeed passed through into consumer prices.12 We will also utilize scanner data on product purchases from representative consumers within the city of Berkeley as a whole and compare the tax effects on quantity consumed by Berkeley consumers versus nearby, control city consumers, taking the analysis beyond the university context of the present paper. Notes 1

There is suggestive evidence that in the first month of the tax, tax revenues increased by $116,000, which

is consistent with demand having not responded in an elastic fashion to the one-cent-per-ounce increase in price (“1st Month of Berkeley ‘Soda Tax’ Sees $116,000 in Revenue.” The Daily Californian. May 19, 2015. Online. [accessed May 21, 2016 ]). 2

“Around $3.4M spent on Berkeley soda tax campaign.” Berkeleyside. Feb. 5, 2015. Online. [accessed

May 21, 2016 ]. 3

One reason to tax distributors instead of customers is to make the price change more salient. There is a

growing literature providing empirical evidence that consumers have an attenuated response to non-salient costs. With a labeling experiment, Chetty et al. (2009) find that the sales of taxable products at a grocery store are reduced when their tax-inclusive price is displayed in addition to the tax-exclusive price. Thus by taxing distributors of SSBs, if the tax is passed on to consumers, this will effect the displayed price and be more salient than a tax at the point-of-sale. 4

Furthermore, identifying the effects of soda tax media coverage on economic outcomes adds to existing

research in this area that has focused on the impact of media expansion and media bias on political attitudes and outcomes (Stroemberg 2004; Gentzkow and Shapiro 2010; DellaVigna and Kaplan 2007). 5

“How One of the Most Obese Countries on Earth Took on the Soda Giants,” The Guardian. 3 Nov.

14

Soda Wars

2015. Online. [accessed 15 Oct. 2016]. 6

“Berkeley Officials Outspent but Optimistic in Battle Over Soda Tax.” The New York Times. Oct. 7,

2014. Online. [accessed May 21, 2016 ]. 7

“Around $3.4M Spent on Berkeley Soda Tax Campaign.” Berkeleyside. Feb. 5, 2015. Online. [accessed

May 21, 2016 ]. 8

While we have data from January 2013 to October 2013, these data are at a more aggregate product

level and do not match the product level data we use in our sample. 9

Diet drinks are dropped from the graph as visually this product group was indistinguishable from zero.

10

As a more rigorous test of parallel trends, we regress quantity on a time trend for the treatment and

control products separately. We find that the point estimates of the trend in treatment and control products are not statistically different from each other. Furthermore, the time series correlation of the sample averages of soda and the control is high, suggesting that the treatment and control products share broadly similar time varying patterns in the pre-campaign period. 11

This was reported to us by campus retail staff and confirmed in the data.

12

The soda tax was implemented on campus in August 2016, more than a year after the election. Currently,

the sample period of our data ends before this date.

References Aguilar, A., E. Gutierrez, and E. Seira (2016). Taxing to Reduce Obesity. Working Paper . Andreyeva, T., M. W. Long, and K. D. Brownell (2010). The Impact of Food Prices on Consumption: A Systematic Review of Research on the Price Elasticity of Demand for Food. American journal of public health 100 (2), 216–222. Brown, D. J. and L. F. Schrader (1990). Cholesterol Information and Shell Egg Consumption. American Journal of Agricultural Economics 72, 548–55. Brownell, K. D. and R. Frieden, Thomas (2009). Ounces of Prevention—The Public Policy Case for Taxes on Sugared Beverages. The New England Journal of Medicine 206 (18), 1805–1808. Cawley, J. and D. Frisvold (2015). The Incidence of Taxes on Sugar-Sweetened Beverages: The Case of Berkeley, California. Technical report, National Bureau of Economic Research.

15

Soda Wars

Chavas, J. P. (1983). Structural Change in the Demand for Meat. American Journal of Agricultural Economics 65, 148–53. Chetty, R., A. Looney, and K. Kroft (2009). Salience and Taxation: Theory and Evidence. American Economic Review 99 (4), 1145–1177. DellaVigna, S. and E. Kaplan (2007). The Fox News Effect: Media Bias and Voting. Quarterly Journal of Economics 122, 1187–234. Falbe, J., N. Rojas, A. H. Grummon, and K. A. Madsen (2015). Higher Retail Prices of Sugar-Sweetened Beverages 3 Months After Implementation of an Excise Tax in Berkeley, California. American journal of public health 105 (11), 2194–2201. Falbe, J., H. R. Thompson, C. M. Becker, N. Rojas, C. E. McCulloch, and K. A. Madsen (2016). Impact of the Berkeley Excise Tax on Sugar-Sweetened Beverage Consumption. American Journal of Public Health 106 (10), 1865–1871. Fernandes, M. (2008). The Effect of Soft Drink Availability in Elementary Schools on Consumption. Journal of American Dietetic Association 108 (9), 1445–52. Fletcher, J. M., D. E. Frisvold, and N. Tefft (2010). The Effects of Soft Drink Taxes on Child and Adolescent Consumption and Weight Outcomes. Journal of Public Economics 94 (11), 967–974. Gentzkow, M. and J. M. Shapiro (2010). What Drives Media Slant? Evidence from U.S. Daily Newspapers. Econometrics 78 (1), 35–71. Gregg, E. W., X. Zhuo, Y. J. Cheng, A. L. Albright, K. V. Narayan, and T. J. Thompson (2014). Trends in Lifetime Risk and Years of Life Lost Due to Diabetes in the USA, 1985–2011: A Modelling Study. The Lancet Diabetes & Endocrinology 2 (11), 867–874. Huang, R. and K. Kiesel (2012). Does Limiting Access to Soft Drinks in Schools Result in Compensation at Home? European Review of Agricultural Economics 39 (5), 797–820. Huberman, G. and T. Regev (2001). Contagious Speculation and a Cure for Cancer: A Nonevent that Made Stock Prices Soar. Journal of Finance 56 (1), 387–96. 16

Soda Wars

James, J., P. Thomas, D. Cavan, and D. Kerr (2004). Preventing Childhood Obesity by Consumption of Carbonated Drinks: Cluster Randomized Controlled Trial. British Medical Journal 328 (7450), 1236. Lusk, J. (2010). The Effect of Proposition 2 on the Demand for Eggs in California. Journal of Agricultural & Food Industrial Organization 8 (1), 1–18. Roberto, C., D. Wong, A. Musicus, and D. Hammond (2016). The Influence of SugarSweetened Beverage Health Warning Labels on Parents’ Choices. Pediatrics 137 (2). Schlenker, W. and S. B. Villas-Boas (2009). Consumer and Market Responses to Mad Cow Disease. American Journal of Agricultural Economics 91 (4), 1140–52. Smith, M. E., E. O. V. Ravenswaay, and S. R. Thompson (1988). Sales Loss Determination in Food Contamination Incidents: An Application to Milk Bans in Hawaii. American Journal of Agricultural Economics 70 (3), 513–20. Stroemberg, D. (2004). Radio’s Impact on Public Spending. Quarterly Journal of Economics 119 (1), 189–221. Van Ravenswaay, E. O. and J. P. Hoehn (1991). The Impact of Health Risk Information on Food Demand: A Case Study of Alar and Apples. In J. Caswell (Ed.), Economics of Food Safety, pp. 356. New York, NY: Elsevier Bioscience TP373.5.E26. Yen, S. T. and H. H. Jensen (1996). Cholesterol Information and Egg Consumption in the US: A Nonnormal and Heteroscedastic Double-Hurdle Model. European Review of Agricultural Economics 23 (3), 343–56.

17

Soda Wars

Table 1: Difference-in-Difference: Effect of Soda Tax Campaign on Campus Retail Soda Sales

Price per Item

Soda=1

(1) Qty Sold -8.981∗∗∗ (1.844)

(2) Qty Sold -4.643∗∗∗ (0.957)

(3) Qty Sold -4.453∗∗∗ (1.015)

(4) Qty Sold -141.142∗∗∗ (51.070)

(5) Log Qty Sold

108.016 (72.779)

Campaign=1

-103.797∗∗∗ (37.693)

-98.564∗∗∗ (28.899)

Post-Election=1

-103.922∗∗∗ (36.164)

-91.348∗∗∗ (28.054)

Post-Policy=1

-114.158∗∗∗ (35.554)

-86.879∗∗∗ (28.775)

Soda=1 × Campaign=1

-16.410 (91.300)

-22.133 (88.128)

-25.979 (35.852)

-70.080∗ (41.196)

-0.079 (0.149)

Soda=1 × Post-Election=1

-146.330∗ (79.335)

-159.601∗∗ (76.093)

-164.097∗∗∗ (42.529)

-272.074∗∗∗ (66.931)

-0.301∗ (0.176)

Soda=1 × Post-Policy=1

-177.501∗∗ (81.300)

-205.644∗∗∗ (78.693)

-204.859∗∗∗ (33.869)

-201.073∗∗ (81.300)

-0.452∗∗ (0.203)

219.673 5631 0.649

-0.341 (0.557) 3.952 5631 0.787

X X

X X

Price per Item (log) Mean of Dep. Variable Num of Obs. R squared Product Group FE Month-of-Sample FE Product FE

219.673 5631 0.009

219.673 5631 0.070 X

219.673 5631 0.103 X X

Clustered errors in parentheses. Clusters are at the product group by month-of-sample level. The outcome variable is the quantity of products sold per month. Products are categorized into ten groups: 1) Soda, 2) Water, 3) Juice, 4) Energy drink, 5) Milk, 6) Coffee, 7) Tea, 8) Diet drinks, and 9) Candy. ∗p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗p < 0.01

18

Table 2: Difference-in-Difference: Effect of Soda Tax Campaign on Campus Retail Soda Sales (Product Group Comparisons—levels) (1) Soda|Water -1538.736∗ (816.223)

(2) Soda|Juice -247.999∗∗ (99.465)

Soda=1 × Campaign=1

7.297 (132.122)

-129.780∗∗∗ (35.336)

Soda=1 × Post-Election=1

-699.934∗ (357.869)

Soda=1 × Post-Policy=1

-517.892 (537.550) 500.339 610 0.603 X X X

Price per Item

Mean of Dep. Variable Num of Obs. R squared Product Group FE Month-of-Sample FE Product FE

(3) (4) (5) Soda|Energy Soda|Milk Soda|Coffee -33.084 -262.659 -8.347 (44.477) (799.328) (18.927)

(6) Soda|Tea 279.634 (409.123)

(7) Soda|Diet -42.142 (379.374)

(8) Soda|Candy 465.901∗ (250.359)

-60.127 (47.171)

279.764∗∗∗ (73.606)

-114.163∗∗∗ (34.050)

-92.383∗∗∗ (27.348)

-105.943∗∗ (47.772)

-15.861 (68.575)

-319.256∗∗∗ (60.532)

-234.655∗∗∗ (65.477)

43.278 (108.858)

-316.617∗∗∗ (68.339)

-316.380∗∗∗ (65.463)

-278.181∗∗∗ (57.326)

-122.961∗∗ (57.230)

-240.695∗∗∗ (81.251) 217.461 2157 0.638 X X X

-183.212∗∗ (69.055) 182.392 1630 0.689 X X X

7.928 (105.614) 272.105 563 0.649 X X X

-282.297∗∗∗ (70.507) 225.589 777 0.619 X X X

-268.990∗∗∗ (69.896) 210.198 879 0.620 X X X

-190.940∗∗ (73.024) 180.788 689 0.594 X X X

-115.878∗ (62.819) 158.499 1028 0.623 X X X

19

Soda Wars

Clustered errors in parentheses. Clusters are at the product group by month-of-sample level. The outcome variable is the quantity of products sold per month (logged). Products are categorized into ten groups: 1) Soda, 2) Water, 3) Juice, 4) Energy drink, 5) Milk, 6) Coffee, 7) Tea, 8) Diet soda, and 9) Candy. Each column includes the products in Soda and the products from one of the other nine product groups. ∗p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗p < 0.01

Soda Wars Figure 1: Google Trends News Search Interest of “Soda Tax” in the San Francisco Bay Area Over Time

Relative Search Interest

120 100 80 60 40 20 0

Source: Google Trends. Online. [accessed May 22, 2016 ]. Note: Numbers on y-axis represent search interest relative to the highest point on the chart for the given region and time. A value of 100 is the peak popularity for the term. A value of 50 means that the term is half as popular. Likewise a score of 0 means the term was less than 1% as popular as the peak.

20

Soda Wars Figure 2: Pre-Soda Tax Campaign: Monthly Quantities Sold by Product Group

Quantity Sold

25000 20000 15000 10000 5000 0 Nov'13

Dec'13

Jan'14

Feb'14

Juice Energy Drink Water Soda

Mar'14

Apr'14

May'14

June'14

Milk Tea Coffee Candy

Note: Beverage and candy products are categorized into nine groups: 1) Soda, 2) Water, 3) Juice, 4) Energy drink, 5) Milk, 6) Coffee, 7) Tea, 8) Diet Drinks, and 9) Candy.

21

Soda Wars Figure 3: Event Study: Effect of Soda Tax Campaign on Campus Retail Soda Sales 600 Campaign Starts

400

Election, Soda Tax Passed

Tax Rollout Begins in City of Berkeley, Not on Campus

Quantity Sold

200

0

-200

-400

-600 Jan 14

Mar 14

May 14

Jul 14

Sep 14

Nov 14

Jan 15 x

Mar 15

May 15

Jul 15

Sep 15

Nov 15

Note: The figure displays the βm coefficient estimates from event study equation 3. The dependent variable is the quantity sold of product i in product group g and month-of-sample m. Upper and lower 95% confidence intervals are depicted in gray, estimated using standard errors clustered by product group by month-of-sample.

22

Soda Wars

A

Appendix

Figure A.1: YES on Measure D Advertisement

Source: Berkeleyside. Online. [accessed May 22, 2016 ].

23