Content uploaded by Marcel Bilger
Author content
All content in this area was uploaded by Marcel Bilger on Dec 28, 2017
Content may be subject to copyright.
Page 1 of 25
© 2013. This manuscript version is made available under the CC-BY-NC-ND 4.0 license
http://creativecommons.org/licenses/by-nc-nd/4.0/. The published version is located here:
https://doi.org/10.1016/j.jhealeco.2012.10.005.
Implications of a Sugar-Sweetened Beverage (SSB) Tax When Substitutions to
Non-Beverage Items Are Considered
Eric A. Finkelsteina,*, Chen Zhenb, Marcel Bilgera, James Nonnemakerb, Assad M. Farooquia,
Jessica E. Toddc
aDuke-NUS Graduate Medical School, Singapore
bRTI International
c Economic Research Service, U.S. Department of Agriculture
ABSTRACT
Using the 2006 Homescan panel, we estimate the changes in energy, fat and sodium purchases
resulting from a tax that increases the price of Sugar-Sweetened Beverages (SSBs) by 20% and
the effect of such a tax on body weight. In addition to substitutions that may arise with other
beverages, we account for substitutions between SSBs and 12 major food categories. Our main
findings are that the tax would result in a decrease in store-bought energy of 24.3 kcal per day
per person, which would translate into an average weight loss of 1.6 pounds during the first year
and a cumulated weight loss of 2.9 pounds in the long run. We do not find evidence of
substitution to sugary foods and show that complementary foods could contribute to decreasing
* Corresponding author: Eric Finkelstein, Health Services and Systems Research Program, Duke-NUS Graduate
Medical School Singapore, 8 College Road, Singapore 169857, (65) 6516 2338 (voice), (65) 6534 8632 (fax),
eric.finkelstein@duke-nus.edu.sg.
Page 2 of 25
energy purchases. Despite their significantly lower price elasticity, the tax has a similar effect on
calories for the largest purchasers of SSBs.
INTRODUCTION
Currenty, roughly 1 in 3 U.S. adults are obese and forecasts predict that obesity rates
could reach 42% by 2030 (Finkelstein et al., 2012). As a result, public health officials are
increasingly looking to identify strategies to contain rising rates of obesity. One strategy focuses
on reducing consumption of Sugar-Sweetened Beverages (Brownell et al., 2009; Paterson, 2008);
SSBs hereafter). Likely due to their low and declining prices relative to healthier food and
beverage items, SSB consumption has increased considerably over the past several decades,
(Brownell et al., 2009) (Brownell et al., 2009) such that today they account for roughly 7% of all
calories consumed (Finkelstein and Zuckerman, 2008). SSB consumption is associated with
increased caloric intake, weight gain, and obesity (Finkelstein and Zuckerman, 2008; Ludwig et
al., 2001; Malik et al., 2006; Vartanian et al., 2007) and there is evidence that a reduction in SSB
consumption could result in weight loss (Chen et al., 2009).
To reduce SSB consumption, a commentary in the British Medical Journal and another in
the New England Journal of Medicine recommended a tax of 1 cent per ounce on SSBs
(Brownell et al., 2009; Kamerow, 2010). Assuming the tax is passed through to consumers, it
would increase the cost of a 20-oz soft drink by roughly 20% and, based on estimates of the price
elasticity of demand, would result in an 8-10% reduction in SSB consumption. The commentary
posits that in New York State alone, where a similar tax was proposed and then defeated (New
York Post, 2009), it would have prevented 145,000 cases of adult obesity.
Whether these taxes would positively influence obesity rates as the commentary suggests
remains open to debate. Some, including prominent scholars (Becker, 2009; Posner, 2009)
Page 3 of 25
challenge the very ability of the tax to significantly decrease caloric intake as those consumers
who like sugar—or even crave it—would have no difficulty finding substitutes for SSBs. Their
beliefs are backed up by recent animal and human studies suggesting an addictive component to
sweet foods (e.g. Avena et al., 2012) and (see Gearhardt et al., 2009 for a review).
The implications of an SSB tax have been partially addressed by several recent studies
that estimated the effect of an SSB tax on the consumption of a wider range of beverages
(Dharmasena and Capps, 2011; Finkelstein et al., 2010; Lin et al., 2011; Smith et al., 2010; Zhen
et al., 2011). Yet, none of the above studies allowed for substitution from SSBs to non-beverage
products that may result from the tax
1
. This is a significant limitation as these studies do not fully
address the concern that individual taste or sugar cravings could result in substituting SSBs with
sugary, highly caloric foods.
Because SSBs are among the most energy dense beverages, any switching away from
SSBs to other beverages is almost guaranteed to reduce total beverage calories. The implications
of the tax are less predictable when switching to non-beverage items is considered. This results
because some food items, such as candy or cookies, are higher in calories per dollar than SSBs.
As a result, if someone switches from SSBs to these foods as a result of a tax, net calories could
actually increase. (Schroeter et al., 2008)’s theoretical model yields similarly counterintuitive
results when predicting the effects of a tax on away from home foods. These results stress that to
assess the net effect of an SSB (or any food) tax, one needs to compute the change in calories
1
Note that another recent study by (Duffey et al., 2010) measures the price elasticity of soda consumption while
taking potential substitutions with whole milk, pizza and hamburgers into account. We argue that both the
number of beverages and of foods categories are not large enough to properly measure overall changes in energy
purchased. For instance, highly energetic beverages such as fruit juices and sports energy drinks are missing, and
no potential sugary foods substitutes such as candy bars are considered.
Page 4 of 25
resulting from the taxed product, as well as increases or decreases resulting from complementary
or substitute products, and in this case, not only from other beverages.
This paper aims to fill this gap by revisiting the extent to which targeted SSB taxes would
reduce caloric purchases and weight when considering substitution to other foods and beverages.
This includes, in addition to the beverages taken into account by (Finkelstein et al., 2010),
candies and cookies, which are both higher in calories per dollar spent than SSBs, and several
other commonly purchased foods that are potential substitutes or complements to SSBs. Using
these data, we estimate the changes not only in calories, but also in fat and sodium purchases
resulting from a tax that increases SSB prices by 20%. These extensions are important as a
healthy diet requires limiting not only caloric, but also fat and sodium intake (U.S. Department
of Agriculture and U.S. Department of Health and Human Services, 2010). Because SSBs are
low in both, it is possible that an SSB tax could increase consumption of both fat and sodium,
thus reducing any health gains that may be achieved through reduced caloric intake.
Unlike prior studies, we estimate the effects of SSB taxes using an instrumental variables
technique which controls for potential price endogeneity. This is an important advancement as
the prices used in prior studies may not be truly exogenous to the purchaser (i.e. driven by
changes in both supply and demand), may be correlated with unobservable factors that influence
demand, or may suffer from significant measurement error. An instrumental variables approach
addresses these concerns. Finally, we estimate quantile regressions to explore the extent to which
SSB taxes differentially affect those who are the largest purchasers of calories from both SSBs
and all included food categories. The former explores whether the taxes work better among the
highest SSB consumers, presumably the intended target, whereas the latter explores the extent to
Page 5 of 25
which the tax may differentially affects those who purchase the most calories (from stores) and
therefore may be more likely to be obese.
Section 2 presents the data sources, variable definitions and descriptive statistics. Section
3 covers our econometric models, identification strategy, and energy-to-weight loss computation
methods. Section 4 presents our estimates for both aggregate and detailed changes in store-
bought nutrients caused by the SSB tax along with the weight loss predictions. Section 5
concludes.
DATA
This analysis relies on data from the 2006 Nielsen Homescan Panel (Nielsen, 2006), which is
now referred to as the National Consumer Panel. This panel includes a representative sample of
U.S. households that scan in their store-bought food and beverage purchases. Purchase records
are reported at the Universal Product Code (UPC) level. Each record contains data on dollars
paid and units purchased. We have limited the data to UPCs in one of nineteen Nielsen defined
categories that are likely to be purchased by the majority of US households, including many that
are likely to be complements or substitutes to SSBs. SSBs are broken down into three categories
(regular soda, fruit drinks and sports energy drinks), the consumption of which is recorded in
addition to four other beverage categories (fruit juices, skim and whole milk, and diet soda), and
to 12 food categories (candy, cookies, salty snacks, ice cream, yoghurt, ready-to-eat cereal,
French fries, pizza, frozen dinners, canned soups, canned fruits and canned vegetables).
Although the data do not include nutritional information, calorie, fat and sodium content for each
product was merged at the UPC level using data available from (Gladson, 2009) and from the
USDA National Nutrient Database (U.S. Department of Agriculture, 2010). The merged data
Page 6 of 25
was then aggregated at the household level for each quarter of 2006. The resulting sample
consists of 114,336 observations over 28,584 households. While aggregating, we computed the
total calories, fat and sodium purchased by each household during each quarter. These
household-level quarterly purchases were finally converted into a daily average and expressed at
the individual level by dividing by the number of individuals older than 6 years old residing
within each household.
Using these dollars paid and quantities purchased, we created a household-specific
superlative Fisher ideal price index for each of the 19 food and beverage categories. Such price
indices are preferable to unit values (i.e. category expenditure divided by quantity) as they
considerably reduce the bias due to within-category substitution (Cox and Wohlgenant, 1986;
Deaton, 1988). Finally, Nielsen also makes available various socio-demographic characteristics
of the individuals and households that took part in the 2006 Homescan panel.
Table 1 presents descriptive statistics on our store-bought food and beverage categories.
The first column shows the average percentage of households that buy at least one item of the
category during a quarter. It can be seen that, with respectively 63.9 and 59.9 percent of
households regularly purchasing, regular sodas and fruit drinks are often purchased in stores. On
the other hand, our third SSB category, sports energy drinks is purchased (from stores) by less
than a quarter of the households (24.2%). The next two columns of Table 1 show the average
energy bought daily by the households. Among SSBs, the most energy comes from regular sodas
(47 kcal per person daily), followed by fruit drinks (14.4), and finally sports energy drinks (3.5).
Overall, daily store-bought energy from SSBs amounts to 64.9 kcal per person. Other major
energy sources are candies with a mean purchase of 62.6 kcal per person daily, salty snacks
(60.9), ready-to-eat sweetened cereals (50.6), whole milk (45.7) and cookies (40.4). Table 1
Page 7 of 25
provides similar information for fat and sodium as well as the average energy that 1 USD could
buy in 2006. The table reveals that cookies are the most calorie dense product per dollar (855
kcal/$), closely followed by salty snacks (828 kcal/$). One dollar also buys 670 kcal of candy
(8th highest on the list of 19 products) while regular sodas come next with 626 kcal per dollar.
Insert Table 1 about here
Table 2 presents the socio-demographic characteristics of our sample. The table shows
unweighted and weighted population estimates from Homescan along with estimates from
Census. It can be seen that whites, older and more educated household heads are overrepresented
in Homescan whereas the lowest and highest income household categories are underrepresented.
Applying Homescan weights, as was done in our analysis, minimizes the differences between the
two samples.
Insert Table 2 about here
METHODS
There is a long history of economic research related to food consumption in what is referred to as
household production behavior (for early papers see Becker, 1965; Gorman, 1956, 1980;
Lancaster, 1966). Under this framework, individuals are assumed to obtain utility from some
underlying goods that cannot be directly purchased. In our context, these underlying goods are
energy, fat and sodium. Consumers do not purchase these directly, but rather produce them via
Page 8 of 25
the purchase of foods and beverages that contain various amounts of these nutrients. When the
price of foods and beverages is altered—for instance by a tax on SSBs—the demand for these
nutrients is also affected.
In this study, we estimate empirical models that directly explain the quantities of
nutrients demanded as a function of food and beverages prices and other control variables. We
analyze the demand for each nutrient (i.e. energy, fat and sodium) in total and separately for
each of our 19 food and beverage categories in efforts to disaggregate the net effects of price
changes. One difficulty with the disaggregate analyses is that not all households purchased every
item during any given quarter and therefore there are many zeros to be dealt with. To address this
issue, we use a two-part modeling strategy (Duan et al., 1983) where the expected quantity of
nutrient purchased q, conditional on vectors of food prices p and of control variables c is
expressed as the probability of having a purchase times the expected quantity of nutrient
purchased when there is purchase:
𝐸(𝑞|𝑝, 𝑐)= 𝑃(𝑞 > 0|𝑝, 𝑐)𝐸(𝑞|𝑞 > 0, 𝑝, 𝑐).
For the first part, i.e. the probability of purchasing nutrients, we apply a logit model. An
advantage of the two-part strategy is that it allows great flexibility when modeling the positive
outcomes in the second part, which is valuable considering that purchases, like medical
expenditures, are generally heavily right-skewed (Manning and Mullahy, 2001). In order to
produce estimations that are fairly robust to misspecification of the functional form, we apply a
model from the GLM family (Nelder and Wedderburn, 1972). Using the Box-Cox (Box and Cox,
1964) and Park (Manning and Mullahy, 2001; Park, 1966) tests we determined that the closest
link function and distribution family are the logarithm and the Gamma family respectively (see
Appendix A), which is a model that is also widely used in health economics for health care
Page 9 of 25
expenditure data (e.g. Blough et al., 1999). As noted above, we also run aggregate regressions
for the total amount of each nutrient purchased over all food and beverage categories to check
the robustness of the total change in nutrients purchased predicted by summing category-level
variations. These aggregate regressions do not involve zero outcomes and a simpler single-part
GLM model can be used.
To predict the effect of a SSB tax, the key covariates in all models are the logarithm of
the price (index) of the SSB categories. Our control variables are the logarithm of the prices of
all the other categories, household per capita income quartiles, the proportion of adults living in
the household as well as an indicator variable for each of the 52 Nielsen markets. Additional
control variables relate to the household head: age, educational degree, ethnicity, and a dummy
variable indicating whether the household head is a woman below 35 years old.
To account for potential endogeneity arising from measurement errors and omitted
variables, we estimate instrumental variable models with instruments for the 19 price
indices. The Homescan data provide the census tract number of each household’s residence.
Using Geographical Information System (GIS), we calculated the Euclidian distance from each
household to all others in the same Nielsen market. The instrument for each price index faced by
a household is the weighted average of the price indices of all other households in the same
Nielsen market and quarter excluding those living in the same census tract. The inverse of the
Euclidian distance is used as the weight. We excluded prices of households living in the same
tract from the construction of instruments to reduce simultaneity bias caused by common tract-
specific demand shocks. The idea of using prices of neighbors to instrument product prices has
been widely adopted in the industrial organization literature (e.g. Hausman et al., 1997; Nevo,
2003). Identification of demand parameters is achieved when measurement errors in neighbors’
Page 10 of 25
prices and neighbors’ idiosyncratic preference shocks are uncorrelated with those of the
household. This type of instrument has proven to be useful when a large number of prices, short
sample period, and specificity of the products make it impractical to obtain supply cost data to
serve as instruments.
To estimate the models using these instruments, we use a two stage residual inclusion
instrumental variable approach (Terza et al., 2008). In this approach, the logarithm of the
category-level prices are regressed on all the exogenous variables of the model and the above
instruments in the first stage, and the residuals of these regressions are included into the
regressions explaining the demand for nutrients in the second stage. The effect of a 20% price
increase on SSBs is then assessed by computing resulting incremental variations in quantities of
nutrients purchased based on the estimated coefficients and the counterfactual of SSB prices
raised by these levels. Estimates for changes in fat and sodium as a result of the tax are
conducted similarly.
We convert the estimated reductions in calories into weight loss by applying the ratios
presented in (Lin et al., 2011). Their approach makes more realistic assumptions on the
relationship between small calorie reductions and short and long term weight loss than the
widely used assumption that a reduction of 3,500 calories in any daily increment results in one
pound of weight loss (e.g. Whitney et al., 2002). The approach taken by (Lin et al., 2011) is quite
complex and notably depends on age and gender as well as on body mass and its composition in
terms of fat and lean tissue. However, they find that for small daily reductions in calories, the
3,500 calories per pound rule overestimates weight loss by 65% in year one. After ten years, they
estimate that the net effect of the calorie reduction is merely 1.15 times the first year estimate;
significantly smaller than the naïve estimate of multiplying the first year reductions times ten.
Page 11 of 25
Lastly, we apply quantile regression (see for instance Koenker and Hallock, 2001) to all
three nutrient models in order to assess the effect of the tax at the 25th, 50th and 75th percentiles
of purchases. The instrumental variable approach applied is also a two-stage residual inclusion
that is adapted to the specificities of quantile regression (Chernozhukov and Hansen, 2008). For
all models, statistical inference is performed by bootstrapping over 1,000 repetitions and
computing both nonparametric confidence intervals and standard errors. All statistical analyses
were performed using Stata/MP 12.0 for Windows (Stata Corporation).
RESULTS
The estimated coefficients of all our models explaining calorie purchases are displayed in
Appendices B (aggregate consumption) and C (category-level), while those for fat and sodium
are available upon request. Appendix A shows the weak instrument and endogeneity tests for all
the models that explain calorie purchases at category-level.
Figures 1A and 1B display the incremental changes in store-bought nutrients resulting
from a tax that increases SSB prices by 20% that are predicted by the exogenous and IV models
respectively. These figures present aggregates that have been computed by cumulating changes
predicted by the models explaining our 19 detailed food and beverages categories. It can first be
seen that SSB taxes do reduce daily store-bought energy by 21.2 and 13.2 kcal according to the
exogenous and IV models respectively. Interestingly, with respectively 2.9 and 1.0 kcal per day
increases, substitution to energy from other beverages is small and not statistically significant.
We find no evidence to support the (Becker, 2009) and (Posner, 2009) hypothesis of substitution
to energy from food. On the contrary, although not statistically significant, we estimate
Page 12 of 25
reductions of 3.2 and 11.2 kcal from store-bought food per day, suggesting that the opposite
could occur.
Overall, both the exogenous and IV models predict a similar reduction in energy from
beverages and foods of more than 20 kcal per day. As for fat, given that SSBs generally contain
none, the direct incremental effect of the tax is negligible in both models. However, there is a
slight indirect reduction of roughly 30 centigrams in both models due to the reduction in
purchases of select foods resulting from the tax. The exogenous and IV models tell slightly
different stories concerning sodium. Both models show a statistically significant reduction in
store-bought sodium (of 10 and 6.8 mg per day respectively), but the IV model suggests that the
reduction in sodium from SSBs is reduced by an increase in sodium from food, and not a result
of substitutions to other beverages as suggested by the exogenous model.
Insert Figure 1 about here
To provide a deeper understanding on what is driving the above results, Table 3 presents
our disaggregated effects. For both the exogenous and IV models, the table displays the
incremental changes in store-bought nutrients resulting from a tax that increases SSB prices by
20% for each of the 19 food and beverage categories. In what follows, we will focus our
attention on the IV results as they correct for potential endogeneity biases. To begin with, most
of the direct effect of the SSB tax on energy purchased comes from regular soda (-5.5 kcal per
day) and fruit drinks (-7.0). Very little impact is observed for sports energy drinks (-0.7) due to
their relatively lower market share. The only substitution towards energy from other beverages
that is statistically significant comes from fruit juices (+2.5), which is not surprising due to their
Page 13 of 25
high natural sugar content. Interestingly, our results only reveal one food category that is a
(statistically significant) substitute for SSBs, albeit with a very low cross price elasticity: canned
soup. An increase in canned soup purchases as a result of the tax generated an increase in energy
purchases of 1.9 kcal per day. The taxation of SSBs does not lead to any changes in energy
purchased from cookies and candies. Moreover, we find that both salty snacks and ice cream
(another sugary substitute) are economic complements with SSBs. The 20% price increase
caused by the SSB tax generates decreases in daily energy purchases by 6.3 and 4.8 kcal for
these categories, respectively. Due to their high fat content, salty snacks and ice cream are also
the two categories that drive the reductions in overall fat purchases by 21.0 and 22.6 centigrams
per day, respectively. The tax also generates decreases in store-bought sodium from regular
sodas (-1.6 mg per day), fruit drinks (-3.9), and diet sodas (-2.4). Two food categories, candy (-
1.9) and ice cream (-1.6) also contribute to this reduction. However, an increase in consumption
of sodium from canned soups (14.7)—a highly salted food category—is sufficient to cancel out
all these modest decreases in sodium purchases
2
.
Insert Table 3 about here
Table 4 presents the incremental changes and arc-elasticities resulting from a tax that
increases SSB prices by 20% for 1) SSB nutrient purchases and 2) total nutrient purchases from
all included food and beverage categories, predicted with a single-part model as well as quantile
regressions. Only the exogenous models are presented as the IV quantile technique
2
On average, 1 oz of canned soup contains 172.4 mg of sodium. Thus the increase in sodium purchased corresponds
to an increase in quantity of only 0.085 oz per day.
Page 14 of 25
(Chernozhukov and Hansen, 2008) did not prove to be efficient enough and led to unstable
results (available upon request). The exogenous model yields a direct elasticity of -1.32 for the
energy purchased through SSBs. However, when using the IV model, this estimate drops to -
0.90. The quantile regression reveals that, as might be expected, low consumers of SSBs would
be far more sensitive to the tax (with a price-elasticity of -4.47 at percentile 25) than heavy
consumers (-1.40 at percentile 75). However, because the lower elasticity for larger purchasers is
applied to greater average purchases, the incremental effects across income categories increase
in absolute terms. A tax would thus be more effective at reducing the SSB purchases of heavy
consumers despite their lower elasticity. Across all foods and beverages, the arc-elasticities
reveal that those who purchase more calories from stores, who may be more likely to be obese,
overall also have a lower price-elasticity (range is from -0.38 at percentile 25 to a mere -0.12 at
percentile 75). However, for the same reason as above, the incremental effects across income
categories are more closely aligned (range is from -23.0 kcal per day for the median to -17.1 for
percentile 75). This suggests that all consumers—including those who may be at the greatest risk
of being obese—would benefit (in terms of weight loss) from an SSB tax.
Insert Table 4 about here
Finally, in order to get a sense of the magnitude of the effect caused by the SSB tax, it might be
useful to convert our predicted 24.3 store-bought kcal per day reduction into potential weight
loss. According to the dynamic method recently proposed by (Lin et al., 2011), predicted weight
loss in year one would be 1.6 pounds and the net effect, ceteris paribus, is 2.9 pounds after 10
years (as opposed to the naïve estimate of 25 pounds using the 3,500 calories per pound method).
Page 15 of 25
Note that these weight losses are not directly comparable with those computed by Lin et al, 2011
as the authors account for away-from-home consumption.
CONCLUSION
In this paper, we estimate the changes in energy, fat and sodium purchases resulting from
a tax that increases SSB prices by 20% as well as the effect of such a tax on body weight.
Although other papers have explored the direct and indirect effects of such a tax on beverages,
the contribution of this paper is to account for potential substitutions/complementarity between
SSBs and 12 food categories, in addition to other beverages. We specifically included cookies
and candies as these categories are greater in calories per dollar and hypothesized to be
substitutes for SSBs due to their high sugar content. As such, we allow for a potential increase in
calories that may result from an SSB tax. Our estimates are generated using a 2 stage residual
inclusion IV method to correct for the potential endogeneity of the price variables. We also run
quantile regressions to measure the impact of the SSB tax on high SSB or overall calorie
purchasers.
Our main findings are that a 20% price increase on SSBs would result in a decrease in
energy purchased in stores of 24.3 kcal per day per household member across the 19 included
food and beverage categories. This equates to a roughly 4.7% reduction. Following the recent
methodology proposed by (Lin et al., 2011), this would translate into an average weight loss of
1.6 pounds during the first year of implementation and a cumulated weight loss of 2.9 pounds
over ten years. The IV results reveal that substitution to other beverages was limited and only
involved fruit juices. Furthermore, we did not find any evidence of substitution to sugary foods
and even found a decrease in calories from ice cream and salty snacks. These decreases could
Page 16 of 25
amount to almost half of the total decrease in energy purchased resulting from the SSB tax. In
addition, due to the high fat content of these two complementary foods, the SSB tax also leads to
a small reduction in the consumption thereof. These estimates suggest the tax would have no
effect on total sodium purchased. Finally, we found the tax would reduce the energy purchased
by both heavy SSB and total calorie purchasers despite their lower price-elasticities.
Because our analysis includes food categories, the calorie and weight change estimates
are not directly comparable to those from prior studies. However, the own calorie elasticity
estimates for SSBs are comparable. Our exogenous model yields a direct elasticity of -1.32 for
the energy purchased through SSBs. This estimates falls into the middle of the range of prior
studies: (Finkelstein et al., 2010), -0.87, (Lin et al., 2011), -0.95 and -1.29 for lower and higher
income respectively, (Zhen et al., 2011), -1.06 and -1.54, and (Dharmasena and Capps, 2011), -
2.26. However, when using the IV model, our price-elasticity drops to -0.90 which is more in
line with earlier studies (see Andreyeva et al., 2010 for a review). Note that, with the exception
of (Finkelstein et al., 2010), the above studies also found very little substitution to other
beverages. The inclusion of food categories and other covariates, use of a different functional
form, or the application of monthly (as opposed to quarterly) prices explain the inconsistency
with the study by (Finkelstein et al., 2010).
As for data limitations in this study, it should be mentioned that the data are self-reported,
which may lead to underreporting. Second, the data are limited to store-bought food and
beverage purchases for 19 food categories, amounting to 518 kcal per day per individual on
average. Extending the model to include non-store purchases and additional food categories,
including random weight foods (i.e., those without barcodes) that are difficult to capture using
the current methodology, would generate more robust conclusions of the net effect of a 20%
Page 17 of 25
price increase on SSBs and should be an area of future research. Because store-bought purchases
represent a diminishing fraction of total and SSB calories purchased, the effect of a 20% price
increase on SSBs is likely to be larger than the estimates presented in the analysis. For example,
the Nielsen data reveal that 64.9 kcal per day of SSBs are purchased from stores whereas
(Popkin, 2010) reports overall daily SSB purchases, that is including food-away-from-home, in
excess of twice this amount.
An important question not addressed in this paper concerns the size of a tax that would
generate a 20% price increase on SSBs. Under perfect competition, taxes undershift in the sense
that a 20% tax would result in a smaller price increase, unless demand is perfectly inelastic,
which the above estimates reveal is not the case for SSBs, or supply is perfectly elastic. In these
instances, the tax is fully passed along to consumers (Musgrave, 1959). However, under
imperfect competition, economic theory predicts that taxes can also increase prices by more than
their nominal value---or overshift---which has been demonstrated for the monopoly (Cournot,
1960; Stern, 1987) and various types of oligopoly (see for instance Anderson et al., 2001;
Besley, 1989; Delipalla and Keen, 1992; Seade, 1985). The above theoretical literature reveals
that milder conditions are required for excise taxes to overshift compared with ad valorem taxes.
Overshifting of SSBs taxes is a possibility given that the soft-drink and related retail
industries are highly concentrated. However, to date, empirical evidence on the effect of select
SSB taxes on retail prices is scarce. In the US, (Besley and Rosen, 1999) find that sales taxes on
sodas that were implemented during the 1982-1990 period overshifted by as much as 29%. In
Denmark, (Bergman and Hansen, 2012) find evidence of overshifting during the 1998 and 2001
excise tax hikes on soft drinks there. Finally, (Bonnet and Réquillart, 2012) apply a structural
econometric model that takes into account the strategic price response to taxation of both
Page 18 of 25
manufacturers and retailers. Using French home-scan data they predict that an excise tax on soft
drinks would overshift by 7% to 33% depending on the brand. On the other hand, an equivalent
ad valorem tax would undershift by 10% to 40%. Empirical evidence on other sin taxes is
consistent with the overshifting of excise taxes in concentrated markets. Excise taxes on alcohol
(Kenkel, 2005; Young and Bielinska-Kwapisz, 2002) and tobacco (Delipalla and O’Donnell,
2001; Hanson and Sullivan, 2009; Harris, 1987) have been found to overshift. Also, studying the
incidence of taxes on saturated fat, (Griffith et al., 2010) find that excise taxes overshift while ad
valorem taxes undershift. Most sin taxes are implemented as excise taxes, perhaps because they
are more likely to raise retail prices, but also because they avoid substitutions from more to less
expensive items. The extent of under- or overshifting of SSB taxes in the US market will likely
only be known with certainty if the taxes are implemented and their effect on retail prices can be
directly assessed.
From a methodological perspective, unlike most other recent studies that account for
substitutions with other beverages (Dharmasena and Capps, 2011; Lin et al., 2011; Smith et al.,
2010; Zhen et al., 2011), we do not perform our analysis in the setting of a utility-theoretic
demand system. Consequently, this analysis does not account for the income effect resulting
from an SSB tax. However, this effect is likely to be limited considering the low budget share of
SSBs. On the other hand, our approach offers several advantages. The first is that our models
directly explain the variables of interest. These are the quantities of nutrients purchased which
we have matched in great detail at the product level. Another advantage of our approach is that
by estimating each food and beverage category separately without imposing utility-theoretic
restrictions such as homogeneity and symmetry, we avoid possible spillover of specification
error in one equation to other equations. We exploit this extra flexibility by using a two-part
Page 19 of 25
model that allows the participation decision and quantity decision to be different processes. Note
that the two-part specification straightforwardly deals with zero purchases which is a complex
issue in the context of a demand system. Notwithstanding this, we are currently in the process of
extending this research to a demand system specifically to test for differential effects by income
strata (Zhen et al., 2011). In this context, a demand system would be more appropriate as it
allows for income effects and welfare analysis.
The principal finding of this research is that an increase of 20% in the price of SSBs
would induce a cumulative per capita weight loss of 2.9 pounds, on average, after ten years,
ceteris paribus. If the revenue raised from these taxes were used to further fund obesity
prevention efforts, the weight losses could be even greater. However, any benefits of such a tax
must be weighed against the potential costs. These taxes are almost certainly regressive as lower
income households spend a larger share of their food budget on SSBs (Finkelstein et al., 2010).
An excise tax would likely be more regressive than a sales tax of equal magnitude because lower
income households tend to purchase lower priced beverages (same cite).
As noted in the introduction, SSB taxes are increasingly being considered in efforts to
raise revenue and/or to positively influence weight outcomes. As such, the implications of these
taxes are of economic interest. This is despite the fact that many economists have called into
question the very justification for such a tax. Although it may have beneficial effects on weight
and health, it is difficult to justify an SSB tax using the classical Pigovian approach to taxation,
as the traditional rationale for market failures appears not to apply (Bhattacharya and Sood,
2011). However, there still may be grounds for such a tax based on behavioral economic
principals.
Page 20 of 25
There is overwhelming evidence showing that people’s food consumption behavior can
be influenced by very subtle cues, thus calling into question people’s ability to make utility
maximizing decisions when it comes to what and how much to consume. For example,
consumers’ judgment of optimal portion size is influenced by the size of the container it comes
in. Larger containers generate increased consumption by altering norms for what is perceived as
an appropriate portion size (Wansink and van Ittersum, 2007). The average size of SSB
containers has increased over the past few decades as suppliers increase sizes in efforts to take
advantage of decreasing marginal costs (Marchiori et al., 2012). Although SSBs are not the only
product whose container size has increased, SSBs and other highly sweetened products may be
more appropriate targets for government intervention because, for genetic reasons, consumers
are predisposed to have a visceral response to these foods and are thus more easily tempted to
overconsume (Gilhooly et al., 2007). For these foods, the capacity for self-control is most limited
and the market has no incentive to correct for potential overconsumption, and in fact, encourages
it through aggressive marketing campaigns in efforts to increase profits. Advertisements for
highly processed foods make up the second largest share of all advertising revenue (Story and
French, 2004). In the behavioural economics literature, lack of self-control is seen as a form of
time inconsistency and has been modeled using hyperbolic discounting (Laibson, 1997; Laibson,
1994). Using such models, (O'Donoghue and Rabin, 2006) have shown that taxing sin goods can
be welfare enhancing when individuals have heterogeneous tastes and where some of them have
time-inconsistent preferences.
The above discussion, combined with the high degree of regret for many consumers, as
evidenced by the multi-billion dollar weight loss industry, suggests that government intervention
to decrease consumption on SSBs and other foods with a large degree of added sugars and added
Page 21 of 25
fats may be justified on economic grounds, at least from a behavioral perspective. As shown in
this analysis, a tax that increases prices by 20%, thus generating a ten year cumulative per capita
average weight loss of 2.9 pounds, would need to be weighed against the distortionary nature of
the tax and other unexpected and potentially undesirable effects.
ACKNOWLEDGMENTS
The authors wish to thank Shawn Karns for considerable assistance in preparing the data for
analyses as well as Eliza Kruger and Chay Junxing for their assistance with NHANES and
census data respectively. This grant was supported by funding from Healthy Eating Research, a
national program of the Robert Wood Johnson Foundation (Grant #65062). The views expressed
in this paper are those of the authors and not necessarily those of any funding body, others whose
support is acknowledged, or the USDA. The funders had no role in study design, data collection
and analysis, decision to publish, or preparation of the manuscript.
Page 22 of 25
REFERENCES
Anderson SP, de Palma A, Kreider B. Tax incidence in differentiated product oligopoly. Journal
of Public Economics 2001;81; 173-192.
Andreyeva T, Long MW, Brownell KD. 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 2010;100; 216-222.
Avena NM, Bocarsly ME, Hoebel BG. Animal models of sugar and fat bingeing: relationship to
food addiction and increased body weight. Methods in molecular biology (Clifton, N.J.)
2012;829; 351-365.
Becker G. A Tax on Sodas? The Becker-Posner Blog. 2009. http://www.becker-posner-
blog.com/2009/05/a-tax-on-sodas-becker.html (accessed Jan 10, 2012).
Becker GS. A Theory of the Allocation of Time. The Economic Journal 1965;75; 493-517.
Bergman MU, Hansen NL. 2012. Are excise raxes on beverages fully passed through to prices?
The Danish evidence. (Ed)^(Eds). University of Copenhagen; 2012.
Besley T. Commodity taxation and imperfect competition: A note on the effects of entry. Journal
of Public Economics 1989;40; 359-367.
Besley T, Rosen H. Sales taxes and prices: an empirical analysis. 1999.
Bhattacharya J, Sood N. Who pays for obesity? Journal of Economic Perspectives 2011;25; 139.
Blough DK, Madden CW, Hornbrook MC. Modeling risk using generalized linear models.
Journal of Health Economics 1999;18; 153-171.
Bonnet C, Réquillart V. 2012. Sugar policy reform, tax policy and price transmission in the soft
drink industry. (Ed)^(Eds). 2012.
Box GEP, Cox DR. An Analysis of Transformations. Journal of the Royal Statistical Society.
Series B (Methodological) 1964;26; 211-252.
Brownell KD, Farley T, Willett WC, Popkin BM, Chaloupka FJ, Thompson JW, Ludwig DS.
The Public Health and Economic Benefits of Taxing Sugar-Sweetened Beverages. New England
Journal of Medicine 2009;361; 1599-1605.
Chen L, Appel LJ, Loria C, Lin P-H, Champagne CM, Elmer PJ, Ard JD, Mitchell D, Batch BC,
Svetkey LP, Caballero B. Reduction in consumption of sugar-sweetened beverages is associated
with weight loss: the PREMIER trial. The American Journal of Clinical Nutrition 2009;89; 1299-
1306.
Chernozhukov V, Hansen C. Instrumental variable quantile regression: A robust inference
approach. Journal of Econometrics 2008;142; 379-398.
Cournot A. Researches into the Mathematical Principles of the Theory of Wealth, trans. N.
Bacon. New York: Kelly 1960.
Cox TL, Wohlgenant MK. Prices and Quality Effects in Cross-Sectional Demand Analysis.
American Journal of Agricultural Economics 1986;68; 908-919.
Deaton A. Quality, Quantity, and Spatial Variation of Price. The American Economic Review
1988;78; 418-430.
Delipalla S, Keen M. The comparison between ad valorem and specific taxation under imperfect
competition. Journal of Public Economics 1992;49; 351-367.
Delipalla S, O’Donnell O. Estimating tax incidence, market power and market conduct: The
European cigarette industry. International Journal of Industrial Organization 2001;19; 885-908.
Page 23 of 25
Dharmasena S, Capps O. Intended and unintended consequences of a proposed national tax on
sugar-sweetened beverages to combat the U.S. obesity problem. Health Economics 2011; n/a-
n/a.
Duan N, Manning WG, Jr., Morris CN, Newhouse JP. A Comparison of Alternative Models for
the Demand for Medical Care. Journal of Business & Economic Statistics 1983;1; 115-126.
Duffey KJ, Gordon-Larsen P, Shikany JM, Guilkey D, Jacobs DR, Jr, Popkin BM. Food Price
and Diet and Health Outcomes: 20 Years of the CARDIA Study. Arch Intern Med 2010;170;
420-426.
Finkelstein EA, Khavjou OA, Thompson H, Trogdon JG, Pan L, Sherry B, Dietz W. Obesity and
Severe Obesity Forecasts Through 2030. American journal of preventive medicine 2012;42; 563-
570.
Finkelstein EA, Zhen C, Nonnemaker J, Todd JE. Impact of Targeted Beverage Taxes on
Higher- and Lower-Income Households. Arch Intern Med 2010;170; 2028-2034.
Finkelstein EA, Zuckerman L. The fattening of America: How the Economy Makes Us Fat, If It
Matters, and What to Do About Its. John Wiley & Sons, Inc.: Hoboken, New Jersey; 2008.
Gearhardt AN, Corbin WR, Brownell KD. Food Addiction: An Examination of the Diagnostic
Criteria for Dependence. Journal of Addiction Medicine 2009;3; 1-7
10.1097/ADM.1090b1013e318193c318993.
Gilhooly CH, Das SK, Golden JK, McCrory MA, Dallal GE, Saltzman E, Kramer FM, Roberts
SB. Food cravings and energy regulation: the characteristics of craved foods and their
relationship with eating behaviors and weight change during 6 months of dietary energy
restriction. Int J Obes 2007;31; 1849-1858.
Gladson. Nutrition Database. 2009.
http://www.gladson.com/SERVICES/NutritionDatabase/tabid/89/Default.aspx (accessed Jan 9,
2012).
Gorman WM. A Possible Procedure for Analysing Quality Differentials in the Egg Market. The
Review of Economic Studies 1956, 1980;47; 843-856.
Griffith R, Nesheim L, O'Connell M. 2010. Sin taxes in differentiated product oligopoly: an
application to the butter and margarine market. (Ed)^(Eds), Cemmap Working Papers. The
Institute for Fiscal Studies; 2010.
Hanson A, Sullivan R. The Incidence of Tobacco Taxation: Evidence from Geographic Micro-
Level Data. National Tax Journal 2009;62; 677-698.
Harris JE. 1987. The 1983 Increase in the Federal Cigarette Excise Tax. (Ed)^(Eds). National
Bureau of Economic Research, Inc; 1987.
Hausman JA, Pakes A, Rosston GL. Valuing the Effect of Regulation on New Services in
Telecommunications. Brookings Papers on Economic Activity. Microeconomics 1997;1997; 1-
54.
Kamerow D. The case of the sugar sweetened beverage tax. BMJ 2010;341.
Kenkel DS. Are Alcohol Tax Hikes Fully Passed Through to Prices? Evidence from Alaska.
American Economic Review 2005;95; 273-277.
Koenker R, Hallock KF. Quantile Regression. The Journal of Economic Perspectives 2001;15;
143-156.
Laibson D. Golden Eggs and Hyperbolic Discounting. The Quarterly Journal of Economics
1997;112; 443-478.
Laibson DI. 1994. Hyperbolic discounting and consumption. (Ed)^(Eds). Massachusetts Institute
of Technology, Department of Economics; 1994.
Page 24 of 25
Lancaster KJ. A New Approach to Consumer Theory. Journal of Political Economy 1966;74;
132-157.
Lin B-H, Smith TA, Lee J-Y, Hall KD. Measuring weight outcomes for obesity intervention
strategies: The case of a sugar-sweetened beverage tax. Economics & Human Biology
2011;9; 329-341.
Ludwig DS, Peterson KE, Gortmaker SL. Relation between consumption of sugar-sweetened
drinks and childhood obesity: a prospective, observational analysis. The Lancet 2001;357; 505-
508.
Malik VS, Schulze MB, Hu FB. Intake of sugar-sweetened beverages and weight gain: a
systematic review. The American Journal of Clinical Nutrition 2006;84; 274-288.
Manning WG, Mullahy J. Estimating log models: to transform or not to transform? Journal of
Health Economics 2001;20; 461-494.
Marchiori D, Corneille O, Klein O. Container size influences snack food intake independently of
portion size. Appetite 2012;58; 814-817.
Musgrave RA. The theory of public finance: a study in public economys. McGraw-Hill New
York; 1959.
Nelder JA, Wedderburn RWM. Generalized Linear Models. Journal of the Royal Statistical
Society. Series A (General) 1972;135; 370-384.
Nevo A. New Products, Quality Changes, and Welfare Measures Computed from Estimated
Demand Systems. The Review of Economics and Statistics 2003;85; 266-275.
New York Post. Paterson pushes soda tax, again. October 20, 2009.
http://www.nypost.com/p/news/local/item_Wg3nnkUrebcb0WdRR9T2IP (accessed January 16,
2012).
Nielsen. Nielsen Homescan Data, ERS-USDA version. 2006. http://www.ncppanel.com
(accessed Jan 9, 2012).
O'Donoghue T, Rabin M. Optimal sin taxes. Journal of Public Economics 2006;90; 1825-1849.
Park RE. Estimation with Heteroscedastic Error Terms. Econometrica 1966;34; 888.
Paterson D. Why we need an obesity tax. 2008.
http://www.cnn.com/2008/US/12/18/paterson.obesity/index.html (accessed Jan 9, 2012).
Popkin BM. Patterns of beverage use across the lifecycle. Physiol Behav 2010;100; 4-9.
Posner R. A Soda or Calorie Tax to Reduce Obesity. The Becker-Posner Blog. 2009.
http://www.becker-posner-blog.com/2009/05/a-soda-or-calorie-tax-to-reduce-obesity--
posner.html (accessed Jan 10, 2012).
Schroeter C, Lusk J, Tyner W. Determining the impact of food price and income changes on
body weight. Journal of Health Economics 2008;27; 45-68.
Seade J. 1985. Profitable Cost Increases and the Shifting of Taxation : Equilibrium Response of
Markets in Oligopoly. (Ed)^(Eds). University of Warwick, Department of Economics; 1985.
Smith TA, Biing-Hwan L, Lee J-Y. 2010. Taxing Caloric Sweetened Beverages: Potential
Effects on Beverage Consumption, Calorie Intake, and Obesity. (Ed)^(Eds). United States
Department of Agriculture, Economic Research Service; 2010.
Stern N. The effects of taxation, price control and government contracts in oligopoly and
monopolistic competition. Journal of Public Economics 1987;32; 133-158.
Story M, French S. International Journal of Behavioral Nutrition and Physical Activity.
International Journal of Behavioral Nutrition and Physical Activity 2004;1; 3.
Terza JV, Basu A, Rathouz PJ. Two-stage residual inclusion estimation: Addressing endogeneity
in health econometric modeling. Journal of Health Economics 2008;27; 531-543.
Page 25 of 25
U.S. Department of Agriculture. Composition of Foods Raw, Processed, Prepared. USDA
National Nutrient Database for Standard Reference, Release 23. 2010.
http://www.ars.usda.gov/nutrientdata (accessed May 9, 2011).
U.S. Department of Agriculture, U.S. Department of Health and Human Services. Dietary
Guidelines for Americans 2010s. U.S. Government Printing Office: Washington, DC; 2010.
Vartanian LR, Schwartz MB, Brownell KD. Effects of Soft Drink Consumption on Nutrition and
Health: A Systematic Review and Meta-Analysis. American Journal of Public Health 2007;97;
667-675.
Wansink B, van Ittersum K. Portion Size Me: Downsizing Our Consumption Norms. Journal of
the American Dietetic Association 2007;107; 1103-1106.
Whitney EN, Cataldo CB, Rolfes SR. Understanding Normal and Clinical Nutritions.
Wadsworth/Thomson Learning: Belmont, CA; 2002.
Young DJ, Bielinska-Kwapisz A. Alcohol taxes and beverage prices. National Tax Journal
2002;55.
Zhen C, Wohlgenant MK, Karns S, Kaufman P. Habit Formation and Demand for Sugar-
Sweetened Beverages. American Journal of Agricultural Economics 2011;93; 175-193.
