You are on page 1of 7

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/227359187

Do Positive Nutrition Shelf Labels Affect Consumer Behavior?


Findings from a Field Experiment with Scanner Data

Article  in  American Journal of Agricultural Economics · January 2011


DOI: 10.1093/ajae/aaq104 · Source: RePEc

CITATIONS READS

52 580

3 authors, including:

Jill Mccluskey Hayley H. Chouinard


Washington State University Washington State University
188 PUBLICATIONS   5,627 CITATIONS    65 PUBLICATIONS   1,032 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

China's Food Consumption and Trade Policies View project

dual career economics View project

All content following this page was uploaded by Jill Mccluskey on 01 September 2014.

The user has requested enhancement of the downloaded file.


American Journal of Agricultural Economics Advance Access published December 9, 2010

DO POSITIVE NUTRITION SHELF LABELS AFFECT


CONSUMER BEHAVIOR? FINDINGS FROM A FIELD
EXPERIMENT WITH SCANNER DATA
JOSHUA P. BERNING, HAYLEY H. CHOUINARD, AND JILL J. MCCLUSKEY

Downloaded from ajae.oxfordjournals.org at Washington State University Libraries on December 15, 2010
Shoppers are exposed to an ever-growing num- in which nutrition labels are associated with
ber of nutrition labels in grocery stores. Manu- quality characteristics that are unfavorable and
facturers display nutrition claims such as “low therefore have a negative effect on the percep-
fat,”“low calorie,” and “zero trans fat” on prod- tion of a food product. A nutrition label could
uct packaging in attempts to draw favorable highlight characteristics that negatively affect
attention to their products. Certain manufac- the perception of the product. If this is the case,
turers have even created their own proprietary shoppers could use nutrition labels to identify
labels, such as Pepsi’s Smart Choice and Kraft’s products that do not possess nutritional char-
Sensible Solution, designed to help shoppers acteristics negatively associated with taste. For
evaluate healthy products within their brands. example, shoppers could use nutrition labels to
Health advocacy groups allow the use of their avoid “low fat” ice cream because they antic-
names on products,such as theAmerican Heart ipate a less satisfying taste than regular ice
Association’s Heart-Check label. Some gro- cream. Although shoppers might abstain from
cery store chains offer their own proprietary purchasing a product after observing a nutri-
nutrition labels, which they affix on products tion claim, they are still better off because they
that meet their own definitions of healthy. Such are allowed to make a more informed decision.
nutrition labels are intended to reduce search In this sense, nutrition labels do not have to
costs for shoppers who want to identify prod- direct shoppers to select healthy choices to be
ucts with specific characteristics. For example, beneficial. While this may not be a favorable
a diabetic shopper may use a label identifying outcome in terms of public policy, nutrition
“no sugar added,” while a hypertensive person labels also increase consumer welfare by help-
may focus on “low sodium” labels. Nutrition ing shoppers to identify preferred products
labels may ultimately complement products by (Teisl, Bockstael, and Levy 2001).
highlighting favorable quality characteristics We conducted an experiment to examine the
(Becker and Murphy 1993). effect of nutrition labels that highlight spe-
While it is reasonable for one to expect cific positive nutrition standards on microwave
that positive nutrition information will com- popcorn sales. We affixed nutrition labels on
plement a product, there may also be instances grocery store shelves below different types of
microwave popcorn in five stores in the East
Bay area of California. Specifying an incom-
Joshua P. Berning is an assistant professor in the Department plete demand system,we estimate the impact of
of Agricultural and Resource Economics at the University of
Connecticut. Hayley H. Chouinard is associate professor in the positive nutrition labels on sales of microwave
School of Economic Sciences at Washington State University. Jill popcorn products that merit a positive nutri-
J. McCluskey is professor in the School of Economic Sciences at tion label based on Food and Drug Adminis-
Washington State University. The authors gratefully acknowledge
funding from the National Research Initiative of the Cooperative tration (FDA) standards and microwave pop-
State Research, Education and Extension Service, USDA, Grant corn products that do not merit a nutrition
#2005-01743. We thank Sofia Villas-Boas and Kristin Kiesel for
their help conducting the store experiments and discussions on
label. We find that nutrition labels decrease
related work. The research also benefited from discussions on sales of healthy popcorn and increase sales of
related work with David Sprott and Ken Manning. This article was unhealthy popcorn across all stores.
presented in an invited-paper session at the 2010 annual meeting
of the Allied Social Science Associations in Atlanta, GA. The arti-
While nutrition labels can reduce search
cles in these sessions are not subjected to the journal’s standard costs for healthier foods, they may also signal
refereeing process. less preferred taste. Shoppers may infer from

Amer. J. Agr. Econ. 1–6; doi: 10.1093/ajae/aaq104


© The Author (2010). Published by Oxford University Press on behalf of the Agricultural and Applied Economics
Association. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com
2 Amer. J. Agr. Econ.

nutrition labels other product characteris- that information regarding fat and sugar con-
tics besides nutritional content. Our find- tent affects consumer responses to flavored
ings highlight an important issue that has milk. Whereas one might expect nutrition
not been extensively researched in food mar- labels to increase purchases of healthy prod-
keting: the possible unintended effects cre- ucts, nutrition labels may also direct consumers
ated by nutrition labels. Although we do not to select products that they associate with
directly measure the effect of the nutrition their preferred taste, which may be relatively
labels on consumer purchase intentions, our unhealthy.
findings of an aggregate effect on sales is
relevant to both policy and industry. Mar-
keting professionals and government agen- Experiment and Methodology
cies need to consider the unintended conse-

Downloaded from ajae.oxfordjournals.org at Washington State University Libraries on December 15, 2010
quences of promoting nutrition. Further, poli- We affixed positive nutrition labels on gro-
cymakers must consider the effect of industry cery store shelves below boxes of microwave
nutrition labels on consumer behavior and popcorn in five stores belonging to the same
not just whether their content meets FDA national chain in the East Bay area of northern
standards. California for four consecutive weeks begin-
ning October 10, 2007. For all labels, the nutri-
tion claims are based on FDA standards. An
Preferences for Food Quality example of three of the labels is provided in
figure 1. The stores differed in size as well as
the number of products that received labels
While a majority of research on nutrition (table 1). Scanner data was collected for the
labels identifies their positive effects on con- entire microwave popcorn product category
sumer behavior, there are several food quality from the five experiment stores and five control
studies which suggest that product labels that stores.
describe nutritional content could also nega- We examined the impact of nutrition labels
tively impact consumer perception of product on popcorn sales by estimating a linear-
quality and taste. In their examination of con- quadratic ideal demand system (LQ-IDS) that
sumer acceptance of soy products,Wansink and was linear in income and quadratic in prices
Park (2002) find that soy labels negatively bias (LaFrance 1990, 2004). The model was derived
taste perceptions and attitudes toward foods from an expenditure function that was also
perceived to contain soy but without any soy. linear in income and quadratic in prices and sat-
Solheim (1992) finds that information about isfied integrability, allowing for exact welfare
fat content influences the response to differ- measures:
ent types of sausage, regardless of the actual
fat content. Aaron, Mela, and Evans (1994) (1) e(p, p̃, s, u) = α  p + p As
report similar effects with fat spread. Wardle

and Huon (2000) find that children rate drinks + 0.5p Bp + θ (p̃, s, u)eγ p .
with “healthy labels” as less pleasant and are
less likely to ask their parents to buy them. In equation (1) p and p̃ are vectors of food
Baranowski et al. (1993) summarize this neg- and nonfood consumer prices; α and γ are
ative correlation by stating the conventional vectors of parameters; A is a matrix of param-
wisdom: if a food tastes good, it must not be eters associated with a vector of demographic
good for me and if a food tastes bad, it is variables, s; and B is a symmetric matrix of
probably good for me. parameters. The term θ (p̃, s, u) is the constant
The effect of nutrition labels on perceived of integration and is increasing in utility, u,
taste appears to differ among products. For but otherwise cannot be identified. Based on
example, Kahkonen, Tuorila, and Rita (1996) equation (1), the indirect utility function can be
find that nutritional content of food affects solved as:
hedonic ratings with fat spreads, whereas
Kahkonen, Tuorila, and Lawless (1997) find (2) v(p, p̃, m, s)
no effect with yogurt and note that yogurt is 
already assumed to be low fat by consumers. = (m − α  p − p As − 0.5p Bp)e−γ p
This difference may not be due strictly to per-
ceived differences of healthy and unhealthy where m is income. The LQ-IDS model is
foods. For example, Shepherd et al. (1992) find consistent with economic theory of consumer
Berning, Chouinard, and McCluskey Do Positive Nutrition Shelf Labels Affect Consumer Behavior? 3

Figure 1. Nutrition label examples

Downloaded from ajae.oxfordjournals.org at Washington State University Libraries on December 15, 2010
The first label was put on products that were low fat. The second label was on products that were
low calorie and low fat and had no trans fat. The third label included a statement that the low
fat claim was based on FDA standards.

Table 1. Number of Labels Used in Each Store and Store Characteristics


Store
Label Description 1 2 3 4 5
Low calorie 6 17 2
Low fat 5 16 13 1
No trans fat 14 14
Low fat/low calorie 0
Low calorie/no trans fat 4
Low fat/no trans fat 1
Low calorie/low fat/no trans fat 14
Total labeled 25 16 13 17 36
Total number of products 64 62 26 62 60
Products labeled, % 39 26 50 27 60
Square footage 30,440 27,178 19,348 26,425 30,168

behavior, allowing for the assumption of a on FDA standards; all other products are
representative consumer when using aggre- categorized as unhealthy. Weekly total sales
gate sales data. The incomplete demand sys- and total quantity of all popcorn UPCs were
tem also avoids the need to assume weak compiled from the first week of 2006 to the
complementarity or two-stage budgeting. With fifth week of 2008 for each of the ten stores.
an incomplete demand system, no adding-up Prices are calculated as total sales divided by
restriction is necessary, and homogeneity is total quantity. From this, weekly quantities and
implicitly satisfied by normalizing all prices by prices for both categories of popcorn are calcu-
the numeraire good. In this estimation, the vec- lated as weighted weekly averages. Estimates
tors of prices p and income m are normalized of weekly income from 2006–2008 are calcu-
using a nonfood consumer price index for the lated for each store as well.To do this,per capita
western United States, given as p̃. The system income by zip code for each store is taken from
of demand equations for the LQ-IDS model is the 1999 U.S. Census. To generate monthly
to be derived from equation (2): per capita income, we multiply the 1999 per
capita income by the ratio of monthly state
(3) q = α + As + Bp level personal income for any month in 2006–
2008 to monthly state level personal income
+ γ (m − α  p − p As − 0.5p Bp) in 1999. Weekly personal income by zip code
is then interpolated from the monthly values.
where q is a vector of quantity purchased. Although this estimate is an inexact measure of
For the purposes of this article, we aggregate income, it captures changes in the overall state
all popcorn universal product codes (UPCs) economy over the two-year period, as well as
into two categories: healthy and unhealthy. income differences among store zip codes.
Healthy products are defined as those that Income endogeneity is tested following the
merit at least one positive nutrition label based procedure outlined in Villas-Boas and Winer
4 Amer. J. Agr. Econ.

(1999). Income is first regressed on all the vari- This model examines the general effect of
ables, and then the residuals of this estimation nutrition labels on the demand for both cate-
are included in the system of demand equations gories of microwave popcorn and treats shop-
as an explanatory variable. In this analysis, the pers as responsive to the provision of product
residuals are not significant, suggesting that information. Consumers may respond in dif-
income is exogenous. Subsequently, income is ferent ways. If the nutrition label positively
treated as an exogenous variable throughout complements the product, there should be
the study. We tested and found prices to be an increase in purchases. Alternatively, if the
endogenous. Lacking data on potential instru- nutrition information negatively complements
ments, we follow Villas-Boas and Winer (1999) the product, there should be a decrease in pur-
and use lagged prices as instruments. chases. During the treatment period, demand
To examine the impact of nutrition labels on for healthy popcorn decreased (−5.58 ounces),

Downloaded from ajae.oxfordjournals.org at Washington State University Libraries on December 15, 2010
the sales of healthy and unhealthy popcorn, and demand for unhealthy popcorn increased
a dummy variable is included in the demand (7.53 ounces), although the latter effect is sta-
system to indicate whether a nutrition label tistically significant at only the 80% level of
was present. This approach attempts to cap- confidence (table 2).
ture the effect of any type of nutrition label Since the demand model accounts for price
on demand. That is, the effect of a “low fat” changes, it could be that the sales of healthy
label is grouped in with the effect of the “low popcorn stayed the same despite a price
calorie” label, as well as all the other differ- decrease. Likewise, unhealthy popcorn sales
ent labels. Fixed effects are included for each may have stayed the same despite a price
store location, as well as a time trend and sea- increase. We examine the sales of each store
sonal effects. Although this demand system is to find how sales have shifted. Specifically, we
incomplete in the sense that only the products compare the average sales during the treat-
of interest are estimated, the demand system ment and nontreatment periods (table 3). In all
includes information regarding all products via but two stores, the total average sales of pop-
the normalization of the nonfood Consumer corn were lower during the treatment period.
Price Index, and no equations are dropped. As In terms of healthy popcorn, the average price
such, the system of demand equations is esti- and sales were lower during the treatment
mated using full-information maximum likeli- period in three of the treatment stores (stores
hood (FIML). As pointed out by Dhar, Chavas, 1, 2, and 4) and one of the nontreatment
and Gould (2003), FIML estimates are asymp- stores (store 7). This indicates that even with a
totically efficient, and this efficiency does not lower price,sales dropped during the treatment
depend on the choice of instruments as with period for healthy popcorn.
three-stage least squares. There is a steep increase in total sales of
healthy and unhealthy popcorn for all stores
in the last week of the experiment, attributed
Results almost entirely to an increase in the sales
of unhealthy popcorn. Although this increase
We estimate the demand system that includes in sales in the last week is present only in
a dummy variable to identify the presence of three of the treatment stores and two of the
a nutrition label on the grocery store shelves. nontreatment stores, we are not able to identify
Table 2. Demand Estimates for Healthy and Unhealthy Popcorn
Healthy Popcorn Unhealthy Popcorn
Parameter Estimate p-Value Estimate p-Value
Own price −383.98 <.0001 −1530.5 <.0001
Cross price 31.53 0.0365 31.53 0.0365
Label effect −5.58 0.0066 7.53 0.2011
Income effect 0.001 0.1742 0.012 <.0001
Time trend −0.068 0.164 −0.67 <.0001
Summer −0.57 0.5135 −0.27 0.9151
Fall 1.31 0.1872 12.29 <.0001
Winter 5.4 <.0001 5.48 0.0311
Adj. R2 0.724 0.718
Berning, Chouinard, and McCluskey Do Positive Nutrition Shelf Labels Affect Consumer Behavior? 5

Table 3. Average Popcorn Sales During Treatment Period vs. Nontreatment Period
Total Average Healthy Unhealthy
Sales Sales Price Sales Price
Treatment stores 1 lower lower lower lower higher
2 lower lower lower lower higher
3 higher higher lower higher higher
4 lower lower lower lower higher
5 lower higher lower lower higher
Nontreatment stores 6 lower higher lower lower higher
7 lower lower lower lower higher

Downloaded from ajae.oxfordjournals.org at Washington State University Libraries on December 15, 2010
8 lower higher lower lower higher
9 lower higher lower lower higher
10 higher higher lower higher lower

Table 4. Demand Estimates for Healthy and Unhealthy Popcorn with Posttreatment Effect
Included
Healthy Popcorn Unhealthy Popcorn
Parameter Estimate p-Value Estimate p-Value
Own price −403.19 <.0001 −1511.82 <.0001
Cross price 23.4 0.13 23.4 0.13
Label effect −8.35 <.0001 9.37 0.13
Posttreatment effect −5.37 <.0001 4.37 0.23
Income effect 0.002 0.06 0.01 0.0002
Time trend −0.06 0.21 −0.66 <.0001
Summer −0.92 0.29 −0.21 0.93
Fall 2.4 0.02 11.18 0.0002
Winter 6.42 <.0001 4.44 0.09
Adj. R2 0.726 0.717

whether this sharp increase is due to some highly significant as well (−8.35). Unhealthy
unidentified marketing effect. To control for popcorn sales are higher posttreatment at
some unidentified effect in this final week of the roughly the 80% level, but now the treat-
treatment, we remove the final treatment week ment period is significant at approximately
from the treatment sample and test whether the 90% level. This could indicate a linger-
the label effect is significant for the remain- ing impact of the nutrition label on the sales
ing three weeks. The sales of healthy popcorn of popcorn or it could identify a shift in sales
are still significantly negative (−8.2), whereas due to some unidentified variable. The post-
the effect on the sales of unhealthy popcorn experiment data last only six weeks after the
is no longer significantly different during the experiment and do not allow us to examine how
experimental time period. While this supports sales change further beyond that period.
the finding that nutrition labels lead to a reduc-
tion in sales of healthy popcorn, it is not clear
whether the increase in unhealthy popcorn is Conclusions
due strictly to some unknown promotion, the
nutrition labels, or a combination of the two. We find that positive nutrition labels affixed to
Without interviewing the shoppers, one cannot grocery store shelves below units of microwave
discriminate between effects. popcorn tend to decrease purchases of
To examine posttreatment sales, we add microwave popcorn that merit positive nutri-
a dummy variable to identify the posttreat- tion claims and may also increase purchases of
ment period as well (table 4). After the treat- unhealthy popcorn. The results of this research
ment period, the sales of healthy popcorn are highlight the importance of understanding
still significantly lower than pretreatment sales how nutrition information impacts consumer
(−5.37), and the treatment period effect is still behavior. From a manufacturer’s and store’s
6 Amer. J. Agr. Econ.

perspective, the implications of this research Dhar, T., J. P. Chavas, and B. W. Gould. 2003.
are relevant to a profit maximization strategy. Issues in Demand Analysis for Differenti-
While one would expect positive nutrition ated Products. American Journal of Agri-
claims to lead to increased purchases, we find cultural Economics 85: 605–617.
evidence to the contrary. The implications are Kahkonen, P., H. Tuorila, and H. Lawless.
also important to an overall store branding 1997. Lack of Effect of Taste and Nutri-
strategy. Perhaps not surprisingly, consumers tion Claims on Sensory and Hedonic
may perceive additional product information Responses to a Fat-Free Yogurt. Food
unfavorably, which could have undesirable Quality and Preference 8: 125–130.
results. Kahkonen, P., H. Tuorila, and H. Rita. 1996.
The implications for policymakers are espe- How Information Enhances Acceptability
cially relevant given the epidemic of obesity

Downloaded from ajae.oxfordjournals.org at Washington State University Libraries on December 15, 2010
of a Low-Fat Spread. Food Quality and
and its related and costly health problems. Preference 7: 87–94.
Our results prompt consideration of impor- LaFrance, J. T. 1990. Incomplete Demand
tant policy questions. Are private nutrition Systems and Semilogarithmic Demand
labels effective in terms of public health or Models. Australian Journal of Agricultural
are they potentially harmful? Should the fronts Economics 34: 118–131.
of packages or shelf labels be required to LaFrance, J. T. 2004. Integrability of the Lin-
also have negative information, such as the ear Approximate Almost Ideal Demand
United Kingdom’s Traffic Light System? This System. Economic Letters 84: 297–303.
research provides a modest example of unin- Shepherd, R., P. Sparks, S. Bellier, and M. M.
tended effects generated by a specific type of Raats. 1992. The Effects of Information
nutrition label with a specific type of prod- on Sensory Ratings and Preferences: The
uct. Further research across more products Importance of Attitudes. Food Quality and
with more types of nutrition labels and claims Preference 3: 147–155.
should be investigated. Solheim, R. 1992. Consumer Liking for
Sausages Affected by Sensory Quality and
Information on Fat Content. Appetite 19:
References 285–292.
Teisl, M. F., N. E. Bockstael, and A. S.
Aaron, J. K., D. J. Mela, and R. E. Evans. Levy. 2001. Measuring the Welfare Effects
1994. The Influence of Attitudes, Beliefs of Nutrition Information. American Jour-
and Label Information on Perceptions nal of Agricultural Economics 83: 133–
of Reduced-Fat Spread. Appetite 22: 149.
25–37. Villas-Boas, J. M., and R. S. Winer. 1999. Endo-
Baranowksi, T., S. Domel, R. Gould, J. Bara- geneity in Brand Choice Models. Manage-
nowski, S. Leonard, F. Triber, and R. Mullis. ment Science 45: 1324–1338.
1993. Increasing Fruit and Vegetable Con- Wansink, B., and S.-B. Park. 2002. Sensory Sug-
sumption Among 4th and 5th Grade Stu- gestiveness and Labeling: Do Soy Labels
dents: Results from Focus Groups Using Bias Taste? Journal of Sensory Studies 17:
Reciprocal Determinism. Journal of Nutri- 483–491.
tion Education 25: 114–120. Wardle, J., and G. Huon. 2000. An Experimen-
Becker, G. S., and K. M. Murphy. 1993. A Sim- tal Investigation of the Influence of Health
ple Theory of Advertising as a Good or Information on Children’s Taste Prefer-
Bad. Quarterly Journal of Economics 108 ences. Health Education Research: Theory
(4): 941–964. and Practice 15: 39–44.

View publication stats

You might also like