Prepared for FTC/NW Conference
November 19, 2009
Can Information Costs Confuse Consumer
Choice?
Nutritional Labels in a Supermarket
Experiment
Kristin
Kristin
Kiesel
Kiesel
and Sofia Villas
and Sofia Villas
-
-
Boas, U. C. Berkeley
Boas, U. C. Berkeley
Research Objectives
• Measure the effects of nutritional information on
consumer purchasing decisions using a field
experiment
• Store level scanner data
Total effect on quantity sold
Willingness to pay (WTP) for labels (preliminary)
• Transaction level data (in progress)
Purchase history
Purchase amount
Entry of new consumers into category
Motivation
• Consumers devote minimal time and effort to processing
grocery product information at the point of purchase
• Does including nutrition information in a “easy-to-process”
shelf-label format
decrease the search costs associated with obtaining
nutritional information and
result in healthier product-selection decisions?
• Grocery retailers have an opportunity to assist consumers in
making healthier purchase decisions.
Costs Processing Available Information?
Low Fat
Low Calorie
No Trans Fat
Low Fat
Low Calorie
Low Fat
Low Fat*
*According to FDA
nutrient content claim
Low Calorie
• Display more salient fashion repetition of already
available information, e.g. No Trans Fat
• Provide a relative scale among alternatives (new format)
Descriptivestatistics:Treatments
A.Treatments
1lowcalorielabel
2lowfatlabel
3lowfatlabelandFDAdisclaimer
4lowcalorieandlowfatlabel
5lowcalorie,lowfat,andlowtransfatlabel
B.TreatmentCharacteristics
12345 12345
LowCalorieLabels 22 ‐‐5 1 21.83(2.04) ‐‐4.71(0.59) 1(.031)
LowFatLabels ‐ 21 ‐ 41 ‐ 21.01(1.94) ‐ 3.91(0.39)
0.971(.167)
LowFat/FDALabels‐‐15 ‐‐ ‐ ‐ 21.01(1.94) ‐‐
NoTransfatLabels ‐‐‐‐12 ‐‐‐‐15.22(1.19)
LowCalorie/LowFatLabels ‐‐‐12 2 ‐‐‐17.11(1.67) 1.86(0.34)
LowCalorie/NoT ransfatLabels ‐‐‐‐3 ‐‐‐‐2.90(0.52)
LowFat/NoTransfatLabels ‐‐‐‐3 ‐‐‐‐3.74(0.52)
LowCalorie/LowFat//NoTransfatLabels ‐‐‐‐16 ‐‐‐‐15.24(1.51)
TotalLabels 22 21 15 21 38 21.83(2.04) 21.01(1.94) 21.01(1.94) 24.92(2.25) 40.99(3.19)
Note:Forthecontrol stores,wereportthemeannumberofproduc tsthatwoul dhavebeentreatedaswell asthestandarddeviationinparenthesis.
treatmentstores controlstore s
-
-
T & C are similar with respect to product assortment & sample
T & C are similar with respect to product assortment & sample
of treated
of treated
products (except T3 smaller store)
products (except T3 smaller store)
-
-
T & C stores serve similar demographics (representative of na
T & C stores serve similar demographics (representative of na
tional
tional
averages)
averages)
-
-
T larger category sales than average controls
T larger category sales than average controls
’
’
sales but within one std dev
sales but within one std dev
Data
Treatments during 4 weeks starting Oct 10, 2007:
• 32 stores (5 treatment and 27 control stores)
• Store level product weekly sales over four years
(focus on narrow window around experiment 14
weeks, five weeks prior and post )
• Socio-demographic statistics provided by the United
States Census Bureau (by zip code) to “match”
Treatment and Control Stores
• Nutritional facts information from products
Empirical Strategy – Difference in Difference
Control Store (C) Treatm Store (T)
Before
Period
Total
units
sold
Effect .
Treatment
Period
Treatment
Period
Before
Period
D
c
D
T
D
c
Effect
D
T
Average Effects
Standard errors clustered at product
Standard errors clustered at product
-
-
store level
store level
Effects by Label Treatment
Standard errors clustered at product
Standard errors clustered at product
-
-
store level
store level
combined
combined
-low fat label: average
decrease of 27.5%
-No trans fat label:
average increase 23%
-But not in combination
with other claims
-All claims label: has
highest information
content but also info
costs, has no effect
Effects on Unlabeled
Results using Store Level Data
• Evidence consistent with information costs mattering
• Increases in quantity sales due to no trans fat labels
• Decreases in quantity sales due to low fat labels (with FDA claim)
• Increase in quantity sales due to low calorie labels (significant at aggregate monthly
effects rather than weekly)
• No inference on unlabeled products (except for low fat FDA claim labels)
• Dissipation of effect when combining claims in single label
• Total category sales decrease 4% due to our labels so labels do not seem to
induce consumption
• we will further investigate with hh data if new consumers enter and how “old”
consumers are affected
• Our results were robust to
Different store and time control structures
Estimation of placebo effects
Additional Evidence
• Results and significance may be affected by remaining uncertainty of
how well average sales in the 27 control stores serve as a
counterfactual
• Synthetic control (SC) method reduces this uncertainty
SC store (created as a combination of all controls)
Best match to the treatment store in pre period
match stores based store characteristics
Investigate significance of treatment effects by estimating placebo
effects for the 27 stores that were never treated
• One treatment unit in this approach, so for each label treatment
we look at aggregate sales by week and store (not by product)
Treatment vs. synthetic control (SC) store
Results confirm DD findings:
1. Low fat label less 27.7 units sold/
week
2. Drop is larger than distribution of
random changes
3. No trans fat increase in sales in the
T relative to SC
4. Low Cal labels increase sales
significantly
5. Other label analysis confirms
results in D in D.
Difference in total units of weekly sales for
low fat labeled products
T-SC (red line)
random changes (placebos/ grey lines)
Recap
• Consumer purchases are affected by nutritional labels
• Effects differ depending on nutritional facts
some claims have NO effect, some +, some -
• Disclosure of source (FDA aprov) discourages sales even more
• More nutrients on label have smaller impacts on change in sales
than a label with just one claim
• Do consumers make inferences about the nutritional content of
non-labeled products? Generally No (except one treatment)
Implications of Results using Store Level Data
• Consumers do not fully incorporate currently available
nutritional information
• Consumers might have taste preferences with respect to
certain nutrients
• Consumers do not perceive FDA approved labels as more
credible in this context
• Consumers do not make inferences on unlabeled products
• Information costs might prevent welfare improving
changes to food choice in context of nutritional labeling
Future Work - preliminary
Willingness to Pay
Demand estimates of no trans fat labels WTP 62 c
Demand estimates of low fat labels WTP of -60 c
Transactions by Household Data
product sales by masked household id over 2 years
No distinct differences in frequency of purchases post T
Higher percentage of new consumers respond to T
Less overall expenditures, larger transaction price savings
Less overall expenditures, larger transaction price savings
Effects seem to dissipate after treatment period
Conclusion
Treatment Effects imply that
• Consumers do not fully incorporate currently available nutritional
information
• Consumers might have taste preferences with respect to certain nutrients
Dissipation of effect when combining claims in single label
Evidence consistent with information costs mattering
Significant Estimates of WTP consistent with reduced form
Treatment Effects
No distinct differences in frequency of purchases before & after
Higher percentage of new consumers respond to treatment
Less overall expenditures, larger transaction price savings
Less overall expenditures, larger transaction price savings
Effects seem to dissipate after treatment period
Thank you!
Just in case questions slides on work
in progress and also for discussant to
see what we did
Demand and Label WTP Estimation Strategy
Indirect latent utility from consumer i choosing product j in week t
U
ijt
= +d
j
+ x
jt
-
i
p
jt
+
jt
+
ijt
d
j
product constant characteristics
x
jt
observed product characteristics, such as our added label
jt
unobserved product characteristics
ijt
consumer preferences about unobserved product characteristics
The probability of buying j among the alternatives is the
probability that j yields maximum U .
Demand and Label WTP Estimation Strategy
Given distribution of
noise of consumer preferences
that will yield
that will yield
a certain probability of purchases as a function of
a certain probability of purchases as a function of
(d,
(d,
,
)
Demand model is estimated to find parameters that give model
Demand model is estimated to find parameters that give model
predicted probabilities of purchase that are the closest to obse
predicted probabilities of purchase that are the closest to obse
rved
rved
frequencies of purchases of brands in the choice set.
frequencies of purchases of brands in the choice set.
To obtain an estimate for the WTP for an attribute
To obtain an estimate for the WTP for an attribute
x
x
in dollars, as
in dollars, as
price is in dollars, divide the estimated marginal U of attribut
price is in dollars, divide the estimated marginal U of attribut
e
e
by
by
marginal U of price
marginal U of price
.
.
Results using Store Level Data
Marginal Utility Estimates for WTP
Marginal Utility Estimates for WTP
Results:
Results:
•
•
Prefer not to buy the
Prefer not to buy the
products we labeled
products we labeled
•
•
-
-
Constant negative
Constant negative
•
•
No trans fat WTP= 62
No trans fat WTP= 62
cents
cents
•
•
Low Fat WTP =
Low Fat WTP =
-
-
60 cents
60 cents
Results using Individual Level Data
Differences in households that respond to labeling
treatment versus households that do not:
• No distinct differences in frequency of purchases before and after
• Higher percentage of new consumers respond to treatment
• Slightly less units purchased when buying labeled products
• Lower individual transaction and total transaction amount for
households responding to treatment
• Responding households buy more on sale/have more savings
• Treatment effects seem to dissipate after treatment period
Differences in consumer type
Differences in households that buy labeled and unlabeled products during treatment period
variable
labeled purchases unlabeled purchases
transaction net amount 85.23 90.24
total transactions amount 1340.79 1385.79
average transaction price promotions 25.35 24.8
average unit price 2.73 2.82
mean
Less overall expenditures
Less overall expenditures
larger transaction price savings
larger transaction price savings
For households that buy in Treatment Period what did
they pre-treatment popcorn purchases look like?
Conclusion:
Higher percentage of new consumers respond to treatment
Households in treatment store
For 25 households w/ observed low fat products that did not buy low
fat (labeled) products during treatment, what did they buy? (to
p
seve
n
products):
Specific treatment effect: low fat (store 2)
POP SECRET MICRO POPCORN HOMESTYLE
POP SECRET MICRO POPCORN BUTTER
POP SECRET HOMESTYLE MICRO POPCORN
POP SECRET MICRO PCRN HOMESTYLE SNACK S
ORV RED MICRO POPCORN BUTTER
ORV RED MICRO POPCORN MVIE THTR BTR
POP SECRET MICRO POPCORN MOVIE THTR BTR
and bought low fat products after the treatment period again
and bought low fat products after the treatment period again
number of hholds %
total households 6641 100.00
households w/observed low fat purchases 2105 31.69
households w/observed purchases during treatment 474 7.14
households w/ observed low fat purchases during treatment 289 4.35
households w/observed low fat purchases and purchases during treatment 186 2.8
households w/observed low fat purchases and low fat purchases during treatmen
t
161 2.42