Food Marketers and Consumers:
Implications of Rationality Asymmetries
for Food Choice and Health
David R. Just
Charles H. Dyson School of Applied Economics and Management
Cornell University
Food Choice and the Rational
Man
► Food
policy is often designed based on the idea
that individuals use information efficiently
 The consumer will use available information
 Can weigh the various consequences of their actions
 Gives appropriate weight to vague or narrow
information
► Individuals
make more than 300 food related
decision each day
 Paying close attention to each would be a waste of time
 We naturally fall back on heuristics, habits and rules of
thumb
The Implications of Heuristic
Choice
► Heuristics
are at best approximations
 They are subject to serious error under the
wrong conditions
 E.g., clean plate rule can be reasonable under
some circumstances and not in others
 May represent misperceptions
► The
implication
 The consumer makes systematic errors
 The consumer could be better off
►Cognitive
costs are prohibitive
Implications of Heuristics
► Policymakers
can make people better off
 Save people from themselves (?)
► We
need to be careful to separate our preferences from theirs
► Could provide a justification for policymakers to impose their
preferences
 People may resist changes that make them better off
► My
► Widens
perceptions are still my perceptions
the set of policy tools
 Before: price, information, content regulation
 Now: Regulating decision context
Some General Principles
► Subtle
►
factors can influence choice
 Suggestive names, lighting, shapes, colors, image, etc.
 Marketers have used this to advantage
In food decisions, much of the problem stems from not
being able to monitor consumption
 Monitoring consumption requires cognitive resources
 Factors that make this task more difficult tend to increase
consumption
 Individuals distort perceived consumption proportional to serving
size
 Containers or packaging can make monitoring easier
► We
are different people when we are distracted
An Extended Example: Portion
Sizes and Purchasing
►A
lot has been said about mega-sized portions
 Super-size me phenomenon
 Long literature documenting the increase in portion
sizes since the 70s
 Not just in restaurants, but also in home recipes
► What




is a normal portion size anyway
Soda can – 12 oz
Starbucks – “Tall” 12 oz (no normative size)
McDonald’s soda – “child” 12 oz (medium is 21 oz)
McDonald’s coffee – “small” 12 oz (medium is 16 oz)
Normative Portion Sizes
► Food
marketers put a lot of effort into these
normative size names
►E.g., Wendy’s offers a Single burger, Double, Triple, a
Jr. and a Double Jr. Deluxe (!)
► Why
do food retailers use such normative
language?
► Two possible motivations
►Informational
►Framing
Standard Models of Portion Size
► Economists
propose that different sizes are
used to price discriminate
 Some value higher quantity, others don’t
 Larger quantities are offered at a volume
discount
 Those who value more benefit from discount
 Those who don’t benefit (weakly) from smaller
quantity being offered
 Profits increase
► Quantity
determines utility
Quantity
► Fast
Food and Cafeterias
 Sizes are usually on display
 Sizes are often posted next to the normative
names
► What
extra information could the labels
provide?
 Information about what others are doing?
 Information about what they want you to buy?
Framing Effects
► How
we phrase the question affects the
answer (Tversky and Kahneman)
►Imagine that the U.S. is preparing for the outbreak of an unusual Asian
disease, whish is expected to kill 600 people. Two alternative programs to
combat the disease have been proposed. Assume that the exact scientific
estimate of the consequences of the program are as follows:
►If Program A is adopted, 200
people will be saved (72%)
►If Program B is adopted, there is
a 1/3 probability that 600 people
will be saved and another 2/3
probability that no people will be
saved (28%)
►If Program C is adopted, 400 people
will die (22%)
►If Program D is adopted, there is a
1/3 probability that nobody will die
and 2/3 probability that 600 people
will die (78%)
Framing Effects
► People
measure utility against reference points
 Lower marginal utility for gains than for losses
 Ratio about ½
► Benartzi
and Thaler find evidence in stock prices
► Tversky and Kahneman suggest that framing
impacts are prevalent in consumption decisions.
► Normative size names could establish a reference
point
 Regular is the status quo
 Upgrades above not as valuable as downgrades below
A Model of Framing Sizes
► Suppose
individuals solve
max U  x |  
xC
► Subject
► Where
to
p  x  x  y
 X is a vector of quantities purchased
 C is the set of available quantity choices (including
origin)
 p (x) is price (possibly nonlinear)
 θ is a vector of quantities given the normative label
A Model of Framing Cont.
► Loss
aversion implies
if xi  0
Ui  x |    0
1
2
1
2
U x  x |    U x  x |   if i  xi  i
if xi  0
Ui  x |    0
if i  j
U x ji  x |    0
i
i
Implications of Framing
► Claim
1: Let x be any positive consumption
quantity available for consumption good i
and suppose that the individual displays loss
averse preferences. Increasing the
reference point must decrease the
willingness to pay for x.
Implications of Framing
► Claim
2: Let x  x be any two
consumption quantities for a consumption
good i and suppose that the individual
displays loss averse preferences. If i   x, x  ,
then increasing θi will increase the
willingness to pay for x relative to x .
Implications of Framing
► Claim
3: Let x  x be the only two positive
consumption quantities available for
consumption good i and suppose that the
individual displays loss averse preferences.
Changing the reference point from i  x to
i  x may decrease the probability of
purchasing either xi  x or xi  x .
Framing
► Three
hypotheses
 Increase norm – ambiguous change in
probability of purchase
(unless only other choice is 0, then decreases)
 Increase norm – increase the WTP to upgrade
 Increase norm – decrease WTP for all sizes of
goods
► Implication
for managers
 Decrease normative sizes.
Our Experiment
► Subjects
were recruited for a lunch
experiment
 Cornell Students
 Conducted at one of the campus dining facilities
► Paid
$15, they purchase lunch and keep the
change
► Items include spaghetti in meat sauce,
salad, pudding, rolls, soda and water.
Our Experiment
► Two
different sizes were offered (one twice
as large as the other) for
 Pudding
 Spaghetti
 Salad
► Sometimes
the sizes were labeled “Half”
and “Regular”
► Sometime they were labeled “Regular” and
“Double”
Part I: An Auction
► 45
Participants (20 Regular/Double, 25
Half/Regular)
► All viewed the items before bidding
► We ran a special nth price auction for each
multiple sized item
 Asked for their maximum willingness to pay for the
smaller
 Asked their maximum additional willingness to pay for
the larger
► 15th
highest bid determines sale price of the larger
► 3rd lowest bid determines the smaller
Demographic Information
Variable
Gender
(1 = Female)
Height (inches)
Weight (pounds)
Age
HALF
0.53 (0.51)
DOUBLE
0.42 (0.51)
Z-Stat
0.641 (0.52)
67.9 (4.3)
145 (26)
19.3 (1.2)
67.9 (5.1)
147 (25)
19.7 (2.2)
0.175 (0.86)
-0.095 (0.92)
-0.033 (0.97)
Average Bids by Treatment
Variable
Small Spaghetti***
Large Spaghetti***
Small Salad**
Large Salad**
Small Pudding**
Large Pudding***
d Spaghetti**
d Salad**
d Pudding***
HALF
Mean
N = 25
$1.07 (1.06)
$1.67 (1.60)
$0.71 (0.98)
$1.08 (1.33)
$0.44 (0.40)
$0.62 (0.63)
$0.63 (0.68)
$0.37 (0.51)
$0.18 (0.31)
DOUBLE
Mean
N = 20
$2.18 (1.02)
$3.50 (1.82)
$1.18 (0.78)
$1.71 (1.06)
$0.89 (0.71)
$1.53 (1.02)
$1.31 (1.25)
$0.58 (0.35)
$0.64 (0.53)
Z-Stat
(P-Value)
-3.400 (0.00)
-3.378 (0.00)
-2.428 (0.02)
-2.199 (0.03)
-2.288 (0.02)
-3.087 (0.00)
-2.539 (0.01)
-2.167 (0.03)
-3.406 (0.00)
Average Treatment Effects
(Controlling for Demographics)
Variable
Small Spaghetti
Large Spaghetti
Small Salad
Large Salad
Small Pudding
Large Pudding
d Spaghetti
d Salad
d Pudding
Model 1
$1.13*** (0.33)
$1.82*** (0.55)
$0.39 (0.31)
$0.38 (0.42)
$0.43** (0.18)
$0.84*** (0.28)
$0.68** (0.32)
$0.05 (0.16)
$0.41*** (0.15)
Model 2
$1.11*** (0.36)
$1.86*** (0.62)
$0.29 (0.37)
$0.21 (0.48)
$0.36* (0.20)
$0.76*** (0.29)
$0.75** (0.35)
$0.01 (0.17)
$0.40** (0.16)
Change in WTP for Larger Size
CDF of Bids
0
1
Cumulative Probability
1
.05
0
5
Bid for Small Spaghetti
CDF of Bids
0
1
Cumulative Probability
1
.05
0
8
Bid for Large Spaghetti
CDF of Bids
0
1
Cumulative Probability
1
.1
0
6
Bid for Spaghetti Upgrade
Interpretation
► There
are substantial incentives to name the sizes
correctly
 Total bids increased from $3.37 to $6.97 (P = 0.00)
► The
effect was much stronger for the hedonic item
 Salad increased in price by 66% and 60% respectively
 Pudding increased by 102% and 147% respectively
 Spaghetti increased by 105% and 108% respectively
► Note
on size of bidding pool
Interpretation
► All
bids increase with smaller normative size
 This is not consistent with LA
► Bids
for larger sizes are bigger for regular to
double than for half to regular
 This is also inconsistent with LA
► Hunger
attenuates significance (somewhat)
 Visceral effects (Loewenstein)?
Part II: Cafeteria Purchasing
► 134
participants
► Same foods and conditions
► Participated for 2 weeks
 First week only the regular
 Second week only the half or the double
► Some
weeks
participants participated only in some
 Allows us to check for order effects (none
detected)
Cafeteria Prices
Small Spaghetti
► Large Spaghetti
► Small Salad
► Large Salad
► Small Pudding
► Large Pudding
►
$2.50
$3.50
$1.50
$2.00
$1.50
$2.00
►
►
►
►
Roll
$0.50
Pepsi
$1.00
Ginger Ale $1.00
Water
$1.00
Consumption Behavior
Variable
HALF
Spaghetti Waste***
Salad Waste**
Pudding Waste
Total Calories
Consumed***
Proportion Purchasing
Spaghetti***
Salad
Pudding
Rolls
Soda
Water
2 (27)
7 (19)
22 (71)
463 (231)
DOUBLE
Z-Stat
(P-Value)
20 (53)
3 (17)
11 (52)
325 (235)
5.348 (0.00)
2.435 (0.01)
1.346 (0.18)
2.854 (0.00)
Large Only
Spaghetti Waste***
Salad Waste**
Pudding Waste**
Total Calories
Consumed***
Proportion Purchasing
Spaghetti**
Salad
Pudding
Rolls***
Soda
Water
0.705 (0.459)
0.302 (0.463)
0.312 (0.466)
0.434 (0.500)
0.169 (0.377)
0.113 (0.320)
0.573 (0.497)
0.566 (0.499)
0.417 (0.496)
0.447 (0.501)
0.563 (0.499)
0.658 (0.478)
Small Only
7 (27)
18 (42)
1 (13)
3 (10)
6 (25)
14 (41)
231 (139)
305 (146)
4.259 (0.00)
-1.420 (0.16)
0.879 (0.380)
0.093 (0.93)
-0.403 (0.69)
-1.267 (0.21)
0.413 (0.495)
0.375 (0.487)
0.163 (0.371)
0.469 (0.502)
0.354 (0.481)
0.500 (0.503)
-2.519 (0.01)
-0.542 (0.588)
-0.961 (0.34)
-2.999 (0.00)
-1.580 (0.11)
-1.028 (0.30)
0.632 (0.487)
0.421 (0.498)
0.228 (0.423)
0.697 (0.462)
0.474 (0.503)
0.579 (0.497)
-3.03 (0.00)
-2.265 (0.02)
-1.986 (0.05)
-2.65 (0.01)
Controlling for Demographics
Variable
Spaghetti Waste
Salad Waste
Pudding Waste
Total Calories
Consumed
Spaghetti Waste
Salad Waste
Pudding Waste
Total Calories
Consumed
Model 2
Model 1
Large Only
-82*** (20)
-86*** (18)
-7** (3)
-5 (3)
-3 (14)
-18 (12)
-83 (53)
-102** (46)
Small Only
12* (7)
13** (6)
1 (3)
1 (2)
10 (8)
8 (7)
58** (28)
76*** (25)
Change in Average Calories
Percent Purchasing
Interpreting
► People
regular
consume more calories when we call it a
 32% or 36% depending on size
► Probability
of purchase goes both ways (only
significant up when regular)
 Seems to depend on type of dish
 LA says decreases
► People
use size names to determine how
much to leave!
Implications
►
Willingness to pay responds to normative language
 Significant increases in WTP by lowering norm (like a
social norm)
 Upgrades more valuable above the norm (violates LA,
also inconsistent with social norms)
 “Regular” increases purchase only for some foods (like
loss aversion)
► Do
we need another model?
 Maybe the label is informative
Labels as Information
► Individuals
respond poorly to visual cues
 Can’t estimate size well (tall vs. short)
 Respond to plate and utensil size
 We eat with our eyes (soup bowls)
► What
if people use names as a crutch
 I know about how much I am willing to pay for a
“regular” – but adjust for perceived size
 Double that for a double (with some adjustment
downward)
 Halve it for a half (with some adjustment upward)
► Might
explain the plate waste
Implications for Retailers
► Higher
WTP for larger names may
encourage larger and larger portions
 Credibility issues may lead to a treadmill
► Responds
like an anomaly
 Exacerbated by stress, hunger, distractions
 Thus, normative sizes more effective in fast
food or convenience food
 What about more hedonic food?
Factors Affecting Food Choice
►
Cognitive Experiential Self Theory (Epstein, 1993)
 Two systems used to evaluate every stimulus
 Experiential system makes snap judgments based on affect
 Cognitive process makes deliberative evaluations based on rational
thinking
 Processing resources (time, stress, distractions) determines which
rules
►
►
►
When resources are few, convenience, affect, and salience
dominate
When resources are many, prices and health information
Hedonic vs. utilitarian foods
 “wants” vs. “shoulds”
 More willing to acquire a should than give up a want
 Significant reactance
Consumption and Control
Preferences
Wealth
Cognitive
(Low Impact)
Consumption
Decision
Affective
(High Impact)
Price
Price of Substitutes and Complements
Attributes (calories, nutrients)
Health Information
Effort
Salience
Structure
Size of portions
Hedonic (Salt, Fat, Sugar)
Atmosphere
Effort/Availability
Distractions
Size of portion
Shape of containers
Primitive
Manufacturer Control
Individual Control
Visceral Factors (hunger)
Mental food accounts
Commitment
Socialization
Habit
The Role of Marketing
►
►
►
Individuals control some of the important factors
Food manufacturers, retailers and marketers control the
majority of factors
Hence, consumption decisions (food and amount) are the
result of a game between manufacturers and consumers
 Consumers are not entirely aware of their
behavior
 Marketers are!
►Or
►
maybe just behave ‘as if’
This is a mechanism design problem
 Could result in a loss of wellbeing due to
incomplete or asymmetric information
 Does asymmetric rationality work the same
way?
The Consumer
► The
consumer problem can be represented as
max U  uc  q   1    ua  q | s   ps
qQs , sS
► Where
ϕ is the level of cognitive resources
determined exogenously
 At least the individual isn’t aware of how their actions
affect it
variable s is an external cue that influences
choice, like portion size
► The variable q is the choice variable (e.g.,
quantity) that influences utility (taste and health)
► The
Food Manufacturers/Retailers
► Food
sellers wish to maximize profits
 To do this they choose the available cues (S ),
prices and the available choices (QS )
 Of course sellers may respond to heuristics also
►I
assume that over time they happen upon
the profit maximizing choice sets
 They behave as if they know the consumer’s
active preferences
A Model of Food Transactions
► The
producer solves
max ps k (q, s )  c  q, s 
S ,Qs , ps 
subject to
max U  uc  q   1    ua  q | s   ps
qQs , sS
► Here
k is the unit for which the individual is
charged
 Package size, or quantity consumed, etc.
Asymmetry of Rationality
► The
seller knows she will receive
ps * k (q*, s*)  c  q*, s *
► The
consumer behaves as if there is
probability ϕ of receiving uc  q *  ps *
and probability (1- ϕ) of receiving ua  q*| s *  ps *
► In actuality the consumer always receives
uc  q *  ps *
The Impacts of Heuristics
► Let q (Qs , S , ps )  arg max uc  q   ps
qQs , sS
 Heuristics are non-trivial if
q (Qs , S , ps )  q *(Qs , S , ps )
► The
most important points:
 Consumers are not the only ones who respond
to policies
 Consumers do not always respond as expected
Is it Always Better to be Rational?
•
The impact on overall welfare also depends
on how the heuristic preferences impact
profit.
• Suppose for instance that individuals are
charged for q, costs are monotonic in q and
that s is costless (~ denotes fully rational eq)
• There are 4 potential cases comparing to fully
rational equilibrium
• Depends on the relationship of
uc ? U *
U * uc ? c * c
An Example: Portion Size
► As
a simple example, suppose we consider
the portion size problem
max p  c  s 
s, p
► Subject

to



uc min q*  s, p  , s  1    ua min q*  s, p  , s | s  p  U
An Example: Portion Size
► Increasing
ds
0
d
iff
distractions increase portion size,


uc '  s   ua  s | s   ua  s | s 
q
s
► If
marginal affective utility is larger than
cognitive utility
 Affective here refers to the deviation from ex
post wellbeing
Sin and Virtue
► This
leads us to define two different types
of foods


 Virtuous
 Sinful
uc '  s  
ua  s | s  
ua  s | s 
q
s
(marginal affective utility lower)


uc '  s   ua  s | s   ua  s | s 
q
s
(marginal affective utility higher)
►
Consumers over consume sinful foods and
under consume virtuous foods
Sin and Virtue
► Result
1: As cognitive resources decrease,
portion sizes will increase more for more
sinful foods and could decrease for virtuous
foods
 We should see some separation
► Result
2: As cognitive resources decrease,
uniformly more sinful foods will be marketed
 Hard to market health as convenience food
 If marketing sinful foods, you want to
encourage distraction
Sin and Virtue
►I
will present some graphical examples using sinful
foods
 Many things flip when we examine virtuous foods
 I will comment on how this changes things without
showing the graphs
► Graphs
assume linear costs, and a strong version
of sinful
 Cognitive arg max is lower, and maximum is lower
 Much more richness if this is not the case, but the
simple model demonstrates how unintended
consequences can happen
100
90
Money Metric Utility
80
70
60
50
Affective
40
Perceived Utility
Cognitive
30
20
10
0
0
2
4
6
8
10
12
14
16
18
20
Portion Size
100
90
80
70
Profit
60
50
Profit
40
Cognitive
Affective
30
20
10
0
0
2
4
6
8
10
Portion Size
12
14
16
18
20
Taxing Size
► The
constraint now becomes




 uc min q*  s, p  , s  1    ua min q*  s, p  , s | s  p  ts  U
► In
equilibrium,
► Producers
𝑑𝑠
𝑑𝑡
=
1
𝑆𝑂𝐶
<0
lose out on profit
 Loss must be greater than ts
► Consumers
welfare will improve if sinful,
and Δ𝑢𝑐 > Δ𝑝 + 𝑡𝑠
100
90
Money Metric Utility
80
70
60
50
Affective
40
Perceived Utility
Cognitive
30
20
10
0
0
2
4
6
8
10
12
14
16
18
20
Portion Size
100
90
80
70
Profit
60
50
Profit
40
With Tax
30
20
10
0
0
2
4
6
8
10
Portion Size
12
14
16
18
20
Welfare: Taxing Large Portions
► Producers
will always perceive a loss of Profit
► Consumers will too if any surplus shared
 This will be unpopular with all involved in the
transaction
► May
or may not increase consumer welfare
 Depends on how sinful the food is
 Reactance can also either increase or decrease welfare
► Incentive
to reduce cognitive resources persists
 If they can influence this at some cost, the equilibrium
will be lead closer to the non-regulation
Welfare: Subsidizing Large
Virtuous Portions
► Profits
to the firm increase
► Consumers may also if surplus is shared
 Thus this may be a very popular program with
all involved in the transaction
► Will
always increase consumer welfare
► Incentive
to increase cognitive resources
persists
► Food may go to waste if subsidy exceeds
decline in perceived utility
Regulating Context
► Suppose
context
instead we considered regulating
 Naming of portion size?
 Requiring an important cognitive cue
► Suppose
we could shift the perceived utility
closer to the cognitive utility
 Could try to increase cognitive resources
(distractions)
 Could try to alter affective utility function
(names, atmospheric cues)
100
90
Money Metric Utility
80
70
60
50
Affective
40
Perceived Utility
Cognitive
30
20
10
0
0
2
4
6
8
10
12
14
16
18
20
Portion Size
100
90
80
70
Profit
60
50
Profit
40
Cognitive
Affective
30
20
10
0
0
2
4
6
8
10
Portion Size
12
14
16
18
20
Welfare: Regulating Context Cues
► By
increasing cognitive resources
 Profits decrease
 Consumers may still perceive they are better off
if surplus is shared
►More
likely the larger the shift in perception
► Consumer
will be better off if small change
► If instead, shifting the affective curve
 Profits may increase
 This may be more politically feasible
Regulating Context for Virtuous
Foods
► Increasing
cognitive resources and/or
reshaping the affective curve will
 Increase profits
 Increased consumer perceived welfare if surplus
is shared
 Thus this would be popular with actors
► Consumer
will be better off
Conclusion
► Ignoring
away
behavioral economics won’t make it go
 Even very traditional policies may interact with
behavioral cues
► Could
make policies ineffective or even self defeating
► Regulators
may be able to use behavioral cues to
create much more promising policy options
 Policies that potentially improve profits and welfare
 Such creative solutions have proven to be very effective
in some contexts
 Could also allow win/win solutions
 Need to know what we cannot observe