1v v,1_
“‘ ‘A $3,, m'a‘
.‘e \ 2;“?!
* 1
L‘. '
{av-33¢”
,1 ":,.,I .I
am. -
ur‘firflv (I:
r, .21: "m
m .
,, .. ..,
.‘.‘V‘,m"
‘,
P
7-: VJ}:
WWW 213.2%". M,
r ' , ‘w W?- «M? away;
“:33” m“. 3:34. {min ~. Wfii‘vfififi- 'x
. . ‘ .. "' '">?E=aw';wé1—5:u ~ ’4 ' '--'r . .V - 7-“
J. ' " ~ {I 5.1.x. . . 1.
~59“? 31.37.5ka ‘3 4" ~: 2'"
~: ‘um '4. '9‘
gauge} " -
‘
.... w
.‘l' ‘
$2
I ~
.v
WWiHHlIHHW|\\|\\|\‘\\H|U\lWWW
was 3 1293 01409
(_3'*.'%’7L"-
This is to certify that the
dissertation entitled
THE EFFECT OF AMBIGUITY ON CONSUMERS' WILLINGNESS
TO PAY FOR PESTICIDE-RESIDUE
CERTIFICATION ON APPLES
presented by
Jennifer Bard Wohl
has been accepted towards fulfillment
of the requirements for
Ph . D . degree in Agricultural Economics
<71/
Major professor
Date December 6, 1994
MSU is an Affirmative Action/Equal Opportunity Institution 0-12771
LIBRARY
Michigan State
Unlverslty
PLACE u RETURN BOXto man this checkout from your mom.
TO AVOID FINES Mum on or baton duo duo.
DATE DUE DATE DUE DATE DUE
WWI
THE EFFECT OF AMBIGUITY ON CONSUMERS’ WILLINGNESS TO PAY
FOR PESTICIDE-RESIDUE CERTIFICATION ON APPLES
By
Jennifer Bard Wohl
A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY
Department of Agricultural Economics
1994
ABSTRACT
THE EFFECT OF AMBIGUI'I‘Y ON CONsWflss TO PAY
FOR PESTICIDE-RasmuwiFICATION ON APPLES}
By
V's
.4
r;
‘C‘
Q
'\
Jennifer Bard Wohl
Consumers often fac ambigui about the health risks posed by pesticide
residues in the food they buy. Wty about the probability 1;;
of an outcome. It results because of the inherent difficulty in assessing the levels and
hazards of pesticide residues in food. Yet standard economic models of decision
making that are commonly used to assess choices involving risk assume that consumers
we“ 9M
can assign numerical probabilities to outcomes with total certainty.
This research develops a conceptual framework of consumers’ willingness to pa
W ._____. -h " If. .
I I O O O O I O O ‘f‘k
(WTP) for pestmde-resndue certification‘ on apples in the face of ambigmty about l‘lSkS. */
MW ,_.-. ,fi- , , M..,,va_w.—»w._...__~_
It uses Segal’s model of choice under ambiguity to show that the marginal value 3
reduction in ambiguity is positive and that willingness to pay for pesticide-residue}
W“ _..___/W I
,/
certification decreases with initialmErceptuions of ambiguin . ’
”mu.“ m ‘0'-
(“~_____—-- ‘ «- ~ “ fin..."-
The study uses the results of a contingent-valuation survey of Michigan
residents’ apple-purchasing behayior and attitudes about the risks from pesticide residues
“”M
\
/
, ,.-- . . a u... "0—”
J“
to test these findings. Heckman’s modeTL used to WWW
W
apples. The model asserts that the demand for certified apples is a two-stage process:
M-
the consumer first decides whether to buy certified or regular apples; he/she then
M.
u...— '—-—o—-~—u
decides how many apples to buy. ,6)
movfl.muwmvm*;fi..,.m."—M
_. mun-w..— w ._..,..,..-v.-v-—V"
' g".
.y ""
The probit results in the first stage indicate tha h initial Weity and
w ~
l / x‘
,1
I
changes in ambiguity associated with switching fro €ng cer‘ified applés are
\'.
important factors compelling consumers to choose certified over regular apples. j
The empirical results show that consumers are willing to pay a 32% premium on ea
WM‘Mrw-n—H ///.r
pound of apples that offers certification that the apples meet certain standards f0
pesticide residues.) This amount varies with perceived ambiguity about risks, suggesting
that consumers are interested in reducing not only the risks from pesticide residues but
W
\m. “J4- ‘0"
also the ambiguity about those risks.
W; certifying that food meets certain standards for
pesticide residues may have value by providing consumers with information about the
current risks they face. The conceptual results suggest that ambiguity about risks should
be incorporated into models of decision making under risk when the probabilities of
outcomes are not "crisp.”
Copyright by
Jennifer Bard Wohl
1994
To my parents
ACKNOWLEDGMENTS
This research was funded by the US EPA and the USDA under Cooperative
Agreement #43-3AEK-2-80039. Funding was also provided by the Michigan
Department of Agriculture and the Michigan Experiment Station. My sincere thanks to
Andy Manale of the US EPA for his support of this research.
I extend my deepest gratitude to my dissertation supervisor and genuine mentor,
Dr. Eileen van Ravenswaay. Her commitment to scholarly excellence profoundly
helped shape this research and facilitated its completion. This dissertation benefitted
tremendously from her keen insights in the area of food safety and from her hawk-eye
editing of numerous drafts (although I assume full responsibility for any remaining
errors).
I would also like to acknowledge the significant contribution of Dr. John l-Ioehn
in helping develop many of the concepts in this research. His willingness to discuss
ideas went far beyond the call of duty and he helped make the notion of ambiguity less
ambiguous.
vi
Many thanks to my other committee members: Dr. Ted Tomasi and Dr. Ching-
Fan Chung for their input at various stages of this research. Frank Lupi and Doug
Krieger also made valuable contributions to this research.
The many relationships I have developed while at MSU have enriched my
graduate studies and made doing this research more pleasurable. Always there along
the way were my good friends and colleagues Karin Stet’fens. Cynthia Donovan, Doug
Krieger, and Cynthia Phillips. They buoyed my spirits. eased my sorrows, and
applauded my accomplishments. Each had an uncanny instinct for knowing when to ask
and when not to ask how the dissertation was going and for identifying the right time
for a coffee break. My dissertation years were also graced by my "little sister," Aisha
Bhatti, who reminded me weekly of the necessity and joy of play.
The biggest thanks go to my parents. Audrey and Herbert Wohl, for their
unwavering confidence in me.
vii
TABLE OF CONTENTS
LIST OF TABLES ......................................... xi
LIST OF FIGURES ........................................ xii
CHAPTER 1: INTRODUCTION ............................... 1
Problem Statement .................................... 1
Objectives .......................................... 4
Illustration of Ambiguity ................................. 5
Definitions of Terms ................................... 6
Organization of Dissertation ............................... 13
CHAPTER 2: LITERATURE REVIEW ........................... 14
Models of Choice ..................................... 15
Model of Choice Under Risk ......................... 15
The Subjective Expected Utility Model .............. 15
Assumptions of the SEU Model .............. 16
Postulates of the SEU Model ................ 17
Failure of the SEU Model in the Presence of Ambiguity ........ 19
The Sure-Thing Principle and the Independence Axiom ......... ‘ 22
Models of Choice Under Ambiguity .......................... 23
Intuitive ...................................... 24
State-Dependent Utility Approach .................. 24
Degree of Confidence in Risk Estimate .............. 27
Data-Based ..................................... 35
Probability Weighting Approach ................... 35
Non-Linear Weighting of Second-Order Probability
Distributions .......................... 41
The Rank-Dependent Expected Utility Model (RDEU) ..... 42
Incorporating Ambiguity into the RDEU .................. 44
Measuring Ambiguity ................................... 50
Range ........................................ 50
The Variance of the Second-Order Probability Distribution ....... 51
Degree of Confidence .............................. 51
Conclusions and Observations .............................. 52
CHAPTER 3: CONCEPTUAL FRAMEWORK ...................... 54
Consumers’ Choice Problem .............................. 54
The Value of Certification When Risks are Ambiguous .............. 60
The Value of Certification ........................... 60\
The Effect of Changes in Risk on WTP for Certification ........ <6
The Marginal Value of Changes in Ambiguity ............... a
The Effect of Baseline Ambiguity on WTP for Certification ......
viii
p
The Effect of Baseline Risk 0n WTP for Certification .......... 68
Assumptions .................................... 68
Measuring Willingness to Pay for Certification ................... @
CHAPTER 4: SURVEY DESIGN AND DATA COLLECTION ............ 74
[I Use of Contingent Valuation Survey Methods fl
to Value Non-Market Goods ................ 74 ‘,
Design of CVM Survey ................................. 74 j
The Shopping Scenario ............................. 76 l
f Previous Survey ................................. 77
Modifications to Previous Survey ....................... 77
; Substitution Possibilities ........................ 78
1 Content of Certification ........................ 79
1" Measuring Risk Perceptions ...............................
Objective Versus Subjective Risk Estimates ................ 8O
Improvements in Risk Perception Questions ................ 81
Measuring Ambiguity Perception ............................ 83
Other Issues ......................................... 85
Prices ........................................ £5)
Implementation of the CVM Survey ........................... L486
Wf—Smey Methods ............................. 87
CHAPTER 5: ECONOMETRIC TECHNIQUES AND EMPIRICAL
ANALYSIS ......................................... 93
General Survey Results .................................. 93
Demographics ................................... 94
Risk Perceptions ................................. 96
Perceived Risk Reduction when Foods are Certified ........... 98
Perceptions of Ambiguity ............................ 98
Attitudes Toward Government and the Scientific Community ..... 100
Perceived Health Effects from Pesticide Residues in Food ....... 102
7 Purchase Intentions ............................... 104
Willingness to Pay for Certification .......................... 104
The Effect of Risk and Ambiguity on
Willingness to Pay for Certification .............. 106
the Demand for Certified Apples as a Two-Stage Process ............ 106
Estimating Consumer Surplus from the TOBIT Demand Equation ....... 110
Estimation .......................................... 112
Variable Definitions .................................... 113
Dependent Variables ............................... 113
Independent Variables .............................. 113
Testing the Hypotheses .................................. 116
Hypotheses ..................................... 116
Estimation Results ..................................... 118
Factors Influencing the Decision to Purchase Certified Apples ..... 118
Risk Reduction ............................. 118
Ambiguity Reduction .......................... 120
Baseline Risk .............................. 120
Baseline Ambiguity ........................... 122
Goodness of Fit ............................. 123
Type of Certification .......................... 124
,. Factors Influencing the Demand for Certified Apples .......... 125
g Willingness to Pay for Certification .......................... 127
Scenarios ...................................... 129
Conclusions ......................................... 131
CHAPTER 6: POLICY IMPLICATIONS AND CONCLUSIONS .......... 133
Conceptual Findings ................................... 133
Empirical Findings .................................... 136
Policy Implications .................................... 137
Survey Design Issues ................................... 139
Measurement Issues .................................... 140
Future Research Needs .................................. 142
APPENDIX 3-1: THE MARGINAL VALUE OF A CHANGE IN RISK ...... 146
APPENDIX 3-2: THE MARGINAL VALUE OF A CHANGE IN
AMBIGUITY ....................................... 147
APPENDIX 3-3: THE EFFECT OF BASELINE RISK ON WTP FOR
CERTIFACTION ..................................... 148
APPENDIX 5—1: SURVEY INSTRUMENT ......................... 149
APPENDIX 5-2: PROBIT RESULTS WITH DUMMY VARIABLE "CERT" . . . 163
BIBLIOGRAPHY ......................................... 164
Table 1-1:
Table 2-1:
Table 2-2:
Table 2-3:
Table 5-1:
Table 5-2:
Table 5-3:
Table 5-4:
Table 5-5:
Table 5-6:
Table 5-7:
Table 5-8:
Table 5-9:
Table 5-10:
Table 5-11:
Table 5-12:
Table 5-13:
Table 5-14:
Table 5-15:
Table A5-1:
LIST OF TABLES
Acts, States, and Outcomes in the Pesticide Residue Case Study . . . 8
Ellsberg’s Urn - First Choice Pair ...................... 21
Ellsberg’s Urn - Second Choice Pair ..................... 21
Ellsberg’s Decision Rule in the Context of the Pesticide Residue
Case Study ..................................... 31
Household Demographics ........................... 95
Household Income ................................ 95
Age of Respondent ................................ 96
Perceived Chance of a Health Problem ................... 97
Perceived Reduction in the Risks from Pesticide Residues When
Foods Meet Federal Standards for Pesticide Residues or Are
Produced Without Pesticides .......................... 99
Sureness about Health Risk .......................... «100
Attitudes Toward the Government and the Scientific Community . . . 101
Perceived Health Effects Associated with Pesticide Residues in
Food ........................................ 103
Willingness to Buy Certified and Uncertified Apples ........... 105
PROBIT Results ................................. 119
Second-Stage Demand Equation with Ambiguity measured as
e'(SURE) ...................................... 126
Willingness to Pay for Certification and Sensitivity Analysis ...... 128
Willingness to Pay for Certification ..................... 128
Changes in WTP due to Changes in Baseline Risk ............ 128
Changes in WTP for Certification due to Changes in Sureness ..... 129
PROBIT Results with Dummy Variable "CERT" ............. 163
xi
Figure 1-1:
Figure 1-2:
Figure 1-3:
Figure 3-1:
Figure 3-2:
LIST OF FIGURES
Definitions of uncertainty, risk. ambiguity. and ignorance ....... 10
FOP probability distribution over outcomes, x (or states) when
there is no ambiguity .............................. 11
SOP probability distribution over outcome, x (or states) when the
probability of each outcome is ambiguous ................. 12
SOP for high initial ambiguity consumer (consumer A) and low
initial ambiguity consumer (consumer B) .................. 67
CV for Certification on Apples ........................ 71
xii
CHAPTER I: INTRODUCTION
Problem Statement
Consumers have become increasingly concerned about the potential health effects
of pesticide residues in food (Food Marketing Institute 1990). Part of the concern stems
from the fact that when making food choices, consumers often face ambiguity about the
health risks posed by pesticide residues. Ambiguity is defined as uncertainty about the
probability of an outcome.‘ It results because of the inherent difficulty in assessing the
levels and hazards of pesticide residues in food. Since pesticide residues are
undetectable in food and since food is generally not labeled for its residue levels,
consumers do not know the exact levels of pesticide residues they ingest. Furthermore,
consumers receive conflicting information about the potential health effects of pesticide
residues, and they may not trust that policies regarding allowable levels of residues are
adequate or are being strictly enforced (van Ravenswaay 1992).
To design policies that effectively address consumers’ concerns, policy makers
should understand how consumers make choices in the face of potential risks from
pesticide residues. However, standard economic models of consumer choice under risk
that are commonly used to assess such choices (e.g., Jones-Lee 1974) assume that
decision makers are able to accurately assess the probabilities of the potential outcomes
associated with a choice (i.e. perceptions about risks are ”crisp” and people can assign
numerical probabilities to outcomes with total certainty). These models predict that
‘ In this research, ambiguity refers to only the uncertainty about the probability of
an outcome; it assumes there is no uncertainty about the types of potential outcomes.
1
2
consumers’ willingness to pay for a reduction in risk increases with both the initial, or
”baseline,” risk and the magnitude of the risk reduction.
Contrary to these theoretical predictions, however, van Ravenswaay and Hoehn
(1991) found empirically that in the case of pesticide residues in food, perceptions of
both initial risks and risk reductions explained very little of the variation in the
willingness to pay for risk reduction. They used the results of a contingent valuation
survey to estimate consumers’ willingness to pay for apples that had been tested and
certified to meet certain standards of pesticide residues. While risk perceptions were
statistically significant in explaining apple purchases, people who perceived large initial
risks from pesticide residues were willing to pay about the same amount for apples
certified for pesticide residue levels as people who perceived small initial risks.
These findings motivated the present research which investigates whether
consumers’ willingness to pay for pesticide-residue certification is affected by the
uncertainty or "ambiguity" they perceive about the risks they face. The study uses the
case of pesticide residues in apples to explore the hypothesis that ambiguity about risks
isfla’nfi important determinant of consumers’ willingness to pay for residue certification.
Segal’s (1987) model of choice which incorporates ambiguity into a generalized
expected utility model is used to develop hypotheses about (1) consumers’ WTP for
(residue certification, (2) the effect of perceived initial or baseline risk and ambiguity on
willingness to pay for certification, and (3) the effect of the perceived reductions in risk
and ambiguity associated with certification on WTP for certification.
These hypotheses were tested using the results of a contingentwygjgatiogw--.
W
telephone survey conducted in Michigan in June/Julylgig. The survey asked Michigan
N7
3
residents about their apple-purchasing behavior in both actual and hypothetical
situations. It also solicited their perceptions of the current risks from pesticide residues
in food and their "sureness" about those risks. As is explained in chapter 4, these
measures of “sureness” serve as proxies for a measure of ambiguity.
The survey considered two hypothetical food-labeling policies that potentially
change consumers’ perception of the risks posed by pesticide residues in apples. One
policy permits certification and labeling of apples that have been produced without
pesticides. The other policy permits certification and labeling of apples that meet
federal standards for pesticide residues SurveyWsWa...--
H.2—
hypothetical shopping scenario arflfgverewaskedmtg..preqmgthsiaapmmmwW -...
“ «Ma-W" .
behavior: Respondents were then asked how their apple purchases and risk perceptions
fl...»
would change if they could choose between regular apples and apples that had been
tested and certified to meet one of these standards for pesticide residues.
The data from the survey were used to evaluate the factors affecting the decision
to buy certified apples and the demand for certified apples. The parameters from the
demand equations were used to estimate consumers’ willingness to pay for certification,
the effects of baseline risk and ambiguity on willingness to pay for certification, and the
effects of reductions in risks and ambiguity on WTP for certification.
If ambiguity about risks is an important determinant of consumers’ WTP for
certification, polices to reduce ambiguity may have value. Consumers may value
knowing the standards for pesticide residues in food, for example, and what those
standards mean. Similarly, information about what is being done to enforce the
standards may be beneficial to consumers. This information helps consumers better
4
assess the risks they face from pesticide residues without necessarily changing the ac
risk that consumers face. Furthermore, a better understanding of the role of ambiguity
in consumer choice may give us insight as to why consumers are willing to pay for
certified apples when initial risks (i.e., the perceived risk associated with the regular
apple) and risk reductions are perceived to be small. It may also have implications for
the interpretation of existing estimates of WTP for reduced mortality and morbidity
which assume no ambiguity.
r" J
Objegaé
,/
The objectives of this research can be summarized as
follows:
1. Weonceptual framework ref willingnesrto ~pay_for_.pesuc1de:residue—-\
cert1 1cation that incorporates perceptions about both risk and ambigui about
MWS model of choice, which incorporatesé®y
into the Rank-Dependent Expected Utility (RDEU) model;
“hm-uni"-
2. Use the conceptual framework to develop hypotheses about the effect of
”baseline” or "initial” risk and W on the willingness to pay for
certification, and the effect of risk and ambiguity reduction on willingness to pay
for c cat1on;\
3. Empirically test the hypotheses using the results of a contingent valuation survey
that asks respondents about their apple purchasing behavior and about their risk
and ambiguity perceptions.
The rest of this chapter illustrates the importance of ambiguity and presents some
concepts and definitions of terms that will be used throughout the thesis. The chapter
concludes with a presentation of the organization of the rest of the dissertation.
5
Illustration of Ambiguity
Ellsberg (1961) illustrates a situation of ambiguity using the following example.
Suppose you have agreed to participate in an experiment in which one ball will be
chosen from one of two urns, both containing 100 balls. In urn I there are 100 balls,
but you do not know the proportion of red and black balls. In um 11 there are 50 red
and 50 black balls. You indicate which color you prefer to bet on (red or black) and
which urn you want the ball to be drawn from. If a ball of your color is chosen, you
win $100, otherwise you receive nothing. If you are like most of the subjects in
Ellsberg’s original experiment, you will be indifferent between betting on a red ball or
betting on a black ball, but once the color is chosen (be it red or black) you would
prefer to have a ball drawn from um 11 (the unambiguous urn) than from um I (the
ambiguous urn). However, if all distributions of red and black balls are equally likely
in the ambiguous urn,2 the expected probability of drawing either color ball is 0.5.
Since the probability of drawing a red or black ball in the unambiguous ball is also 0.5,
you should be indifferent between the urns. That you are not indifferent about the urns
suggests that the uncertainty about the probability of drawing a particular ball in the
ambiguous urn may affect your choice of urn.
2 All distributions would be equally likely if you used the "Principle of Insufficient
Reason" which states that "if there is no evidence leading one to believe that one event
from an exhaustive set of mutually exclusive events is more likely to occur than
another, then the events should be judged equally probable" (Luce and Raiffa 1988, 48).
6
Definitions of Terms
This section establishes the terminology used throughout the rest the dissertation
when discussing consumer choice under uncertainty.
In this research, the term "uncertainty" refers to a situation in which one does
not know for certain what the outcome or consequence of an action will be. In
Savage’s (1954) framework for decision making under uncertainty, the decision maker
chooses between ”acts” (also called "lotteries" or "alternatives").3 These are
probabilistic choices available to the decision maker. Nature is said to choose among
”states of the world" (also called "states of nature" or ”states"). Each act/state
combination yields an "outcome" or "consequence" from which the decision maker
derives utility. When making a choice, the decision maker does not know which state
of the world nature will choose, and thus chooses an act under uncertainty. The
decision maker assesses the probability of each possible state of the world (and thus the
probability of each outcome). Each act, then, has associated with it a probability
distribution over outcomes. The outcome is thus a random variable with a probability
distribution.“
The states of nature are usually assumed to "form a mutually exclusive and
exhaustive listing of those aspects of nature which are relevant to this particular choice
problem" (Luce and Raiffa 1988). The states of the world are considered independent
’ The terms ”act,“ ”action,” and ”lottery” are used interchangeably throughout the
text.
‘ For a given ”act," the probability distribution over "states" is identical to the
probability distribution over "outcomes." One can therefore talk about the probability
distribution over either without a change in meaning.
7
of the act chosen. The probabilities of the states of the world "rain" and "no rain," for
example, are independent of the acts ”carry an umbrella" and "do not carry an
umbrella." If the probabilities of the states depend on the act chosen one may
alternatively assign to states ”conditional" probabilities (Jeffrey 1983) which depend
upon which act is chosen.
This study presented respondents with a hypothetical market in which there were
available both regular apples (the kind they usually buy) and apples that had been tested
and certified by the federal government to meet federal standards for pesticide residues
(or, for half the sample, to have been grown without pesticides). Respondents were
asked to indicate whether they would select the regular apples, the tested apples, some
of both, or none at all. The choice of apple type is considered an act. The states of
nature are (1) ”experience an adverse health outcome someday because of pesticide
residues in apples” and (2) ”do not experience an adverse health outcome someday
because of pesticide residues in apples.” The probabilities of the states of nature are
”conditional" upon which type of apple is chosen. The outcome depends on which type
of apple was chosen, as well as which state of nature prevails. The choice scenario for
apples is depicted diagrammatically in Table 1-1.
5 The state of nature could alternatively be characterized as (1) pesticide residues
are safe and (2) pesticide residues are not safe. This depiction of the states conforms to
the Savage (1954) framework in which the probabilities of the states are immutable and
do not depend on the act chosen. With this representation, however, one cannot take
actions (buying certified apples, for example) to reduce the risks from pesticide
residues. To develop a framework for the WTP for certification that depends on the
probability of experiencing an adverse health outcome someday because of pesticide
residues, this research uses "conditional" probabilities that depend on which type of
apple one chooses.
.m 2:“: 8:: x 8: 88. 88:8: 8888: u mx
88.,
88888 5.8: 288: 8:::
8: 88 .8.3: .8: 88:88 8: 88 «x
888.8 5.8: 3.88:
8::: .8.3: 88:88 8: 88 H":
1?. n 4x. 88:: 3.3
88888 5.8: 38.8: 8::: 8:
88 .8.3: 8 83. :8: 88:88 ax
n
.h u 4x 88:: an:
8888: 5.8: 88.8: 8:::
.8.3: .8 83. :8: 88:88 ”Mu
m.-. u .x 8.5 8::
88888 5.8: 3.88:
8::: 8: 88 .8.3: 828. 88:88 ”M:
ah n ax 88:: A5.—
88880 5.8: 88.8:
8::: .8.3: 828: 888:8 H m:
«:7— H Q 88:: 3?.
as u «x 88:: A53
88888 5.8: 888:
8::». 8: 88 .8.3: we: 888:8 ”mg
" III-.1]
88888 5.8: 888:
8::: .8.3: we: 888:8 H":
1..-. u .x 88:: A5:
.h n .X 88:: 83:
8.3: .9: 8: 88 ”4x
8.3:
8 83. 5e: 8 888 .3: ”ax
8.3: 828. 3:0 .8: ”fix
1
8.3: 5:8: :8 .9: . 8x
0882.6 :3.
8.3: :. 88.8.. 88888.. 8.3: :. 88.8.. 88888:
.8 8:88: 5:888 8888: .8 8:80: .8888 0888:
5.8: 38.8: :: 88.898 “um 5.8: 388: :: 88:38 ”a
85m
88m 8:0 38.81 088883 05 :_ 88880 8:: .885 .8<
”7— 2:5.
9
Most of the literature on risk discusses choices between "acts" such as those
presented in Table 1-1 as choices among "lotteries;" that convention will be used here.
The following notation will be used to represent such lotteries: X,=(x},1r{;
x§,w§;...x§,,1rf,) is a one-stage lottery (or act) called Xi that offers the prize x}, with
probability it}, x; with probability 1;, up to x}, with probability mi.“ The text later uses
a two-stage lottery to introduce ambiguity into the decision calculus. A two-stage
lottery is a lottery in which the prizes are tickets to another lottery; it is represented by
A=(X,,q‘; X2,q2;...X,.,q'”). The symbol "1r" is used to denote the probabilities in one-
stage lotteries; is used in two-stage lotteries to denote the probabilities of winning
cl
tickets to a one-stage lottery. Capital letters are used to denote lotteries; lower-case
letters are used to denote prizes or final consumption goods.
”Risk," "ambiguity,” and ”ignorance" fall under the general rubric of
"uncertainty" and are distinguished by the degree to which one can accurately specify
the probability distribution over states (Einhorn and Hogarth 1985).
"Risk" refers to uncertainty about which outcome (or state) will prevail. There
is no uncertainty about the probabilities of outcomes (or states). Decisions under risk
can be described by one-stage lotteries. If one knows with certainty the probability that
pesticide residues in apples cause adverse health outcomes, the example in Table 1-1 is
considered a decision under "risk."
° If there are only two possible outcomes to a lottery the subscript denoting the
outcome is often dropped; the superscript denoting the lottery number is dropped when
the discussion concerns only one lottery or a general set of lotteries.
10
”Ignorance" refers to a situation in which one has no knowledge about the
probability distribution over states.
"Ambiguity" refers to uncertainty about the probabilities of outcomes (or states).
When there is ambiguity, the probability of each state is itself a random variable. Each
point on the probability distribution over states then has its own probability distribution
over probabilities (called "second-order probability distribution," or SOP) (Marschak
1975). For example, food-consumption choices when pesticide residues may be present
are made under ambiguity when one does not know whether an adverse health effect
will result and one cannot assign probabilities to health states with certainty. Ambiguity
can alternatively be described as a continuum, with risk at one extreme and ignorance at
the other. A case of zero ambiguity would be considered "risk"; complete ambiguity
would be considered "ignorance. "
The definitions of ”uncertainty," "risk," "ambiguity," and "ignorance" (adapted
from Einhom and Hogarth 1985) are shown schematically in Figure 1-1.
Uncertainty: one does
not know what the
outcome of an action
will be
risk: one knows with
certainty the probability
distribution over
outcomes
ambiguity: one does not
know with certainty the
probability distribution
over outcomes
ignorance: one has no
information about the
probability distribution
over outcomes
Figure 1-1:
Definitions of uncertainty, risk, ambiguity, and ignorance
11
Figure 1-2 below shows a continuous first-order probability distribution (FOP -
represented by f(x)) over outcomes, x (or over states) when there is no ambiguity about
the probability of each state of the world. Each point on this distribution is a point
estimate of the probability of a particular state (objectively or subjectively determined).
In Figure 1-3 each point on the FOP distribution in Figure 1-2 has its own second-order
probability (SOP) distribution (represented by g[f(x)]). The amount of ambiguity is the
spread, or variance, of the SOP distribution. The next chapter shows how the spread of
the SOP distributions of Figure 1-3 can be accounted for in a model of choice under
uncertainty.
Figure 1-2: FOP probability distribution over outcomes, x (or states) when there is
no ambiguity ‘
12
Figure 1-3: SOP probability distribution over outcome, x (or states) when the
probability of each outcome is ambiguous
The concept of the probability of an outcome is itself subject to interpretation.
There are two main types of probability:
1)
2)
Objectivistic (or frequency): The probability of a given outcome is considered
a property of that outcome. This property can be determined by repeated
experimentation. The probability is the proportion of the experiments that result
in the outcome. The probability of a coin toss landing heads, for example, may
be determined objectively by performing an experiment in which the coin is
tossed numerous times. One may then objectively estimate the probability of
heads as 0.5.
Subjectivistic (or personal): This view holds that probability is a subjective set
of beliefs about risk which may or may not be based on a frequency
interpretation of probability. Two individuals faced with the same evidence may
state divergent probabilities, both of which may be reasonable. One individual
may hold the subjective view that the probability of rain tomorrow is 0.9, while
another may estimate the probability as 0.7; both probabilities are reasonable.
Our research maintains that people have subjective views about the objective
probability (or frequency) of an outcome. People also have a certain degree of
confidence in their estimate. If, for example, someone saw a coin tossed twice and it
l3
landed heads once and tails once, he/she would estimate that the probability of that coin
landing heads was 0.5. Similarly, if he/she saw the coin tossed 1000 times and the
result was 500 heads and 500 tails, he/she would again assign 0.5 probability that the
flip of the coin would result in heads. However, he/she would have more "confidence"
in the second estimate than in the first and his/her SOP distribution would have a
smaller variance.
Organization of Dissertation
The rest of the dissertation is organized as follows. chapter 2 reviews the
subjective expected utility (SEU) model, the assumptions and postulates undergirding it,
and the failure of the model to account for ambiguity. It then presents several models
of choice that incorporate ambiguity. The purpose of the literature review is to
highlight the importance of ambiguity in decision making and to evaluate the various
models that incorporate ambiguity in order to select the best model. Chapter 3 presents
the conceptual framework used in this study to derive the hypotheses about consumers’
willingness to pay for residue certification, the effect of baseline risk and ambiguity on
WTP for certification, and the effect of reductions in risk and ambiguity on WTP for
certification. It also discusses how demand analysis can be used to test the hypotheses.
Chapter 4 discusses the survey used to measure consumers’ apple purchasing behavior
and the measures of risk and ambiguity used. Chapter 5 presents some descriptive
statistics of the survey results. It then discusses the empirical model used to test the
hypotheses and an presents an . analysis of the results. The summary and policy
implications are presented in chapter 6.
CHAPTER 2: LITERATURE REVIEW
The expected utility (EU) model as axiomatized by von Neumann and
Morgenstern (1947) and the subjective expected utility model (SEU) as axiomatized by
Savage (1954) are normative models of decision making under risk. The expected
utility model asserts that if a decision maker conforms to a set of axioms that govern
"rational" behavior, then his/her choices among alternative acts can be ranked according
to their expected utility. The decision maker should choose the act that results in the
highest expected utility, where the weights attached to utility in various states of the
world are the objective probabilities of those states. In the SEU model, decision makers
do not use ”objective” probabilities to weight the utilities. but rather use subjective
beliefs about the likelihood of various states.7
This chapter first reviews the SEU model and the postulates that undergird it. It
then demonstrates how the model fails when probabilities are ambiguous. The chapter
reviews several models of choice that explicitly accommodate ambiguity. These models
can be divided into three types: (1) intuitive models that offer non-axiomatized
alternatives to expected utility, (2) data-based models that weight the expected
probability of an outcome to account for ambiguity, and (3) theoretical models that
weight the entire second-order probability distribution (SOP) over outcomes non-linearly
by relaxing one or more of the axioms of SEU (usually the "sure-thing" principle, or
the independence axiom in EU).
7 The rest of the discussion in this section uses the subjective expected utility (SEU)
model when demonstrating the importance of ambiguity. However. most of the
discussion applies to the expected utility model as well.
14
15
The purpose of this chapter is to demonstrate the importance of ambiguity in
decision making under uncertainty and to select a theoretically sound model of choice
that accommodates ambiguity. Segal’s model of decision making under ambiguity is
described at the end of the chapter; this model is used in chapter 3 to develop
hypotheses about consumers’ willingness to pay (WTP) for pesticide-residue certification
on apples and the effect of risk and ambiguity on WTP.
Models of Choice
Model of Choice Under Risk
'v i ' M l
The subjective expected utility model offers a normative framework of how
people should rank "acts" in an uncertain world. Subjective expected utility for a given
act is the weighted average of the utility of the possible outcomes (or consequences)
associated with that act in different states of the world, where the weights are the
subjective probabilities of the various states occurring. Given an individual’s
preferences over consequences, the individual should choose the act that has the highest
expected utility.8
Consider the simple example of a decision maker deciding between two acts (X.,
X2). Nature will choose one of three possible states (5,, 52, or 5,). Each act/state
’ Hirschleifer and Riley (1992) make the "crucial distinction" between the utility of
consequences and the utility of actions. Utility attaches directly to consequences but
only derivatively to actions. This study maintains this distinction and follows their
notation of u(x) to represent a person’s preference-scaling function (or elementary-utility
function) over the consequences, x; U(X) will be used to represent the person’s derived
preference ordering over acts, X.
16
combination yields an outcome: (x1, 11%, xg) are the consequences associated with act Xl
under states 1, 2, and 3; (xi, x3, xi) are the consequences associated with act X2 under
states 1, 2, and 3. The decision maker’s subjective beliefs about the unconditional
likelihood of states s,, 5;, and 53 are represented by 1,, «2, and «3. The decision maker
calculates the following:
senor.) = 1,1104) + 12ml) + «311(8)
SEU(X2) = 1r,u(xf) + «2u(x§) + 1r,u(x§)
He/she then chooses the act with the highest subjective expected utility.
When the states of nature are continuous between 1 and n, the outcomes
associated with act i are continuous between x: and x,‘,. The individual then chooses the
highest among:
SEU(X,)=f1t,u(x,‘)dx (2-1)
.r-l
SEU(X2)= f «311(42): (2-2)
3-1
Assumptions of the SEU Model
The SEU model is valid only under certain assumptions about the decision
maker’s behavior. The model requires that decision makers adhere to certain axioms or
postulates of behavior that lead to these behavioral assumptions. This section reviews
17
those assumptions and postulates. The section following demonstrates situations in
which it may be inappropriate to apply them to our decision maker.
The assumptions of the subjective expected utility model are listed as Al through
A4 below (adapted from Gardenfors and Sahlin 1988).
A1:
A2:
A3:
A4:
The values of the outcomes in a decision situation are determined by a utility
measure which assigns numerical values to the outcomes.
When determining the value of a decision alternative the only information about
the decision maker’s wants and desires that is exploited is the utilities of the
possible outcomes of the alternatives, i.e., factors other than the utilities of the
outcomes - such as risk involved - do not influence the value of the alternative.
A decision maker’s beliefs about the states of the world in a given situation can
be represented by a unique probability measure defined over states. For each
state s the decision maker thus assigns a probability value P(s) so that the sum of
all these values taken over the set S is 1. These probabilities are subjective. ‘
For all states and all alternatives, the probability of the state is independent of
the act chosen.
The foundations for these assumptions are Savage’s (1954) axioms or rationality
postulates P1 through P4 below (adapted from Ellsberg 1961).
Postulates of the SEU Model
P1.
There is a complete ordering of gambles or "actions" (either Act 1 is preferred
to Act 2, Act 2 is preferred to Act 1, or one is indifferent between Act 1 and
Act 2).
The choice between two actions must be unaffected by the value of payoffs
corresponding to states for which both actions have the same payoff. (Sure Thing
Principle)
18
P3. Knowledge of an event9 cannot establish a new preference among consequences
or reverse an old one, and no preference among consequences can be reduced to
indifference by knowledge of an event.lo
P4. The choice of which state a person prefers to stake a prize should not be affected
by the size of the prize.
These axioms justify the use of the SEU model as a framework for decision
making under uncertainty. Further, A3 justifies reducing all decisions under uncertainty
to decisions under risk. The subjective expected utility model assumes decision makers’
beliefs about the likelihood of states of the world can be represented by a single
probability estimate. Savage himself was aware of the problem ”ambiguity" might pose
to the SEU model as demonstrated in the following passage:
...there seem to be some probability relations about which
we feel relatively ”sure" as compared with others. When
our opinions, as reflected in real or envisaged action, are
inconsistent, we sacrifice the unsure opinions to the sure
ones. The notion of "sure" and "unsure" introduced here
is vague, and my complaint is precisely that neither the
theory of personal probability, as it is developed in this
book, nor any other device known to me renders the
notion less vague (Savage 1954, 57-58).
9 An event is a set of states. An event can also be a set of one state, in which case
a state and an event are the same thing.
‘° This postulate simply helps define what is an "act" and what is a "consequence.”
The example given by Savage (1954, 25) is that of a person going on a picnic that will
be held at either the beach or near tennis courts. If the ”consequences" are the
possession of a tennis racket or the possession of a bathing suit, clearly the preference
over "consequences” would depend on where the picnic were held. P3 stipulates that
the "consequences" should instead be defined as "enjoy a swim,” or "enjoy a game of
tennis,” etc.
19
Failure of the SEU Model in the Presence of Ambiguity
Ellsberg (1961) demonstrates that in situations of high ambiguity, people do not
conform to the Savage postulates (P1 and P2 in particular); the subjective expected
utility model is then invalidated. Ellsberg (1961) defines ambiguity as the degree of
confidence one has in his/her estimate of relative likelihoods of outcomes.
Ellsberg’s (1961) um example described in chapter 1 demonstrates a situation in
which the Savage postulates are violated. Reconsider the two urns: In um I, there are
100 red and black balls in a proportion unknown to you. In urn 11 there are exactly 50
red balls and 50 black balls. Ellsberg then asks the following questions:
1. ”Which do you prefer to bet on , Red, or Black,: or are you indifferent?” That
is, drawing a ball from Um I, on which "event" do you prefer the $100 stake,
red or black: or do you care?
2. "Which would you prefer to bet on, Red.I or Black.,?"
3. ”Which do you prefer to bet on, Redl or Redn?”
4. ”Which do you prefer to bet on, Blackl or Blacku?"
If you are like most people, you will be indifferent in both the first and the
second case. Yet once the color is determined (be it red or black) you would prefer to
bet on the unambiguous um (um 11).“ This pattern of choice violates Savage’s first
postulate that there is a complete ordering of gambles or "actions” and it demonstrates
" This example falls under the "economics of uncertainty,” where one must make a
terminal decision, rather than under the "economics of information” in which the
decision maker has the option of paying to acquire more information. For more on this
distinction, see Hirschleifer and Riley (1992).
20
that personal probabilities cannot always be determined by observing choices among
gambles (as is implied by Ramsey (1931) and Savage (1954)).
If you prefer to bet on Red,, rather than Red, (i.e., you are not indifferent),
assumption A3 above leads us to conclude that you must view Red,, as more likely than
Red, (i.e., P(Redn) >P(Red,)).l2 Similarly, if you prefer to bet on Black" over Black,
this should reflect your belief that Black,, is more probable than Black, (i.e., P(Blacku)
> P(Black,)). But if the latter holds then it must also be true that P(not Red,,) > P(not
Red,), which is a contradiction to the former. It is thus impossible to infer meaningful,
unique probabilities from your choices and you are violating Savage’s first postulate
since there is no clear ordering over preferences. The justification for using the
subjective expected utility framework is thus in doubt.
Ellsberg (1961) also presents an example that shows how the sure-thing axiom
(P2 above) is violated. Suppose you are presented with two choice pairs. In the first
choice pair, there is an urn filled with 30 red balls and 60 yellow and black balls in a
proportion unknown to you. One ball will be chosen from the urn. You may bet that
the chosen ball will be red or you may bet that it will black. The prize will be $100 for
a correct prediction. The payoffs corresponding to each choice are presented in Table
2-1. Would you prefer to bet on red (choice I) or on black (choice 11)?
‘2 A3 suggests that one can always assign a unique probability to a state of the
world. One can thus always compare the probability of one state to the probability of
another.
21
Table 2-1: Ellsberg’s Urn - First Choice Pair
[W =
Payoff
Choice Red (N =30) Black (0 s N s 60) Yellow (0 s N 560)
I. Bet on Red $100 $0 $0
II. Bet on Black $0 $100 $0
3 = NumEr of balls r
In the second choice pair, you must choose between betting on (red or yellow)
and betting on (red or black). This choice and its payoffs are depicted in Table 2-2.
Table 2-2: Ellsberg’s Urn - Second Choice Pair
Choice Red (N =30) Black (0 s N 560) Yellow (0 s N s 60)
III. Bet on (Red or $100 $0
Yellow)
IV. Bet on (Black or $0 $100
Yellow)
Most people prefer Choice I to Choice 11, but prefer Choice IV to Choice 111.
This pattern of choice violates the Sure-Thing Principle (P2 above). In the first choice
scenario, the payoff is the same for both actions if a yellow ball is drawn (payoff of
$0). The Sure-Thing Principle says that this payoff should therefore be ignored.”
'3 Savage (1954, 21) gives the following example to demonstrate the "sure-thing
principlez' ”A businessman contemplates buying a certain piece of property. He
considers the outcome of the next presidential election relevant to the attractiveness of
the purchase. So, to clarify the matter for himself, he asks whether he would buy if he
knew that the Republican candidate were going to win, and decides that he would do so.
Similarly, he considers whether he would buy if he knew that the Democratic candidate
were going to win, and again finds that he would do so. Seeing that he would buy in
. ——
a = “1.11115! OI Balls
22
Similarly in the second choice scenario, the payoff for yellow is the same for both
actions ($100 if yellow is drawn). This payoff should also be ignored. Ignoring the
payoff under yellow, the payoffs of the two choice situations are identical. According
to the Sure-Thing Principle, if you prefer choice I, you should also prefer choice III.
The Sure-Thing Principle and the Independence Axiom
The sure-thing principle can also be stated as follows: "Suppose that we are
considering two actions A (bet on red) and B (bet on black) and that these will have the
same outcome ($0 payoff) if E (yellow ball drawn) occurs, but will yield different
outcomes when E does not occur. Suppose that A is preferred to B. Now consider two
acts A' (bet on [red or yellow]) and 8' (bet on [black or yellow]) such that
- A' yields the same outcome as A provided E does not occur
- 8' yields the same outcome as B provided E does not occur
- A' and B' yield the same outcome whenever E occurs.
Savage argues that if A is preferred to B, we should also prefer A' to B’
(Quiggin 1993, 9). When choosing between acts, we should ignore states in which acts
yield the same outcome.
The sure thing principle is analogous to the independence axiom in EU which
states that if for acts X and Y, XzY, then for any number r€[0,1] and any act 2,
rX+(1-r)Z z rY+(l-r)Z. Preferences between two lotteries should be independent of
any common components .
either event, he decides that he should buy, even though he does not know which event
obtains.”
23
The independence axiom (sure-thing principle) is the cornerstone of the Expected
Utility (Subjective Expected Utility) theory. This axiom restricts the preference
functional to be "linear in the probabilities.” That is, it must be representable as the
mathematical expectation of some von Neumann-Morgenstern utility index defined over
the set of pure outcomes” (Machina 1988, 216).
However, as just shown, the independence axiom is commonly violated in
situations of ambiguity. Without the independence axiom, the preference functional is
not restricted to being linear in the probabilities. Several of the models that have been
developed to accommodate situations of ambiguity are not linear in the probabilities.
The next section reviews several of these models as well as other models of choice
under ambiguity.
Models of Choice Under Ambiguity
There has been a plethora of empirical studies over the last twenty five years
demonstrating that people prefer a choice situation in which they have more information
regarding the probabilities of states to one in which little is known about the
probabilities of states, even if the mean probability of states in the two choices is the
same.“ In other words, pe0ple prefer situations of low ambiguity to situations of high
ambiguity. Yet in their traditional form, the EU and SEU cannot accommodate
ambiguity because probabilities about probabilities are averaged out. Instead, both EU
" See Camerer and Weber (1992) for a review of several empirical studies of
ambiguity.
24
and SEU use point estimates of probabilities even though the decision maker may not
know the probability distribution over states with certainty.
Ever since Ellsberg’s (1961) seminal paper, several researchers have tried to
incorporate ambiguity into models of decision making under risk. There have been
three types of models: 1) non-axiomatic intuitive models, 2) models based on data
generated by laboratory experiments, and 3) theoretical models that weight the second-
order probability distribution non linearly. This section reviews some of the approaches
taken in each category of model.
Intuitive Models
5 42 l I! 'l' E I
Expected utility theory assumes that the utility derived from a consequence is
independent of the state in which the consequence occurs. State-dependent utility
theory, on the other hand, allows utility to depend on the state in which the consequence
occurs. Although this approach violates A2 above, it has been used to account for the
effect of ambiguity on decision making. Smith (1969) proves that in the Ellsberg (1961)
um example, preferences are consistent when the probabilities are the same for the two
urns, provided that the utilities of payoffs in the two urns are different. He says "the
utility of money or other rewards is not independent of the circumstances under which it
is obtained. The utilities in the payoff matrix may have arguments other than what
appears to be the ’objective’ reward" (Smith 1969, 325). In other words, ambiguity
about'the probability of states may influence the utility of the consequences obtained in
the states.
25
Consider the case of apple consumption. An apple eater may derive utility from
not only the quantity of apples consumed, but also from knowing that the pesticide
residues on those apples will not cause adverse health effects someday. The probability
of such an occurrence as well as the ambiguity about that probability may be direct
arguments in the utility function.
Winlder (1991) also argues that the effects of ambiguity operate through the
decision maker’s preferences rather than through the probabilities. Ambiguity may
increase the feelings of “regret" or "blame” if a poor outcome results (such as adverse
health effects from pesticide residues), and satisfaction or joy if a good outcome results.
This justifies an expanded consequence space that may include attributes (such as risk or
ambiguity) other than the monetary payoff received.
Sarin and Winkler (1992) propose a model that permits a decision maker’s
preferences to depend on the ambiguity about the probability of an event. Consider a
simple lottery X, that yields a monetary payoff x, in state 1 and x2 in state 2. Assume
that x, >x2, the assessed probability of state 1 is 1r, and the assessed probability of state 2
is 1-1r. The expected utility of this lottery is then:
U(X)=1ru(x,) +(1 -1r)u(x2) (2'3)
where 11 represents the decision maker’s von Neumann-Morgenstern utility function.
If 1 is ambiguous then the utilities are replaced by modified utilities, V:
V[u(x,)|u(x2)] and V[u(x2)|u(x,)].“ The modified utility depends of the degree of
‘5 The capital letter V represents the value function when the value of the lottery is
not represented by its expected utility.
26
ambiguity present in the probabilities of each state. As x, approaches x2, V[u(x,)|u(x2)]
approaches u(x,) and V[u(x,)|u(x,)] approaches u(x,). Although the modified utilities
could be appropriate in non-ambiguous situations as well, the authors argue that the
sense of elation or regret "would seem to be heightened and especially salient when
ambiguity is present" (Sarin and Winkler 1992, 394).
The general approach of accounting for the effects of ambiguity by expanding
the consequence space of utilities is similar to "regret theory" developed by Loomes and
Sudgen (1982). Regret theory compares what is received with what could have been
received if a different action had been taken but the same state of the world obtained.
Sarin and Winkler compare what is received with what could have been received if the
same action was chosen but a different state of the world obtained.
State—dependent utilities are not unique to situations of ambiguity. Non-
ambiguous risk can also lead to feelings of regret or the assignment of credit or blame.
This approach is thus appr0priate for situations of both risk and ambiguity.
Camerer and Weber (1992) argue that the choice of whether to model decision
making under ambiguity by adjusting the probabilities or by altering the preferences is a
matter of taste. "The choice between the two approaches turns crucially on whether one
believes likelihood estimates and decision weights must be equal. Those who advocate
modifying utilities are reluctant to let likelihood and decision weights differ (e.g.,
Winkler 1991). Those who are willing to allow a difference may find utility
modification cumbersome or tautological" (Camerer and Weber 1992, 343).
27
r f nfi n 'nRik im
Ellsberg (1961), Gardenfors and Sahlin (1982, 1983), and Levi (1974, 1982)
present models of choice that account for ambiguity by assigning estimates of
"confidence" to probability distributions over outcomes. Ellsberg (1961) demonstrates
with his urn examples described above that there are some situations in which people do
not act according to the Savage axioms and the justification for using expected utility
framework disintegrates. Ellsberg (1961) does not suggest abandoning the Savage
Axioms; he only suggests that there are some situations to which they do not apply.
His thesis is that one must account for the "ambiguity of information" which is "a
quality depending on the amount, type, reliability and ’unanimity’ of information, and
giving rise to one’s degree of ’confidence’ in an estimate of relative likelihoods"
(Ellsberg 1961, 657). Although Ellsberg (1961) does not make the explicit analogy
himself, higher confidence corresponds to a tighter SOP.l6 According to Ellsberg
(1961), the level of confidence in a probability distribution over outcomes is a function
of the ”ambiguity of information." This research is concerned primarily with the level
of ambiguity rather than with the sources of ambiguity. It thus concentrates on
Ellsberg’s (1961) “confidence” and not on the ”ambiguity of information."
Ellsberg’s (1961) model of choice accommodates ambiguity by including a
measure of the degree of confidence in the probability distribution over states as a
“ To be consistent with the distinction between "risk," "ambiguity," and
"ignorance“ made in Chapter 1, if ambiguity is to be measured by the degree of
confidence one has in the probability estimate of a state, it would have to exclude the
endpoints. That is, "total confidence“ would be a situation of risk, and "total lack of
confidence” would be a situation of ignorance.
28
parameter in the choice model. Let p (0
1r for small r) and large probabilities to be
underweighted (w(1)w(r°)+w(r‘) then there is a sufficiently small 6 such that
2‘ This model is used here only for purposes of illustration. The same result would
follow from any of the simple probability weighting models described in this section.
41
w(1r"+1r‘)x > w(1r")(x+5) +w(7r‘)x (2‘9)
Lottery X, will result in a higher utility level than lottery X2, even though X2
stochastically dominates X,.
As Weber and Camerer (1987) point out, a descriptive theory should be able to
account for violations in stochastic dominance but not predict them in a wide variety of
situations.
Theoretical Models
Although the probability models just described assume a SOP around the
probabilities of outcomes, they apply weights to only the expected probability (13(1)) of
the SOP to account for ambiguity. As such, they lead to violations of stochastic
dominance and can lead to sub- or super-certainty. Another class of models applies
non-linear decision weights to the entire SOP. These models are able to account for
ambiguity without violating first-order stochastic dominance.
Kahn and Sarin (1988) present a variant of the probability weighting model for
predicting consumers’ choices under ambiguity in which the probability of an outcome
is a function of the entire SOP. If the probability density function (SOP) for 1r is 4>(1r),
the decision weight is defined as follows:
1
w(n)=E(n)+ f (n -E(n))el-*>VO¢(n)dn (2-10)
:00
42
where:
E(1r) = the expected value of the probability of an outcome
¢(1r) = the probability density function of 1r
A = reflects an individual’s attitude toward ambiguity in a given context
and
(2-11)
I
o= f (n-E(n))’¢(n)dn
V s-O
is the standard deviation of the random variable a.
This model departs from the SEU by assigning a non-linear decision weight to
the probability of an outcome. If there is no ambiguity then w(1r)=E(1r) and the model
reduces to the SEU. Similarly, if the individual is ambiguity neutral (A=O), then again
w(1r)=E(1r) and the model reduces to SEU. This model differs from the probability
weighting models discussed in the previous section in that the decision weight w(r) is a
function of the entire second-order probability distribution. ¢(1r), and not just of the
mean of the distribution.
II B l-D I E ”1.1. H HIBDEIII
The Rank-Dependent Expected Utility Model (RDEU), developed by Quiggin
(1982), uses a nonlinear weight of the entire probability distribution over outcomes
(although not over probabilities of outcomes). Segal (1987) modifies this model by
allowing for nonlinear weighting of the SOP.
Because the RDEU is the basis for Segal’s (1987) approach to modeling decision
making under ambiguity, it will be reviewed here. Without Segal‘s modification,
43
however, the RDEU does not explicitly accommodate ambiguity. RDEU is an
extension of the simple probability weighting model (described above) to multiple
outcomes. It is a generalized utility model that maintains several of the features of the
expected utility model (transitivity, completeness, continuity, and first-order stochastic
dominance). The model was developed to help explain empirical violations to expected
utility such as Allais’ Paradox, and the systematic overweighting of small probabilities
and undervveigthing of large probabilities.
As in the probability weighting models. the RDEU uses "decision weights"
instead of objective probabilities of outcomes. In the RDEU, however, the decision
weight of the probability (ark) of outcome x, depends on the probabilities (1r, V s=1...n,
s¢k) of the other possible outcomes (x, V s=1...n, s¢k) and the rank, or desirability,
of outcome x. relative to all other x,. If the lottery X offers n outcomes (i.e., outcomes
x. where s=1...n), and the outcomes can be ranked in order of preference (x, < ...x, <
...x,), the RDEU takes the following form:
RDEU = V(X) = ih’(n)u(x‘) (2'12)
.r-l
where:
8 3-1
’33“) =fl31t] -fl2n}) (2-13)
j-l 1-1
1' = the vector of probabilities (1,,rz,...1r,,) of other outcomes
44
and the function f ”transforms" the probability (which always lies between 0 and 1) such
that the transformed probability remains between 0 and I (i.e.. f[0.l| » [0,1]). It is
assumed that f(0) =0 and f(1) = 1.
The decision maker chooses the act X that has the highest RDEU.
Incorporating Ambiguity into the RDEU
Although the RDEU allows for the over- and underweighting of the probabilities
of outcomes, it does not account explicitly for ambiguity. Segal incorporates ambiguity
into the RDEU model by assuming that choices involving ambiguous probabilities are
like two-stage lotteries. The first-stage lottery is over the probability of an outcome; .the
second stage is the lottery over outcomes, using the result‘of the first stage as the
relevant probability. This is consistent with the notion that ambiguity can be
conceptualized as a second-order probability distribution (the probability distribution
over probabilities).
For our apple consumer, for example, the first-stage lottery would be a lottery
over the probability that an adverse outcome would result someday because of pesticide
residues in food; the second-stage lottery would be the lottery over the health outcome,
using the result of the first-stage lottery as the probability.
Assume a decision maker faces the following two-outcome lottery:
X = (x,1t; 0,1-1t) (2'14)
Now assume that 1r is itself a random variable 17‘ where g=1...m (m is the total
number of possible probabilities of a given outcome). The probability that the actual
45
probability of outcome x is 1r‘ is q‘. That is, the probability that the lottery one will
face is actually (x, a"; 0, 1-1r') is q‘. The different lotteries that the individual might
face are the following:22
X,: (x,1r'; 0,1-1r‘) where P(x') = q'
X2: (x,1'2; 0,1-1'2) where P(wz) = q2
Xm: (x,1r"‘; 0,1-1r‘“) where P(1r'“) = qm
The standard approach to evaluating this lottery is to use the Reduction of
Compound Lotteries Axiom (RCLA)2’:
The Reduction of Compound Lotteries Axiom (for two outcomes):
Let X, be a one-stage two-outcome lottery: X,=(x‘, a"; 0, I-w‘), i = 1,...m. Let A be a
two—stage lottery in which the possible prizes are tickets to the one-stage lotteries above
(X,). Define R(A) as the one-stage lottery that is the actuarial equivalent of the
combination of X, and A (i.e., R(A) is derived by combining the probability of winning
a ticket to the lottery with the probability of getting the prize once the lottery is played).
R(A) = (Lil‘q‘; 0,1-51441‘). By definition, this reduces to the same lottery as
hi (I!
2’ Reminder: in this lottery, there are only two outcomes, x and 0. The probability
of outcome x may take one of m possible values, however. That is, the probability of
receiving outcome x may take any one of 1|"...1r'“ values.
2’ For simplicity, the RCLA is stated only for the two-outcome case. For the
general version of the RCLA, see Segal (1990).
‘ ' "Ft. i'o‘qa-rfiw'ucaivgflozi 74:5. :7
::. '73. £071 :'«nv:.‘w.-'-.r_f. 1w
46
equation 2-14. The decision maker is thus indifferent between the two-stage lottery A
and the one-stage lottery R(A).“
Even in the presence of ambiguity, if one uses the RCLA in processing multiple
layers of risk, ambiguity falls out of the model and is behaviorally insignificant.
Segal assumes that instead of using the RCLA, decision makers adhere to the
Compound Independence Axiom (CIA) which is stated as follows:
Compound Independence Axiom: Let A be a two-stage lottery: A=(X,,q';...X,,,,q“‘)
and define CE(X,) by (CE(X,),1)~X,.25 This notation says the decision maker is
indifferent between getting the certainty equivalent of the X, lottery (CE(X,)) and facing
the lottery X,. The decision maker is then also indifferent between facing lottery A and
facing the lottery ((CE(X,),1),q‘;...(CE(X,,,),l),q"‘).26 Consider our apple consumer.
He/she might be indifferent, for example, between a situation in which the probability
of experiencing an adverse health outcome because of pesticide residues is somewhere
between 0.1 and 0.00001, with an average probability of 0.050005. and the situation in
which he/she knows for certain that the probability is 0.07.
2‘ The same conclusion would be reached if there were more than two outcomes in
the one-stage lottery.
2’ The lottery (CE(X,),1) is a degenerate lottery in which the decision maker
receives CE(X,) with certainty.
2‘ The RCLA and the CIA together imply the Mixture Independence Axiom (MIA)
which states that for every one-stage lottery X, Y, and Z, and for a 6 [0,1], XzY if
and only if (xx + (l-a) Z a aY + (1-oz)Z. The MIA implies that the preference
functional is linear in its probabilities. By dropping the RCLA but maintaining the CIA,
we retain the spirit of the independence axiom while rejecting the assumption that a two-
stage lottery is equally as desirable as its one-stage actuarial equivalent.
47
If there is a "certainty equivalent" for each first-stage lottery above then:
uICE(X,)] = [1m] (2-15)
Solving equation 2-15 for CE(X,) yields:
CE(X,) = u"[V(X,.)] (2-16)
Now substitute u-l [V00] into the RDEU in which there are only two outcomes (x
and 0) to the lottery Xi and assume u(0)=0. This yields:”28
RDEUWA = V(X) = u(x,M)f(1tl)
+ u(IM)‘22[7(n‘) -f'0) and he/she faces an initial probability
1° (0< 1°< 1) of death during the current period, his/her expected utility is:
Etw=<1-u')L(W°)+n°D(W°) (3'9)
where L(W°) and D(W°) denote, respectively, L(W) and D(W) evaluated at W=W°.
If the individual is offered the opportunity to reduce the probability of death
during the current period from 1° to 1‘ (where 1‘ (1°), he/she should be willing to pay
some amount to face the new probability. This amount is the compensating variation
(CV) and is defined as the amount of money that will leave the decision maker with the
3’ This type of model is also used by other researchers. See, for example, Viscusi,
Magat, and Huber (1987). Smith and Desvouges (1987) amend the model by
considering a two-stage process in which the individual has a risk, R, of being exposed
to hazardous wastes during the time horizon described by the model and, if exposed, a
separate risk, q, of premature death from the exposure.
‘° This model can also be used to model the demand for risk reduction when the
”states of the world” are ”no health outcome" and "health outcome," as in this research.
To be consistent with Jones-Lee (1974), however, the presentation of the model will use
the states "life" and "death.”
62
same level of expected utility as in the initial situation. CV will thus satisfy the
following equation:
(1 -1t‘)L(W°-CV)+1:‘D(W°-CV) =(1 -1t")L(W")+1t°D(W")
(3-10)
CV is the Hicksian compensating variation in wealth for a change in probability from 1°
to 1‘.
The compensating variation in the case ‘of ambiguous risks is the amount of
income the decision maker is willing to give up in exchange for a reduction in
probability from 1' to 1‘ and a reduction in the ambiguity from 6' to 6°. That is, CV is
the amount the decision maker is willing to pay for the certification. It satisfies the
following equation:
v(15,247.17! - CV)? (1‘ + 7‘s“)
+ ..p,.p..M- Caz-1%; + W) for + wecnfitql)
j-i
v(p,.1r2,.p...M)? (1’ + We?
+ ,.p,.p,.M)§2 [hr + re) 4% + we'nfigql)
(3-1 1)
From this model, we deveIOp hypotheses about the effect of baseline and
changes in risk and ambiguity on willingness to pay for residue certification.
63
The Effect of Changes in Risk on WTP for Certification
The change in WTP for certification due to a change in mean risk is represented
by -(dCV/d1°).“ This expression evaluated at 1‘=1‘ is the marginal value of an
decrease in risk from the initial level (Jones-Lee 1974).“2 We use equation 3-11 to
find this expression. If consumers are ambiguity averse (i.e., the transformation
function, f, is concave)“, and if ambiguity is unaffected by changes in the mean risk
and the marginal utility of income is positive, equation 3—11 can be used to show that in
situations of ambiguous risks:
_ acv
ar
> 0 (3-12)
1’
where [2.2] indicates that the expression is evaluated at 1'.
61‘ _,
We thus have the following hypothesis:
Hypothesis 1. The marginal value of a decrease in the mean probability of an adverse
health outcome (when ambiguity present) is positive
" In this particular case 61" = 1'-1°.
‘2 Note that this is the marginal value of a decrease in risk and is therefore
positive. By contrast (dCV/da") is the marginal value of an increase in risk (Jones-Lee
1974).
‘3 Given the plethora of empirical results demonstrating that in most situations
people do not like ambiguity, and given the serious nature of the possible health effects
from pesticide residues, these are reasonable assumptions. There are some situations,
however, in which people systematically prefer ambiguity to no ambiguity. These are
situations characterized by very small probabilities of very large gains. See Segal
(1987, 187) for examples. The assumption of ambiguity aversion should be explored
further.
64
This calculation is the shown in Appendix 3-1. As in the case when risks are not
ambiguous, the amount consumers are willing to pay to reduce risks should increase
with the amount of risk reduction they obtain (regardless of the level of ambiguity).
The Marginal Value of Changes in Ambiguity
Segal (1987) shows using the RDEUWA that a less ambiguous lottery has a
higher value (in terms of the RDEUWA) than a more ambiguous lottery. He also
shows that a non-ambiguous lottery has more value than an ambiguous one.“ Using
the general formulation for utility, this can also be written as:
“0 = “py’pc!pr9foer9M) < v(py'pc!pr’f‘éc‘M) = “1
(3-13)
where:
e° 0 (3'15)
If Segal’s assumptions hold, and the value of a less ambiguous lottery is higher
than a more ambiguous one, then the willingness to pay for a marginal reduction in
66
ambiguity from 5' to e° must be positive. Consumers should thus be willing to pay some
amount above what they would pay to get risk reduction only to get ambiguity
reduction. The second hypothesis of this research is:
> 0 (3-16)
_ dCV
66"
"
where [ aCV] indicates that the expression is evaluated at 6'.
de‘ G,
Hypothesis 2. The marginal value of a decrease in ambiguity is positive.
This hypothesis can also be developed by using equation 3-11 to solve for -(6CV/de°).
This is shown in Appendix 3-2.
The Effect of Baseline Ambiguity on WTP for Certification
Figure 3-1 shows the SOP distribution of an adverse health outcome from
pesticide residues for certified and regular apples for two consumers (consumer A and
consumer B). The consumers perceive the same baseline risk (1’) and the same amount
of risk reduction (1'-1‘) when switching from the regular apples to the certified apples.
However, consumer A has a higher level of initial ambiguity than consumer B. The
amount of risk reduction consumer A obtains when switching from regular to certified
apples is uncertain because of the large variance in the SOP distribution. The risk faced
after moving to the certified apple when risks are highly ambiguous may not be
significantly different from the risk associated with the regular apple. The motivation to
move to certified apples is then gone. The risk reduction for consumer B, on the other
67
f(z)
eammerA
: 4
tc 1" fl
f(n)
constnnerB
' l
‘36 “r 1:
Figure 3-1: SOP for high initial ambiguity consumer (consumer A) and low initial
ambiguity consumer (consumer B)
hand, is clear. Although the mean risk reduction is the same in both cases, the
consumer with the high ambiguity should not be willing to pay as much for the
certification as the low ambiguity consumer.“ The following hypothesis results:
Hypothesis 3. The WTP for certification decreases with initial ambiguity, holding
ambiguity reduction, baseline risk, and risk reduction constant.
‘5 This scenario assumes there is no ambiguity reduction when switching from
regular to certified apples. It looks only at the effect of baseline ambiguity on WTP for
certification.
68
This hypothesis implies that consumers who perceive higher ambiguity from the
regular apple should be willing to pay less for certification than those people who
perceive lower ambiguity, all else held constant.
The Effect of Baseline Risk on WTP for Certification
Jones-Lee (1974) shows that the marginal value of risk reduction increases with
initial (or baseline) risk, 1'. He determines this by differentiating the expression for the
marginal value of a change in risk with respect to initial risk level. The finding in the
case of ambiguous risks is that consumers who perceive higher baseline risks from the
regular apple should be willing to pay more for certification than people who perceive
lower baseline risks.46 That is,
- :61
61"
61’
a (3-17)
">O
Hypothesis 4. The WTP for certification increases with initial risk.
Assumptions
In order to use the RDEUWA model to develop the above hypotheses, it is
necessary to make several assumptions. This section outlines the assumptions used.
This research assumes that there are only two possible states of the world: an
adverse health outcome does result from pesticide residues, and an adverse health
‘6 This calculation is shown in Appendix 3-3.
69
outcome does not result from pesticide residues.“7 The utility and indirect utility
associated with no adverse health outcome are normalized to 0, i.e., unh(-)=0 and
v,,,()=0,“ If the health effects from pesticide residues have a negative effect on
utility, then utility and indirect utility in the state of the world in which one experiences
an adverse health effect from pesticide residues (u,,(-) and v,(-)) are negative.
Although the restrictions outlined in Chapter 2 regarding the bounds on 7‘s still
hold, 7‘ does not vary across decision makers. The parameter 7‘ varies with the
possible probabilities, but does not vary by individual. The scale of 7‘ is such that
whatever the scale of 6, 7‘5 will be between 0 and l. The term 6, then, is not restricted
to lie between any bounds.
The approach here also assumes that q‘, the probability that the j"‘ probability is
the actual probability, is the same for all decision makers.
People may take actions (averting behavior) to affect the probability of
experiencing a health problem someday because of pesticide residues in food. They
may wash or peel their fruit and vegetables before eating them, for example.“9 These
actions may be complements or substitutes to acquiring the certification, i.e., people
‘7 Pesticide residues may result in several types of health outcomes with different
degrees of severity. However, to keep the analysis simple, we consider only two health
outcomes (none and any). The model can be modified to accommodate several (or
continuous) health outcomes.
“ u,,,() is utility in the state of the world in which no adverse health outcome from
pesticide residues results and v,,,(-) is the indirect utility in the same state. u,() is utility
in the state of the world in which an adverse health outcome from pesticide residues
occurs and v.0 is the indirect utility in the same state.
‘9 There are several other actions pe0ple might take. They might grow their own
produce, or buy organic produce. ,
70
may continue to take the actions they usually do (actions and certification are then
complements) or they may discontinue taking those actions because they perceive the
certified apple offers the reduction in risk they were achieving by their actions (actions
and certification are substitutes).so The questions developed in the survey about the
risks from pesticide residues are all framed in terms of pre-averting behavior. The
probabilities obtained from the survey are thus all pre-averting behavior.
Measuring Willingness to Pay for Certification
Freeman (1993) shows that the CV for a change in the price of a good is equal
to the area to the left of the Hicks-compensated demand curve between the initial and
new prices. In this research, when a consumer switches from the regular apples to the
certified apples he/she moves from a situation in which the price of the certified apples
is prohibitively high (or, if the certified apples are not available, the price of certified
apples is effectively infinite) to a situation in which the price of the certified apples is
such that consumption of certified apples is positive. The CV for the certification is
then the area to the left of the Hicks-compensated demand curve (h°) between the price
at which demand for certified apples is zero (the "choke” price) and the price at which
the certified apples are offered. This is a measure of the surplus from the certified
5° This research does not develop hypotheses about the relationship between risk
reduction acquired through consumers’ own actions and risk reduction acquired through
buying certified rather than regular apples (i.e., it does not develop hypotheses about
whether these actions are substitutes or complements). It does, however, measure
empirically the relationship between the two. This is discussed further in Chapter 5.
71
apples that is due strictly to the presence of the certification indicating the pesticide
residue level. This is shown graphically in Figure 3-2.
who
Penan-
x“
(certified apple.)
Figure 3-2: CV for Certification on Apples
When a consumer chooses the regular apple at a market price, he/she also
receives some surplus; this surplus is due to the “appleness” of the regular apple that is
worth something to the consumer above the market price. Since the regular and
certified apples are identical except for their price and certification, when the consumer
chooses the certified apple, he/she could have consumed the regular apple and still
gotten the 'appleness." The surplus associated with the certified apple is thus due only
to the certification and not due to its "appleness."
The CV can also be written in terms of the expenditure function, i.e., it is the
difference in the expenditure required to keep the individual at the initial level of utility,
u°, when the price changes from pi to p1. Thus:
72
CV = e(p,.2,1r‘,e‘,u") - e .,.‘,1r",e‘,u ") (3'18)
where p}, is the price of certified apples when the consumer chooses regular apples (i.e.,
p: is the ”choke” price or the price at which the demand for certified apples is zero),
and pi is the market price of certified apples at which there is positive demand for the
certified apples.
Since CV is defined as the difference between two levels of expenditure at two
different prices, it can also be written as (Freeman 1993):
P:
p.‘
d ’1‘. ‘3 "
I e0). 6 u )6”). = Jhcwcnl’maw’) dpc (3-19)
P: a _
dp‘ P:
This measure of the welfare associated with the certified apple is inconvenient
because the Hicks-compensated demand curve upon which the true CV is based is
unobservable. While the ordinary demand curve upon which the Marshallian consumer
surplus is based is observable, the Marshallian consumer surplus is flawed as a welfare
measure as it assumes constant marginal utility of income. The question then is can we
use the Marshallian consumer surplus as close approximation to the true welfare
measure? Willig (1976) found that the differences between the Marshallian measure and
other welfare measures, such as the CV, are likely to be small. He says ”the error will
often be overshadowed by the errors involved in estimating the demand curve" (W illig
1976, 589). Freeman (1993, 61) also states that "the differences among the measures
appear to be small and almost trivial for most realistic cases. "
73
Several methods have been developed for estimating exact welfare measures
using either a Taylor’s series expansion of the indirect utility function based on terms
that are derivatives of the ordinary demand functions, or by specifying a flexible
functional form for the indirect or direct utility function (Reaume 1973; McKenzie and
Pearce 1976, 1982; Mckenzie 1976, 1983).SI Freeman points out that the Taylor’s
series expansion approach “has been criticized as being cumbersome...potentially
inexact because successive terms of a convergent Taylor’s series are not necessarily
monotonically decreasing...and unnecessary” (Freeman 1993, 67).
Because the issue of the appropriate approach to measuring welfare has not been
resolved in the literature, this research uses the consumer surplus based on the ordinary
demand curve as a measure of the welfare associated with the certified apple.
The next chapter describes the survey used in this research to test the hypotheses
developed in this chapter. Chapter 5 presents the econometric techniques used to test
the hypotheses and presents both descriptive and analytical results.
5‘ These approaches are reviewed in Freeman (1993).
CHAPTER 4: SURVEY DESIGN AND DATA COLLECTION
Use of Contingent Valuation Survey Methods
to Value Non-Market Goods52
This chapter describes the design of the contingent valuation survey (CVM) used
to test the hypotheses developed in chapter 3. The first section describes the design of
the CVM survey used in this research. It then presents the methods for measuring both
risk and ambiguity. The third section describes how the CVM survey was
implemented. The concluding section uses Carson’s (1991) validity criteria to evaluate
the major strengths and weaknesses of the research design.
Design of CVM Survey
Contingent valuation survey methods (CVM) offer a technique for eliciting the
value consumers’ attribute to non-market goods or to characteristics of goods. In
contingent-valuation surveys, respondents are asked questions about hypothetical
scenarios; the value they place on the non-market good or characteristic can then be
determined. Respondents may be asked to directly state their willingness to pay for the
non-market good, or they may be asked about the purchase decisions they would make
at various prices and about their perceptions of the characteristics (such as pesticide
residue levels) of the good.”"‘ In the latter case, the data can be used estimate the
’2 The rest of this chapter draws heavily from van Ravenswaay and Wohl (1993)
and from van Ravenswaay, et al (1992).
’3 This later approach is technically called "contingent behavior" rather than
”contingent valuation" (Freeman 1993).
74
75
demand for the good as a function of the non-market characteristics of the good. Shifts
in the demand due to changes in the product attributes are then used to estimate the
willingness to pay for changes in the attributes.
This research uses a contingent valuation survey (presented in Appendix 5-1) to
measure consumers’ purchase decisions under alternative policy scenarios. It also elicits
consumers’ risk and ambiguity perceptions. The results of the survey are used to
estimate the value to consumers of two pesticide residue policies that reduce risk and/or
ambiguity. The first policy (Policy I) maintains the current federal standard for
pesticide residues on food, but permits product claims to be made that the federal
standard is met. This policy may not change the actual risks that consumers face from
pesticide residues (if the apples already met federal standards), but it may change
peoples’ perceptions of both the risks they face and the uncertainty or ambiguity about
those risks. The second policy (Policy II) permits certification that products have been
produced without pesticides. This policy reduces risks to consumers and may also
change peoples’ perceptions about the risk and ambiguity from pesticide residues. In
both scenarios, the policy may reduce the ambiguity about risks from pesticide residues
by simply providing information about the levels of pesticide residues in apples.
The constructed market for apples under these two policy scenarios was
presented to consumers using a contingent valuation survey instead of an experimental
setting in which participants are offered actual products. This approach permits the use
“ The pesticide residue level on apples is considered a "characteristic" of apples;
risk and ambiguity are based on consumers’ perceptions of residue levels, but are not
themselves characteristics of the good.
76
of a large and representative sample at lower cost than the equivalent experimental
setting.
The contingent valuation survey used in this research had several goals that
influenced its design. One goal was to create a hypothetical shopping scenario for
respondents that would prompt them to reveal their apple-purchasing behavior given
certain prices of apples and levels of pesticide residues. Another goal was to elicit
respondents’ perceptions of the risks they face from pesticide residues and how those
risks change with residue levels. A final goal was to examine peoples’ uncertainty
about risks to test whether ambiguity is a factor in determining willingness to pay for
pesticide residue reduction.
The Shopping Scenario
The survey created a realistic shopping scenario by asking people to imagine that
they were shopping in the fall and planning to buy some apples.” This guaranteed
that all respondents were thinking about the same shopping scenario.
The survey did not ask people to directly state their willingness to pay for
reduced pesticide residues or reduced risks since this is not a situation they are ever
likely to face. Instead, it asked them about their apple-purchasing behavior under
various scenarios. These are decisions people actually face when they buy apples; the
responses are thus more likely to reflect real decision behavior.
’5 Respondents were first asked if their household had bought any fresh apples in
the past year. The hypothetical shopping scenario was presented only to the ninety two
percent of respondents who indicated they had.
77
Previous Survey
The questionnaire used in this study built upon the questionnaire developed by
van Ravenswaay and Hoehn (1991) to estimate consumer willingness to pay for reduced
risks from pesticide residues in apples. Since it used many of the same questions
developed by van Ravenswaay and Hoehn, this section briefly describes the
development of their questionnaire."5
The questionnaire used in the van Ravenswaay and Hoehn study asked
respondents about their current purchases of apples. their purchase intentions for regular
apples at specified prices, and their purchase intentions for apples that were said to be
certified to have "no pesticide residues," ”no detectable pesticide residues," and finally
”no pesticide residues above federal limits.” Respondents were given a range of
different prices and asked how many apples they would buy if they were planning to
buy apples on a typical shopping occasion in the fall. The season was specified so that
all respondents would be considering similar supply conditions.
Modifications to Previous Survey
In the present research several modifications were made to the van Ravenswaay
and Hoehn (1991) survey. First, this survey was conducted by telephone rather than by
mail. Since van Ravenswaay and Hoehn were interested in studying the levels of pest
Wage that consumers would accept, their survey required that respondents react to
photographs of apples with pest damage; a mail survey was thus necessary. In the
5‘ For more details of the van Ravenswaay and Hoehn survey, see either van
Ravenswaay and Hoehn (1991) or van Ravenswaay and Wohl (1993).
78
present research, however, no variations in visual aspects of apples were necessary; a
phone survey was thus suitable.
A telephone survey offers several advantages over mail surveys. It allows the
use of Random Digit Dialing (RDD), which generates more representative samples, it
reduces non-response bias since interviewers are able to clarify questions and prompt
respondents for further information, complicated skip patterns can be easily incorporated
into the questionnaire design, and, since the telephone survey is computerized, data are
entered directly into the computer as respondents answer questions, thus minimizing
processing errors.
i ' i ' ' i
As in the van Ravenswaay and Hoehn (1991) questionnaire, the purchase
intention questions in our survey were developed to reveal the quantity of apples
respondents would likely buy at different prices during a typical grocery shopping
occasion in the fall. However, in this survey respondents were given a choice between
certified and uncertified apples, whereas in the van Ravenswaay and Hoehn (1991)
questionnaire, apples were either all certified or all uncertified. Our modification
allowed the substitution between different ”brands" to be explicitly accounted for. In
the van Ravenswaay and Hoehn (1991) questionnaire, the consumer could substitute
another fruit for apples, but the price and type of that fruit were unknown. The survey
used here assumed that the closest substitute to an uncertified apple was an apple
certified for the level of pesticide residues. It therefore offered respondents both
products and specified their prices (prices varied acress households). As in the van
79
Ravenswaay and Hoehn (1991) survey, respondents were told to assume that all fruits
other than apples were not certified for pesticide residue levels. The two studies
represent two different policy alternatives. The van Ravenswaay and Hoehn (1991)
study sought to evaluate a change in residue standards; the present study evaluates
residue-labeling policies.
' l i n
The certification on the apples was also changed. The survey omitted the "no
detectable residues" certification since van Ravenswaay and Hoehn (1991) found it was
statistically indistinguishable from the certification "meets federal standards for pesticide
residues." The certification "no pesticide residues" was changed to "produced without
pesticides” since it is technically impossible to guarantee "no pesticide residues" when
pesticides may be airborne, waterborne, or soilbome at the farm where they were
grown, the warehouse where they were stored, or the store where they are sold.
In order to make the questions shorter and more tractable. half the sample was
asked to make a choice between regular apples (no certification) and apples with the
certification "produced without pesticides;” the other half was asked to choose between
regular apples and apples with the certification "no pesticide residues above federal
limits." The type of certification was randomly assigned to respondents based on the
last digit of the respondents’ phone number.
80
Measuring Risk Perceptions
Objective Versus Subjective Risk Estimates
In order to test the hypotheses developed in chapter 3. the study estimates the
demand for certified apples as a function of both risk and ambiguity. Market data does
not allow this, however, as it is difficult to know the levels of pesticide residues present
in the apples people buy; and it is difficult to know peoples’ perceptions of the levels of
pesticides present in the apples they buy. Some researchers have applied ”objective” or
"scientific" estimates of the risks from pesticide residues to data on the demand for
organic products to assess consumers’ willingness to pay for reduced residues and
reduced risks (I-Iammitt 1986). However, focus groups with consumers conducted. by
van Ravenswaay and Hoehn (1991) found that organic labels on produce were not
interpreted in a consistent fashion by consumers. Consumers buy organic products for
several reasons, only one of which is the reduced pesticide residues they offer. For
example, some consumers believe organic products have higher nutrient value and are
more tasty than conventional products. Furthermore, both the risks and ambiguity about
risks from pesticide residues are "subjective," i.e., individuals vary in their perceptions
of the risks they face from the same product. When subjective estimates of risk diverge
significantly from the ”objective" or "scientific” estimates (Kahneman and Tversky
1979, Eom 1994), assuming that all consumers use the "objective" estimate of the risks
when making purchase decisions distorts the true willingness to pay for risk reduction.
Since consumers make decisions based on their own assessments of the riskiness of
products, and not on scientists’ estimates (of which they are generally not aware), it is
important to understand peoples’ subjective risk estimates.
inpro
ask ;
C0111
red;
“1K
fit
01’
81
Improvements in Risk Perception Questions
To develop measures of both baseline risk and changes in risks. several
improvements were made to the van Ravenswaay and Hoehn (1991) methods of eliciting
risk perceptions. The van Ravenswaay and Hoehn (1991) study found that there was
considerable variation across households in the actions households undertook to avoid
pesticide residues. One third of their sample reported that they washed fresh produce
with soap and water, bought organic produce, or grew their own produce in order to
reduce or avoid pesticide residues in food. To ensure that each household was
comparing certified apples to uncertified apples rather than to apples that might have
reduced pesticide residues because of one’s own actions, the survey asked respondents
to estimate the risks from pesticide residues to a person from a household like theirs
who did nothing at all to avoid or reduce pesticide residues in food.
Another improvement in the risk perception questions was the increased
specificity in the meaning of quantitative risk. In the van Ravenswaay and Hoehn
(1991) survey, respondents were asked what the chances were that someone from their
household would have a health problem someday because of pesticide residues in food.
To ensure that respondents to the present survey were considering therisk to a person
in the whole population, the survey asked respondents in this survey to imagine that
there were a million people from households like theirs who did nothing at all to reduce
or avoid pesticide residues in food. It then asked them what they thought the chances
were that a person from one of these households would have a health problem someday
because of pesticide residues in food. To help ensure the respondents were estimating
population risks, the response categories used in the van Ravenswaay and Hoehn (1991)
82
survey were modified to specify numbers of people out of one million who would be
expected to have a health problem someday because of pesticide residues in food. They
could choose from the response categories: one in a million. one in 100,000, one in
10,000, one in 1,000, one in 100, one in 10, one in 5. one in 2. certain to happen.
Responses to this question measure perceptions of baseline risk.
A third improvement to the risk perception questions was that respondents were
asked for a qualitative statement of the risks from pesticide residues before being asked
for a quantitative estimate. Respondents were asked whether they would say there is
"no chance," "it is very unlikely,” ”somewhat unlikely,” "somewhat likely," "very
likely," or "certain to happen" that someone from a household like theirs would have a
health problem someday because of pesticide residues in food. The interviewer used
this qualitative answer to prompt the respondent for a quantitative estimate of risks.
This approach permits an examination of the correlation between a qualitative, but more
easily understood estimate, and a quantitative, but more precise estimate. In fact, the
two measures are highly correlated.
To measure perceptions about changes in risk, the survey also asked respondents
how much (in percentage terms) their risks would be reduced if the federal government
tested and certified that the apples met federal standards for pesticide residues (or, for
half the sample, that apples had been produced without pesticides). This information is
used to calculate the perceived risks associated with the certified apples.
The risk perception questions were all in terms of the chance of a health problem
someday because of pesticide residues in food. The responses thus represent
respondents’ perceptions of the risks from pesticide residues in all foods, not only the
83
risks from residues in apples. Pretests done for the van Ravenswaay and Hoehn study
(1991) revealed that this was the easiest way for people to communicate their
perceptions of the risks from pesticide residues; expecting people to accurately assess
the risks due specifically to pesticide residues on apples is unrealistic. The
questionnaire was thus designed to facilitate eliciting accurate measures of peoples’ risk
perceptions.
The quantitative measures of the perceptions of risk from all foods (both baseline
and changes) are included as an explanatory variables in the estimated demand model in
the form obtained from the survey, i.e., as lifetime risks from all foods.’7 This is
appropriate if one assumes that the risks from apples are a constant proportion of the
risks from all sources (van Ravenswaay and Hoehn 1991). However, when using the
model for sensitivity analysis to see how a marginal change in risks affects willingness
to pay for certification, we adjust the willingness to pay for certification to calculate
annual willingness to pay a change in annual risk.
Measuring Ambiguity Perception
The conceptual framework developed in chapter 3 shows that ambiguity, defined
as the “spread” of the second-order probability distribution, is theoretically an important
determinant of the willingness to pay for residue certification. The problem arises as to
how to measure ambiguity without knowing the shape of the respondents’ second—order
probability distribution. Since ambiguity perceptions cannot be directly measured,
‘7 The Heckman two-stage model that is used to estimate the demand for certified
apples is described in detail in Chapter 5.
84
several proxies were developed to reveal respondents’ underlying perceptions of the
spread of their second-order probability distribution. First. respondents were asked how
"sure" they were about both their qualitative and their quantitative risk estimates. A
Likert scale (1=very sure, 2=somewhat sure, 3=somewhat unsure. 4=very unsure)
was used to measure sureness in risk estimates. This study assumes that the less "sure"
someone is, the wider is his/her second-order probability distribution. One of the
problems with this measure is that it assumes that the ordinal "sureness" scale employed
in the survey is interpreted similarly by all respondents.
There are several factors that may influence someone’s level of ambiguity about
the probability of a health problem resulting from pesticide residues in food. They
include the level of trust that the government is setting appr0priate standards, the level
of trust that once these standards are set, they are being met, and the level of trust that
the scientific community understands the risks from pesticide residues and is truthful
about that knowledge. These trust levels are alternative proxies for measures of
ambiguity. The empirical section in chapter 5 tests whether they have any effect on the
demand for certified apples.
The proxies just described for ambiguity are all proxies for baseline ambiguity.
They are related to the ambiguity associated with the regular apples. This allows us to
gauge differences in ambiguity across people. Although the survey did not ask
respondents about their perceptions of the changes in ambiguity when switching to the
certified apples, this research develops a proxy for the change in ambiguity across
certification type. People who do not currently trust that the standard for pesticide
residues is being met, and who feel that if the government tested and certified apples
85
for their residue level that such a program would be effective are likely to experience
the most ambiguity reduction when switching to the certified apples. A dummy variable
in the demand model tests the effect of reducing ambiguity on the demand for certified
apples.
Other Issues
Prices
Respondents were told the prices of both regular and certified apples. Each
respondent received one of forty price combinations of regular and certified apples
ranging from $0.49 to $1.19 for the uncertified apple and $0.49 to $1.59 for 'the
certified apples. To ensure randomization of the price combinations given to
respondents, price combinations were assigned on the basis of the last four digits of the
respondents’ phone number. Respondents were told that all apples would look the same
as those they usually buy. They were then asked if they would buy all of one type of
apple (certified or uncertified), some of both, or none at all, and the quantities of those
apples they would likely buy at the given prices.
The contingent valuation survey used in this research was an "experiment” in
which the prices of regular and certified apples were systematically varied to ensure
price variation in the data. The survey thus generates respondents’ purchase intentions
for apples on a typical shopping occasion. However, the estimation of the surplus
associated with the purchase of certified apples is based on specific prices for both
regular and certified apples that are the same for all consumers. The price of the
regular apple is the average retail price (per pound) of red delicious apples in Michigan
86
in June/July, 1992 (the time period of the survey) of $0.98 per pound (United States
Department of Agriculture 1993).
Apples certified exactly as they were described in the survey are not currently
available on the market. In order to estimate the cost per pound of certifying apples to
have been produced without pesticides (or to meet federal standards for pesticide
residues) this study used cost data from Nutriclean, a private organization located in
California that certifies produce for pesticide levels.”59 This calculation produced an
estimate of $0.06 per pound. The price of the certified apples was thus $1.04 ($0.06
higher than the regular apple price).
Implementation of the CVM Survey
The target population for this study was Michigan households that purchase
food. A sampling error of j: 1% required 1000 completed surveys. To reach this
goal, we purchased a sample of approximately 2600 Michigan telephone numbers from
Survey Sampling Incorporated (SSI). The sample purchased from 881 was drawn using
random digit dialing. Of these, only 1730 phone numbers had to be used to reach the
goal of 1000 completed interviews. Of the eligible households contacted. 67% (1003)
5' The cost to the government of certifying apples would most likely be different
from that of a private firm. However, the cost to Nutriclean gives us a "ballpark"
figure.
‘9 Nutriclean’s cost per sample to certify apples (regardless of the size of the
operation) is approximately * $1200 (Jim Knundson, Nutriclean, telephone
communication, September, 1994). We use this data and the 1987 Michigan Census
Data on total pounds of apples harvested and number of farms harvesting to calculate
the price per pound of certification.
87
completed the survey. The telephone interviews were conducted by the Institute for
Public Policy and Social Research (IPPSR) at Michigan State University.
The surveys were conducted with adults over the age of 18 who did most of the
food shopping for their household. They were conducted by telephone during June and
July, 1992. Each survey averaged 16 minutes in length but varied from 7 to 35
minutes.
All questions that were added to or revised in the van Ravenswaay and Hoehn
(1991) questionnaire were pretested using face-to-face interviews with typical household
food purchasers. Further pretests of the entire questionnaire were also done in
conjunction with IPPSR using telephone interviews with Michigan consumers. The
pretest interviews, as well as the final interviews, were conducted by IPPSR personnel
who were specially trained to administer the survey. The pretests improved not only
the survey instrument, but also allowed us to develop detailed instructions on how
interviewers should handle unusual or difficult situations, thus improving the reliability
of data collection.
The CATI software used by IPPSR to conduct the telephone survey allowed for
skip patterns that depended on how respondents answered certain questions.
Interviewers were thus able to tailor each survey to the particular responses of the
individual being surveyed.
A Critique of Survey Methods
The results from contingent valuation surveys are often criticized because of the
large divergence between what people say they will do. and what they actually do.
88
Carson (1991) has developed a series of criteria for designing constructed markets to
increase the reliability of the results. His criteria focus on the theoretical accuracy and
policy relevance of the scenario offered in the survey, as well as on the extent to which
the scenario is understandable, plausible, and meaningful to respondents. This section
uses these criteria to evaluate the strengths and weaknesses of the contingent valuation
survey design used in this study.
To guarantee theoretical accuracy, the researcher must ensure that features of the
created scenario are compatible with economic theory. For example, property rights
must be clearly specified, the respondent’s budget constraint must be binding,
substitution possibilities must be clearly indicated, and the payment mechanism for the
good in question should result in accurate estimates of value.
The theoretical framework outlined in chapter 3 warrants the use of a private-
goods market. Individual demand curves for the basis of the measure of consumers’
willingness to pay for marginal changes in risk and ambiguity. The private market for
apples is thus an appropriate setting for measuring willingness to pay for risk reduction.
The budget constraint of the respondent is likely to be binding since respondents
were given the prices of goods in a market setting. It is doubtful that respondents
would strategically exaggerate the number of apples they would buy at these prices. It
is also unlikely that people would have difficulties making decisions about products.
Most people buy apples; it is easy for them to accurately predict their purchasing
behavior.
It is important in survey design to incorporate substitutes for the good in
question. Otherwise, people are responding to questions out of the context in which
89
they would actually be required to make payments. The constructed market used in this
research consisted of both certified and uncertified apples, with the prices for both
indicated. Since certified apples are probably the closest substitute for regular apples
(assuming the prices for regular and certified apples are not too divergent), this
approach contributes to the creation of a realistic buying scenario.
The questionnaire was designed to isolate the effects of perceived risk and
perceived ambiguity on the demand for apples. The shopping scenario was structured
so that respondents would assume that product attributes other than these did not vary.
This was accomplished by asking respondents to consider the variety of apples they
normally bought and to assume that the quality of all apples was the same as they
normally observed in the fall.
It is important that the results of the survey be relevant to policy makers.
Understanding the relationship between willingness to pay for pesticide residue
certification and for marginal changes in risks and ambiguity can aid policy makers in
designing policies regarding pesticide residues that target the concerns of consumers
making decisions in the face of risks. This study considers not only how consumers
value changes in the risks and ambiguity from pesticide residues. but also the effect of
food labeling on consumer choice. These are all of interest to decision makers in the
area of food-safety policy. The methods used to elicit perceptions about risks from
pesticide residues and ambiguity about those risks can be used in any application that
requires an understanding of consumers’ decision making under uncertainty.
One could question the policy relevance of a scenario that considers a market
where apples are the only product tested and certified for residue levels. If policy
9O
makers were to require apples to be tested and certified, they would likely also require
that other produce be tested and certified. However. respondents are not likely to be
able to accurately predict and describe their behavior in the more general scenario
where all foods are described as being tested for pesticide residues. This scenario
would require changing the prices of all substitutes for apples; creating an easily
understandable scenario then becomes infeasible.
As the name implies, the results of contingent valuation studies are contingent
upon the scenario presented to respondents. The highly specific scenario given to
respondents in this survey makes it difficult to generalize the results. However, from
the point of view of economic theory and respondent comprehension. the specific
scenario offers the advantage of yielding results more valid than those obtained from
less specific scenarios.
Respondents will be more able and motivated to accurately predict their behavior
if the scenario presented to them is understandable, plausible. and meaningful.
Although contingent valuation surveys commonly ask respondents to directly state their
willingness to pay for specific goods or services, this research did not use this approach.
Instead, it asks respondents how many apples they would likely buy under different
scenarios. The demand analysis then uses this information to estimate willingness to
pay for risk reduction. Since consumers are more likely to have to make market
choices than to be asked to pay directly for reduced risks. the market scenario is more
understandable, plausible, and meaningful respondents. Furthermore, consumers are
typically price takers in apple markets.
91
Apples provide a good vehicle for studying the effects of risk and ambiguity on
WTP because most households purchase apples, apples have been associated with
pesticide residues in the media, and they allow people to predict their behavior under
very familiar and likely circumstances.
One of the goals of this research was to develop reliable methods of soliciting
perceptions of baseline risk and ambiguity as well as the perceptions of changes in risk
associated with certified apples. Extensive pretesting of the survey instrument was
therefore conducted to ensure that questions were thoroughly understood. Cross
tabulations between qualitative and quantitative responses to risk questions suggest that
respondents did, in fact, understand the nature of the quantitative risk assessments they
were asked to make. Careful consideration was also given to the order of questions.
The survey was designed to first get respondents thinking about their current apple
purchasing behavior before asking them to make hypothetical purchasing decisions.
Many studies involving decisions under risk present respondents with scientific
or objective risks and then assume that respondents use those estimates in their decision
calculus. In some cases, this approach may be useful, but the presentation of the
objective risks should be coupled with questions that examine how such information
alters risk perceptions. Researchers cannot assume that respondents accept objective
estimates of risk without consideration of their own experience and other information.
The survey used in this research asks respondents to make their own assessment of the
risks involved in pesticide residue consumption. This is a more realistic scenario since
consumers are generally forced to make risk assessments before making purchase
decisions. Furthermore, scientific or objective risk assessments are not generally
92
available to consumers. To ensure that respondents understood the questions, risk
perception questions were worded to allow respondents to think in terms of numbers of
people affected rather than just in probabilities.
The survey used in this research also allows examination of baseline ambiguity
about risk. Questions were asked about how "sure" respondents felt about their risk
estimates and how much they trusted the government and the scientific community.
Because there is a lot of uncertainty about the risks from pesticide residues, ambiguity
may be an important factor in explaining consumers’ willingness to pay for risk
reduction. Allowing respondents to express their uncertainty about the health risks from
pesticide residues may improve the validity of the risk perception measures. When
respondents are given the opportunity to express their reservations, they feel less
pressure to be "right," and are then more likely to give their best estimate of the risk
instead of giving worst- or best-case estimates.
CHAPTER 5: ECONOMETRIC TECHNIQUES AND EMPIRICAL ANALYSIS
This chapter presents the results the CVM survey of Michigan consumers’ apple-
purchasing behavior and attitudes about pesticide residues infood. The chapter first
presents some descriptive information from the CVM survey to provide a broad
overview of the structure and content of the data. It then summarizes the general
methods used to estimate the willingness to pay for pesticide residue certification based
on the demand for certified apples. This is followed by a discussion of the Heckman
two-stage demand model used to estimate the demand for certified apples. After
defining the variables used in the estimation. the results of the first-stage PROBIT
analysis and the second-stage demand estimation are presented. This is followed by a
discussion of how the estimated coefficients are used to calculate the consumer surplus
associated with the certified apples and to conduct the sensitivity analyses. The chapter
then presents the results and discusses their implications.
General Survey Results
The tables in this section present some descriptive results from the CVM survey
used in this research. These results help assess how representative the data are of the
Michigan population. The data are presented in the following categories: demographics,
risk perceptions, perceptions of ambiguity, perceived risk reduction when foods are
certified, perceived health effects from pesticide residues in food, attitudes toward
government and the scientific community, and purchase intentions.
93
94
Demographics
To assess how representative of the Michigan population the sample used in this
study is, the results from the present survey are compared to the data from the 1990
Michigan Census or the "Current Population Reports." The census data and the data
from the ”Current Population Reports” are based on the population of Michigan when
possible, the US when not.
Table 5-1 shows that the average household in this study is about the same size
as the average household in Michigan in 1990. However, the current sample
overrepresents households with children under the age of 18 and underrepresents single-
person households.
Table 5-2 indicates that this sample may underrepresent low-income households
and overrepresent wealthy households. However, the mean before-tax household
income is comparable to the Michigan average.
Table 5-3 shows age of respondents. Elderly people (over 75) and very young
people (under 25) may not be adequately represented in our survey.
95
Table 5-1: Household Demographics
This 1990
study Census
Characteristic N = 1,003 (Michigan)
Average household size (# of persons) 3.0 2.7
Percent of households with children 54.8 49.2
under 18 % %
Percent of single-person households 13.3 23.6
. % %
* Figures may not add to 100% due to rounding
Table 5-2: Household Income
This 1990
study Census‘ (US)
Income N = 1,003
Less than $10,000 3.7% 14.9%
$10,000 - 49,999 58.7% 60.4%
$50,000 or more 28.2% 24.6%
No answer / don’t know 9.6% NA
Mean before-tax household income $41,025 $37,403
1. from Current Population Reports. Series P-60, No. 174. 1990.
"' Figures may not add to 100% due to rounding
Age of Respondent
Table 5-3:
Age of 1990 Census Age of
Respondent (Michigan)‘ This Study Respondent
18-24 14.7% 8.1% 25 or younger
25-44 43.6% 22.2% 26-35
23.6% 36-45
45-54 13.9% 17.1% 46-55
55-64 11.6% 13.8% 56-65
65-74 9.6% 9.9% 66-75
75+ 6.6% 4.9% 76+
0.4% No Answer
* right-hand age categories apply to this study; left-hand age categories apply to
census data. .
1. percentages apply to population over 18 years of age.
The demographic results indicate that this survey probably represents the
Michigan population well. It is probably safe to use the results of our survey to
extrapolate to the Michigan population as a whole.
Risk Perceptions
The perceived baseline risk level from pesticide residues in food is presented in
Table 5-4. The average perceived risk is between 1 in 10,000 and l in 1000. This is
the same mean perceived risk found by van Ravenswaay and Hoehn (1991), although
the wording of the questions differs between the two surveys. However, many fewer
households in our study perceived no chance or a chance of 1 in a million. This result
97
Table 5-4: Perceived Chance of a Health Problem
"Suppose there were a million people from households like yours who did nothing to
reduce or avoid pesticide residues in food. What do you think the chances are that a
person from one of these households would have a health problem someday because of
pesticide residues in food?"
Response (prompted) Eercent Respondents —]
No chance 2.4% I
l in a million 4.1% II
1 in 100,000 14.1%
1 in 10,000 23.0%
1 in 1,000 22.8%
1 in 100 10.8%
1 in 10 8.4%
Certain to happen 8.2%
Don’t know/no opinion /refused to 6.3%
answer r
Note: N = 1,003. Figures may not add to 100% due to rounding
was expected because the earlier survey did not control for the possibility that
consumers may have been taking their own actions (such as washing or peeling fruits
and vegetables) to reduce or avoid pesticide residues in food.
These results show that perceptions of the baseline risks from pesticide residues
in food vary significantly across people. Consequently, surveys that assume that risk
perceptions among all consumers are the same would lead to incorrect estimates of
willingness to pay for risk reduction.
98
Perceived Risk Reduction when Foods are Certified
The risk levels presented in Table 5-4 are the perceived risks associated with the
consumption of conventionally produced foods. If foods are certified for different levels
of pesticide residues, many people will perceive that those risks change. Table 5-5
shows the percentage risk reduction respondents felt they would be getting if foods were
certified by the government as indicated.“0 Respondents felt they got more risk
reduction when foods are "produced without pesticides" than when foods "meet federal
standards." However, the difference in risk reduction between the two types of
certification is not large, suggesting that consumers may perceive that either the current
federal standards adequately protect them from the risks of pesticide residues, or that
they do not distinguish the exact content of the certification when making their
assessments of risk reduction.
Perceptions of Ambiguity
In addition to the questions about baseline risks and the changes in risks from
pesticide residues in food, Michigan respondents were also asked to indicate their level
of sureness about risk. More specifically, they were asked " how sure are you that the
chance of a health problem is ? (blank filled in with respondent’s estimate)." The
answer to this question is used as a proxy for ambiguity. Despite the difficult nature of
quantitatively assessing risks, Table 5—6 shows that 23% of respondents were very sure
about their risk estimates, 45% were somewhat sure, 20% were somewhat unsure, and
‘° The perceived risk levels associated with the labels can be calculated by applying
the perceived risk reduction with the labels to the risk levels in Table 5-5.
99
Table 5-5: Perceived Reduction in the Risks from Pesticide Residues When Foods
Meet Federal Standards for Pesticide Residues or Are Produced Without
Pesticides
"Now suppose that all foods met the federal standard for pesticide residues (split sample
received: Now suppose that pesticides were not used in producing foods). What percent
do you think that would reduce the chances of a health problem happening someday to
people who currently do nothing to reduce or avoid pesticide residue?" (open-ended)
Certification: "Meets Certification: "Produced
federal standards" Without Pesticides"
Percent Percent Respondents
Reduction
0% 0.9% 1.7%
0% to 20% 10.6% 13.4%
20% to 40% 14.3% 7.2%
40% to 60% 28.1% 29.9%
60% to 80% 21.9% 19.4%
80% to 99% 9.4% 13.2%
100% 4.1% 15.3%
Refused / no 10.6% 0.0%
answer
Note: N =852. Respondents who answered ”there was no chance of a
health problem,” or who answered "don’t know/no Opinion/refused” to the
question in Table 5-4 were not asked this question. Figures may not add to
100% due to rounding.
100
Table 5-6: Sureness about Health Risk
"I-Iow sure are you that the chance of a health problem is ? (blank filled in with
respondent’s estimate)"
Response (prompted) Percent Respondents
Very sure 23.4%
Somewhat sure 44.7 %
Somewhat unsure 19.9
Very Unsure 5.3
Don’t know/no opinion 6.7
Wares may not add to Mg.
only 5% were very unsure. The results of this question demonstrate that people have
different levels of confidence in their risk estimates (i.e., the spread of the SOP varies
by individual).
Attitudes Toward Government and the Scientific Community
Table 5-7 shows the results to questions about trust in the government and the
scientific community. This research assumes that trust in the government in scientific
community influences consumers’ ambiguity. levels. Almost 50% of respondents said
they either "somewhat disagree” or "strongly disagree” with the statement "I trust the
federal government to set the same standards that I would set in limiting the amount of
pesticide residues allowed in food.” Even if there were certification guaranteeing that
foods met federal standards, many respondents felt that standard would not be one they
would agree with. There is similarly sentiment about the federal standards for pesticide
residues.
9.6.56. 9 one Roo— o. 3:8: .38 :9.sz .82 "Z .862
5.5.858
2.2.58: 2.. .3 8980...... 9.: 5.65.
:03 2: coo. E 3:28. 026.38 «o 226.
fivfi $wd— QumdN .80de fiofim 2.2.56 .23 c3:_oo:m: 3%.. 5.3.. 2E.
.8223. 33.3.. Ea... $7.... 5.3...
Son: 265. me... .23 Son: .35.... on
101
.3. e . ..N .33 e2: .33., c. 8...... a. :8 $5.588 2......23 a:
85......
.8... .8... ...a a... . :8. 0.. .... ..a
.3... .38 :3: «5:: 8:. a: 85...... .83. o... 8.... .2. .3... .
...8.
E 326:: 8:28.. 0238.. me .558: 05
«5:8: 5 .3. 2.53 _ .2. 8::—.53... 05::
$2. $93 $5.3 :wmfiu RN: 05 .3. e. ...e..........>eM .83». o... 7...... _
8:3:38: .595— 288355
3:32 - .
225% o: 3&8... 02mg... 3.»: 3.»:
355. ...50 >_w=ebm 5.32:8 . .EBoEem bucobm
555.80 8......8: 2.. ...a .85528 0.. .83.... 83...... .3 2...:
102
those standards." These respondents also do not trust the enforcement efforts.
Of the respondents, 36% said they either somewhat disagree or strongly disagree
with the statement that ”the health risks associated with current levels of pesticide -
residues in food are well known and understood by the scientific community.”
Furthermore, 46% of respondents said they either somewhat disagree or strongly
disagree with the statement ”the scientific community can be trusted to be truthful about
what they know about the health risks from pesticide residues." There is substantial
lack of confidence that the scientific community has presented trustworthy information
about the risks associated with pesticide residues in food.
It appears that ambiguity does not stem from only one source, although it does
seem that ambiguity derives less from scientific uncertainty about risks than from the
trustworthiness of scientists and government regulators.
Perceived Health Effects from Pesticide Residues in Food
Table 5-8 shows the types of health effects people associate with pesticide
residues in food. The survey asked respondents: "suppose someone from a household
like yours had a health problem someday that resulted from the current levels of
pesticide residues in food. In your opinion, what would the health problem most likely
be? (open-ended)." While respondents perceive a variety of health problems associated
,with pesticide residues in food, more than 50% believe that cancer is the most likely
illness. Although these results are difficult to compare to the results from van
Ravenswaay and Hoehn because of the different wording of the question in that survey,
they also found that cancer was considered the most likely health effect to occur.
103
Table 5-8: Perceived Health Effects Associated with Pesticide Residues in Food
”Suppose someone from a household like yours had a health problem someday
that resulted from the current levels of pesticide residues in food. In your
opinion, what would the health problem most likely be?" (open-ended)
.6.8'/., gastro-mtestmal problems.
.6.17., allergies.
.4.8'/., respiratorg problems.
l1.97., nmhmg.
- 11.97., other.
-13.5°/., don’t know / refused.
0% 10% 20% 30% 40% 50% 60% 70% 80% 907. 100%
104
Purchase Intentions
Table 5-9 shows respondents’ purchase intentions for certified and uncertified
apples at different prices. The table gives the percentage of respondents indicating they
would buy that type of apple if both certified and uncertified apples were available.
The data do not indicate the quantities purchased. There was not a large difference in
purchase intentions between the subsample that evaluated the "meets federal standards”
certification and the subsample that evaluated the "produced without pesticides"
certification. This is evidence that there may be substantial value in reducing ambiguity
about whether foods meet current federal standards for pesticide residues.
Table 5-9 suggests that the difference in price between certified and uncertified
apples is an important factor in respondents’ decision to purchase certified apples.
Respondents were interested in certified apples only if the price was low enough relative
to the uncertified apple. Specification of substitution possibilities is clearly important.
Willingness to Pay for Certification
As described in chapter 3, the area to the left of the ordinary demand curve for
certified apples above a specified price and below the choke price approximates the
willingness to pay for residue certification. This study uses the data from the survey of
Michigan households to estimate the demand for certified apples.“I The consumer
surplus is calculated for each respondent; the average surplus for the sample is then
calculated to obtain the average willingness to pay for certification.
6‘ The survey is described in detail in chapter 4.
105
6085.3 on 2.53 ..an 30.. 8: .825
.83» 2. 2...... ..o 090 «5823. 892.828 2.. 03... c. 33:...2 @5952 o. 0.6 88— o. E... .2. as... 8.8.... .m8_nz 882
88.3.
em... 8.. an. .3... 3...... ..So
e3. .3... .3... .3... __< a 252
e _ .3 .2... saw. .3... 5.... .0 2.5m
.33 s3... e2... e ...e 8.9.... 8......8
seem e . .2 $2 .32 83.2 33......
80E .5 Omdm
”muzmmmmua mUEm
82 .o omdm
.mUmemmn—E moan.
8.0:. ..o Omdm
.mUZmdmn—EQ mUEm
m8_ .0 8.8
”MUmemn—n—E moan—
8Eu_.8m 255:5 .8269... "=e_.3..__...ou
83.5.5 .22....
2...... 8:28.. 2.5.8.. ..2 523250
mJn—Az. ho mm>h >Dm Oh. mmflZUZ—JEB UZEEUEZ— whszZOmmm—m hzmummm
Eur—mt.
hO ”ma—>5.
8...... BEES: ...a 85.30 a... 0. 382......» .3 2...;
106
The Effect of Risk and Ambiguity on
Willingness to Pay for Certification
The willingness to pay for the residue certification is estimated given the
respondents’ perceived levels of risk and ambiguity associated with the regular and
certified apples. The marginal value of a change in risk or a change in ambiguity is
calculated by using the coefficients from the demand analysis to conduct sensitivity
analysis on the total willingness to pay for certification. That is, we see how the total
willingness to pay for certification changes with small changes in either risk or
ambiguity, holding the levels of all other variables at their original values.
Modeling the Demand for Certified Apples as a Two-Stage Process
The survey used in this research asked respondents to consider a hypothetical
shopping scenario in which both regular and certified apples were available at stated
prices. Respondents were first asked if they would choose regular apples, certified
apples, some of both, or no apples at all. They were then asked how many apples of
that particular kind they would buy.
The decision to buy the certified apples can be thought of as a two—step process.
The consumer must first decide whether or not to buy the certified apples. Once the
decision to buy them has been made, he/she must decide how many to buy.
Limiting the range of the values of the dependent variable in the second stage to
only the positive values of those who consumed the apples leads to a nonzero mean of
the disturbance term and to biased and inconsistent least squares estimators. A TOBIT
model of the demand is often used in such situations to account for the bias that results
107
when estimating the demand for purchasers only. The TOBIT model is described as
follows:‘52
Let
Y,‘=a+px,.+e,.‘ (i=l,2,...,n) (54)
be a regression for which all basic assumptions are satisfied. Let the dependent
variable, Y', represent the consumption of some commodity (or an index of desire), X
represent a vector of explanatory variables, 6 represent the vector of coefficients on X,
and n is the number of households. For the households that consumed positive
quantities of the commodity, Y' is the actual quantity. For households that did not
consume any of the commodity, Y' represents an index of the "desire” to purchase'the
commodity.
For households that did not consume the good, Y' is not observed and is
recorded as zero. Thus, instead of observing Yi', we actually observe Yi, which is
defined as
Yi = Y,‘ if Y,‘ > O,
= 0 if Y,‘ s 0.
Equation 5-1 then becomes
Yi=a + BXfiei (5'5)
where Yi is truncated at zero and e is truncated at {art-BX} The probability
distributions of Y, and e, are both cut off and the probabilities are piled up at the cut-off
‘2 The presentation of the TOBIT model borrows heavily from Kmenta (1986, 560-
563).
108
point. The mean of Yi is thus different from that of Y,’; the mean of ei is different from
the mean of e,’, which is zero. This is true whether the points for which Yi=0 are or
are not included in the sample (i.e., this is true whether the sample is truncated or
censored).
The log-likelihood function for n observations is given as
- a-BX,
L=1-Z.lF
‘21:“ )0g( (56)
+ z [-- 2|og(21ra'2)-_2._(Y.-a 6X)?”
where:
zi if vi >0.
1
o ifY, <0
This function can be maximized to obtain the maximum likelihood estimates
(MLE) of the model parameters.
However, the TOBIT model assumes that the factors that affect the decision to
buy a product are the same as those that affect the quantity purchased. Several
researchers (Haines, et al 1988; Yen 1993; Blaylock and Blisard 1992) use Cragg’s
(1971) double—hurdle model to explicitly account for the two—step process involved in
consumption decisions. This model consists of estimating a PROBIT model for the
indicator of whether the dependent variable is positive, and a truncated regression model
for the non-limit observations. In this model, however, the relationship of the
disturbance in the latent regression underlying the PROBIT model and that in the
truncated regression is unclear (Greene 1992).
109
Heckman (1976, 1979) also proposed a two-step procedure to correct for the
problem of censored data. Suppose that of all n observations there are m (m0. The regression equation for these observations is
Y‘=a +pX‘+e‘. (i =1,2,...m) (5‘7)
where Y, and e; are truncated normal variables. Since the regression is Only for those
observations for which Y,‘ > 0, the conditional expectation of Yi given Yi‘ > 0 is
E(Y,|Y,‘>0) a+px,+5(e,|y,‘>0)
a +BXi+E(ei |e,‘>-a -px'.) (5-8)
(1' = (1,2,...m)
Given that ef~ N(0,o’), the mean of the corresponding truncated variable, 6,, is
E(e.-Ie.-‘>-a-BX,)=ox, <5-9)
where
+ X
fl“ 5 ‘)
1‘.__°__ (5-10)
(“BX‘
F( )
and f(*) represents the density and F(*) the cumulative distribution function of a
standard normal variable. The variable X, is the inverse of the Mill’s ratio. To allow
for the non-zero mean of e, the equation
110
Y,=a+[3X,+o).,+e, (i=l,2,...,m) (5'11)
is run on the m observations for which Yi'>0. The error term in equation 5-8 then has
a mean of zero.
If K, were observable, this equation would have a disturbance with zero mean
and the least squares coefficients would be unbiased. Since x is not observed, it must
be estimated. The estimation procedure is as follows: (1) probit analysis is run using a
dichotomous variable for consumption and non-consumption as the dependent variable,
(2) equation 5-8 is estimated with A, as one of the explanatory variable using the m-
observations for which Y,>0. The estimators of a and B are consistent and
asymptotically normal.
This research estimates a Heckman two-stage model and uses the results of the
demand estimation in the second stage to calculate consumer surplus.
Estimating Consumer Surplus from the TOBIT Demand Equation
Hellerstein (1992) explains that substituting the estimated coefficients from the
censored model (such as that described by the TOBIT or Heckman’s two.stage
procedure) into a linear functional form in order to calculate the consumer surplus is
incorrect. When exogenous factors change, the individual's choke price changes, and
this has non-linear effects on consumer surplus that are not accounted for if one uses the
coefficients from the Heckman model and then integrates under the demand curve using
those coefficients.
The correct measure of expected consumer surplus, is the following:
111
Pan
5le1 = j («p was + (up) (1,) (5-12)
where a is the standard deviation of the disturbance term. e. <1> and qb are evaluated at
(BXi/ai), and P0,, is the observed price at which the integral is evaluated“. and Pchoke is
the value at which demand approaches zero (selection of this price is described below).
This measure of consumer surplus explicitly accounts for the fact that the choke price
depends on the stochastic term.
As Hellerstein points out, since the integrand in equation 5-9 is strictly positive,
the choice of a "choke" price is no longer obvious. This research follows Hellerstein
and uses the cutoff price from the linear deterministic model: if the linear deterministic
-(fiX - Bp‘
6,.
P
model is Y = XB, the cutoff price is f) .6“ This is the choke price we
use when we estimate equation 5-9.
The value of equation 5-9 is calculated for each respondent, at a specified price
for regular and certified apples. This calculation yields the total surplus associated with
the purchase of certified apples. Since the dependent variable in the demand equation
(equation 5-11 below) is the quantity of certified apples purchased in a fall season; the
generated surplus measure is the surplus associated with the purchase of certified apples
‘3 The choice of the observation price is described in chapter 4.
“ Hellerstein (1992) notes that this is not the price at which expected demand goes
to zero, and is therefore not necessarily better than some other value (such as the
maximum price observed in the sample).
112
in the fall quarter. The surplus per pound of certified apples is the fall surplus divided
by the expected number of apples bought in the fall season. This amount is estimated
using the integrand of equation 5-9.
Estimation
The general two-stage model of demand for certified apples estimated in this
research is the following:
W
The following PROBIT model is run to determine the factors that affect the decision to
select certified apples.
Z = an + apcpc + amp, + afir’ + arm/Air + 01,6’ + amAe + onM + 0:55 + e
mm
For households that choose certified apples:
Q = 5,, + Bap. + 3A1), + Eff“ + BA‘Ae + aMM + aS.S' + e
where:
Z = a dummy variable defined as: Z=l if Yf>0 (the household buys
certified apples), Z=O if stO (the household does not buy
certified apples).
Q = the quantity of apples consumed
a0 = constant term in PROBIT equation
5. = constant term in the demand equation
1), = the per-pound price of certified apples
p, = the per—pound price of regular apples
A1
113
perceived baseline probability of an adverse health effect resulting
someday because of pesticide residues in food when one consumes
regular apples
perceived change in the probability of an adverse health effect
resulting someday because of pesticide residues in food when one
consumes certified apples
perceived ambiguity about 1r'
perceived change in ambiguity about 1r' when switching to
certified apples
Income
demographic characteristics of the household related to choice of
apple
demographic characteristics of the household related to the
quantity of apples bought
disturbance term
The next section describes various measures of the variables in the two-stage model that
were solicited from survey respondents.
Variable Definitions
Dependent Variables
a dummy variable defined as: Z=l if the household selected
certified apples), Z=0 if not.
quantity of apples consumed in a fall season. The survey asked
respondents about their apple-purchase intentions for a given
shopping occasion. This variable is constructed by extrapolating
the quantity purchased on each occasion to the quantity purchase
in the fall based on information obtained in the survey about
respondents general apple-buying patterns.
Independent Variables
Response to the question "suppose there were a million people
from households like yours who did nothing to reduce or avoid
pesticide residues in food. What do you think the chances are
1"(VLCT H) =
1"(SUSL) =
WA:
At (ABS) =
Al'(%) =
At(%OWN) =
e'(SURE)
114
that someone from one of these households would have a health
problem someday because of pesticide residues in food?"
Although the response categories are discrete choices ranging
from O to one, the variable is treated as a continuous variable.
(Corresponds to Q4)
Qualitative response to the question "suppose someone from a
household like yours did nothing at all to reduce or avoid
pesticide residues in food. What do you think the chances would
be that someone from that household will have a health problem
someday because of pesticide residues in their food?" This is a
dummy variable that takes the value of 1 if the respondent
answered "very likely,” or "certain to happen." and the value of 0
if the respondent gave any other response. (Created from
responses to 02).
Qualitative response to the same question as for 1r'(V1..CT H).
This is a dummy variable that takes the value of 1 if the
respondent answered "somewhat unlikely." or "somewhat likely,"
and the value of 0 if the respondent gave any other response.
(Created from responses to Q2)
perceived (absolute) change in the probability of an adverse health
risk when switching from regular apples to certified apples
perceived (percentage) change in the probability of an adverse
health risk when switching from regular apples to certified apples
The amount of risk reduction (in percentage terms) respondent
believes he/she gets through his/her own actions (washing,
peeling, etc.) (Corresponds to Q7)
perceived probability of an adverse health effect resulting
someday because of pesticide residues in food when one consumes
certified apples. Because perceptions of this probability were not
asked directly in the survey, it is calculated as 1r’-A1r(ABS).
The survey asked people how sure they were about their
risk estimate. The response categories were "very sure,"
((NTRSTSTD)
£'(QUAL)
e’(VSEFF)
Ac
1’:
Pg
115
”somewhat sure." "somewhat unsure." and "very sure."
This variable is a dummy variable created from the
responses to that question. e’(SURE) has a value of 1 if
the response was "somewhat unsure." or "very unsure;" 0
otherwise. (Created from answers to 05)
The survey asked respondents whether they "strongly
agree," "somewhat agree." "somewhat disagree," or
"strongly disagree" with the statement "I trust that once
the federal standards [for pesticide residues] are set, all the
food I buy will meet those standards." This variable takes
a value of one if the respondent answered "somewhat
disagree," or "strongly disagree,” and 0 otherwise.
(Created from answers to Q9)
Same as e'(SURE) but this variable is the "sureness"
associated with the Qualitative. rather than quantitative
measure of risk. (Created from 03a).
= The survey asked respondents ”suppose the federal
government did testing and certifying [of pesticide residue
levels]. How effective do you think such a program
would be in ensuring that foods had no pesticide residues
above federal standards [or, that foods had been produced
without pesticides?” This variable is a dummy variable
that has a value of 1 if the respondent said "very effective"
or ”somewhat effective;" 0 otherwise. (Created from
Q12).
The survey did not ask respondents about their perceptions of the
changes in ambiguity when switching to the certified apples. This
variable is a proxy for changes in ambiguity. It is an interaction
term created by multiplying the variables e'(VSEFF) and
e'(SURE) together. If the value is l, the respondent perceives a
higher level of ambiguity reduction than if the value is 0.
Each respondent was given one of 40 price combinations for
regular and certified apples. This variable is the per pound price
of regular apples the respondent was given.
The per pound price of certified apples the respondent was given.
r ini
HHSIZE
SCHOOL
UND18
116
' I
n n
Household’s annual (1991) income before taxes. Although the
responses were categorical. we treat income as a continuous
variable, using the midpoint of the range as the income.
(Corresponds to Q35, Q35a, Q35b, Q35c. Q35d. 035e, and
Q350
The number of household members in the respondent’s household.
(Corresponds to Q31, Q3la. Q3lb. and 031c)
The number of years of education of the respondent.
Respondents were asked the highest level of education they had
finished. These data were converted to years of education using
the number of years that level would most likely represent.
(Corresponds to Q34)
A dummy variable indicating the presence or absence of children
under the age of 18 in the household. The variable takes a value
of one if the household had at least one member under the age of
18; 0 otherwise. (Created from answers to Q31 and 0313)
Testing the Hypotheses
This section discusses how the estimated coefficients from the model are used to
test the hypotheses developed in chapter 3.
Hypothesis 1.
Hypotheses
The marginal value of a decrease in the mean probability of an
adverse health outcome (when ambiguity present) is positive.
If this hypothesis is true, then the higher the perception of risk reduction (Air)
associated with the certified apples, the more likely one would be to purchase the
certified apples. We therefore expect the coefficient on the variable that measures Air
to be positive in the PROBIT model. This hypothesis also suggests that the coefficient
117
on 1" in the demand equation for certified apples should be negative. as a higher 1r“,
holding baseline risk constant, would mean lower risk reduction.
Hypothesis 2. The marginal value of a decrease in ambiguity is positive.
1f the marginal value of a decrease in ambiguity is positive. then we would
expect that a perception of ambiguity reduction would increase the probability of
selecting the certified apples (the coefficient on As should be positive in PROBIT
model). Similarly, a perception of ambiguity reduction should increase the quantity of
certified apples purchased (the coefficient on As should be positive in demand equation).
Hypothesis 3. The WTP for certification decreases with initial ambiguity,
holding ambiguity reduction, baseline risk, and risk reduction
constant.
The higher the perception of baseline ambiguity, the less likely one should be to
buy certified apples (coefficient on measure of 5' should be positive in the PROBIT
equation).
Hypothesis 4. The WTP for certification increases with initial risk.
People who perceive high risks from pesticide residues should be more likely to
choose the certified apples. We would thus expect the coefficient on 1r' to be positive in
the PROBIT model.
118
Estimation Results
Factors Influencing the Decision to Purchase Certified Apples
The results of three PROBIT specifications are presented in Table 5-10. In the
three models presented, the most important variables (based on significance level)
influencing the decision to purchase the certified apples seem to be the prices of the two
apples (certified and regular), and the presence of children in the household under the
age of 18. The higher the price of the regular apples, the more likely it is that the
respondent will buy certified apples; the higher the price of the certified apples per
pound, the less likely they are to choose certified apples. The presence of children
under the age of 18 in the household increases the likelihood that the household will
purchase the certified apples.
W29
All three probit models indicate that risk reduction from the certification
(At(%)) and the risk reduction achievable through one’s own actions (Air(%OWN)) are
important determinants the decision to buy the certified apples. As expected, the more
risk reduction one perceives one gets from the certified apple, the more likely he/she is
to purchase the certified apple. holding the initial risk level constant. Similarly, the
more risk reduction people think they can get from their own actions, the less likely
they are to buy the certified apples, suggesting that the risk reduction one gets from
one’s own actions and the risk reduction one gets from certified apples are substitutes.
If they were complements, people would continue to undertake their own actions, even
though they acquired additional risk reduction from the certification, and the variable
119
Table 5-10: PROBIT Results
MODEL 1 MODEL 2 MODEL 3
Dep Var: Z Z Z
Coefficient Coefficient Coefficient
Variable (standard error) (standard error) (standard error)
a, 0.29 0.15 -0. 19
(0.51) (0.51) (0.53)
pr 2.27” 2.19‘” 2.07”
(0.60) (0.60) (0.60)
p, -2.68"‘ -2.62“' -2.45"‘
(0.54) (0.54) (0.53)
SCHOOL 006” 0.06” 0.06”
(0.03) (0.03) (0.04)
INCOME 0.00001” 0.00001” 0.00001”
(0.000003) (0.000003) (0.000003)
UND18 0.42’” 0.40'” 0.39‘“
(0.12) (0.12) (0.12)
1' 0.32 0.29
(0.26) (0.26)
f(VLCTH) 0.53”
(0.21)
1’(SUSL) 0.42”
, (0.21)
A‘l'(%) 0.50” 0.54” 0.49”
(0.23) (0.23) (0.23)
At(%OWN) 1 -0.56" -O.49" ~0.43'
(0.24) (0.24) (0.24)
e'(SURE) -0.35’
(0.22)
Ac 0.44' —0.42
(0.24) (0.28)
€(QUAL) -0.24‘
(0.14)
€(NTRSTSTD) 0.19
(0.15)
e'WSEFF) 0.14‘
r (0.07)
% of buying Ellis 72% 72% 73%
Log (L) -289.39 -290.95 -295 .88
x2 67.71 68.16 67.00
Pseudo R2 0.10 0.10 0.10
% correct predictions 75 74 73
* significant at the 10% level
*" significant at the 5% level
*** significant at the 1% level
120
At(%OWN) would not be significantly different from zero in determining the choice of
apple.
Ambiguity Reduction
The CVM survey did not explicitly ask respondents for their perceptions of the
reduction in ambiguity associated with switching from regular to certified apples.
However, we assume that among all respondents, those who have more baseline
ambiguity and who think that a certification program run by the federal government
would be effective would achieve the highest level of ambiguity reduction when
switching to the certified apples. This is tested using the variable Ac in model 1 and
model 2. This variable is positive and significant in model 1, indicating that ambiguity
reduction is important in determining consumers’ choices. The variable is not
significantly different from zero in model 2.
Further research is needed on ways of measuring perceptions of changes in
ambiguity before conclusive statements about the effects of ambiguity reduction on
choice can be made.
3 I. B. l
The effect of baseline risk is tested in model 1 and model 2 using a quantitative
measure of initial risk perceptions (1'). A qualitative measure of baseline risk is used in
model 3 (t'(QUAL). Although these models yield the expected positive sign on the
coefficient for the baseline quantitative risk variable, the variable is not significantly
different from zero in influencing respondents’ choice of certified or regular apples.
121
We cannot, however, conclude that risk perceptions are not important in the choice
process. It may be that a quantitative measure of risk such as a" does not adequately
measure peoples’ concern about the health risks posed by pesticide residues on regular
apples. Furthermore, because the survey asked respondents to imagine they were going
shopping intending to buy apples, baseline risk may not affect the choice of which
apple one buys while the change in risk does.
The qualitative measure of risk used in model 3 (t'(QUAL) is positive and
significant in determining consumers’ choice of apple type. Relative to people who
perceive a health effect resulting from pesticide residues in food to be "very unlikely"
or "no chance," (as captured by the constant term) people who perceive the chance to
be ”somewhat unlikely,” or "somewhat likely” (r'(SUSL)=l) and people who perceive
the chance to be ”very likely," or "certain to happen” (ii-'(VLCTH)=1) are more likely
to purchase the certified apple. Even moderate concerns about the health risks from
pesticide residues compels people to buy the certified apples. Although we would
expect the quantitative and qualitative measures of risk to be highly correlated, the
difficulty of putting a number on one’s risk assessment (which does not apply to the
qualitative assessment) may be the reason that the quantitative variable (1") does not
capture the effect of concerns about pesticide residues on the demand for certified
apples.
If people have different assessments .of the risks from pesticide residues on
regular apples for children than they do for adults, a dichotomous measure such as
UND18 may better reflect consumers’ concerns. Although the variable a" should reflect
this (since it is a risk assessment for the whole household). the variable UN 018 may do
122
a better job of distinguishing these dichotomous concerns. This would be consistent
with the National Research Council (1993) finding that children face the greatest risk
from pesticide residues in food (they consume more fruits and vegetables than do adults,
unlike adults, they are subject to developmental effects. and they have a longer time
span over which to accumulate residues in their system).
The variables. e'(SURE) and e'(QUAL) in model 1 and model 3 measure
peoples’ ”sureness" about their estimate of the baseline risk. e'(NTRSTSTD) is the
measure of peoples’ trust that the standard set by the government for pesticide residues
in food (whatever that may be) meets the standard for pesticide residues
(¢'(NTRSTSTD)=1 indicates a person does not trust that food meet the standards).
Although hypothesis 4 suggests that the sign of the coefficient on ¢'(SURE) and
e'(QUAL) should be negative, the expected sign on e'(NTRST STD) is unclear. We
assume that people who mistrust that the food they buy meets the federal standards for
pesticide residues have higher ambiguity (indicating that the sign of its coefficient
should be negative). However, this variable probably also captures their higher risk
perception as people who feel that food does not currently meet the standard probably
perceive higher levels of pesticide residues in their food. The sign on the coefficient
might then be positive. This variable probably confounds the two perceptions; and the
expected sign of its coefficient remains undetermined.
Both "sureness” measures of ambiguity (e'(SURE) and e'(QUAL) in model 1 and
model 3 are significant and negative at the 10% level; the "trust" measure of ambiguity
123
in model 2 (e'(NTRSTSTD)) is not significantly different from zero. As was shown in
chapter 3, the more ambiguity people feel about their risk estimate, the less likely they
should be to buy the certified apple, since, if they are unsure about their risk estimate.
they will be unsure about the risk reduction they will be getting with the certified apple.
W
Several goodness of fit measures suggest that the three PROBIT models
presented in Table 5-10 are all significant. A likelihood ratio test of the hypothesis that
all the coefficients are zero shows that all three PROBIT models are highly significant.
The x2 value for each model (67.71 with 10 degrees of freedom for model 1, 68.16
with 11 degrees of freedom for model 2, and 67.00 with 10 degrees of freedom form
model 3) is well above the critical value at 0.001 level of significance. Furthermore, all
three models make over 70% correct predictions. Although the pseudo R2 are low for
all the models, we hesitate to emphasize this measure of goodness of fit as Greene
(1990) suggests that "values between zero and one have no natural interpretation"
(Greene 1990, 682).
Because all three models fare equally well in the goodness of fit measures, the
calculations of the willingness to pay for pesticide-residue certification is based on
model 1.“ This model uses the measure of ambiguity most closely related to the
definition of ambiguity used throughout the text. The results in the other models do
‘5 The second-stage estimation for both model 1 and model 2 is presented in Table
5-11.
124
provide, however, further evidence of the importance of ambiguity in the consumers’
choice calculus.
T ifi i
In all three models, the "type" of certification one received did not influence the
decision to buy the apple. The decision to purchase the certified apples was not
influenced by whether one received the "produced without pesticides" certification, or
the ”this product meets federal standards for pesticide residues” certification.“
This result could stem from two factors. The presence of the certification,
regardless of the exact content, may be enough to assuage peoples’ fears about pesticide
residues. In other words, there is a sort of "warm glow” effect (Andreoni 1990,
Kahneman and Knetsch 1992). Pe0ple may be willing to pay for the certified apple as
long as it gives them some improvement over the regular apples. The ”quantity” of the
improvement may not be important as long as people think they are getting some
reduced risk.
‘6 The effect of the label was tested using a dummy variable. CERT, which was 0
if the respondent received the ”meets federal standards" label, and 1 if the respondent
received the ”produced without pesticides” label, in the PROBlT model. These results
are presented in Appendix 5-2.
125
Factors Influencing the Demand for Certified Apples
This section discusses the results of the second stage demand estimation based on
the first stage in model 1.“ The results in Table 5-11 showthat the most important
factor determining the quantity of certified apples, once the decision to buy them has
been made, is household size. In model 1 and model 2 household size (HHSIZE) is
significant and positive. The models also demonstrate that the demand for certified
apples is price and income inelastic. Price and income influence the choice to buy the
certified apples, but once the decision to buy them has been made, people seem to buy
the quantity based on the size of the household.
The risk associated with the certified apple (1r‘) is significant in both models,
although it is not the expected sign. We expected that higher perceived risk on the
certified apples would compel people to buy fewer certified apples. However, this
finding probably reflects the measurement of 1r' rather than 1r‘. The survey did not ask
respondents to directly state their perceptions of the risks from certified apples. Instead
it asked for perceptions of the risk from regular apples and the percentage of perceived
risk reduction when apples were labeled. The risk from the certified apples is
calculated based on this information. Because perceived risk reduction is generally
small, the constructed variable may be overwhelmed by the influence of the risk of
regular apples, thus explaining the positive sign on a“.
‘7 Because the results of the PROBIT analysis showed that people did not
distinguish between the two labels when making their decision about certified apple, we
estimate the demand for all certified apples together. i.e., we do not distinguish between
the types of label.
Table 5-11:
126
Second-Stage Demand Equation with Ambiguity measured as e'(SURE)
MODEL 1 MODEL 1
Dependent QC QC
Variable:
Coefficient Coefficient
Variable (standard error) (standard error)
CONSTANT 15.40 19.52
(9.267) (8.99)
pr 22.18 11.03
(20.09) (18.75)
pc -27.65 ~15.11
(20.09) (18.79)
HHSIZE 3.68'” 3.41'”
(1.22) (1.18)
INCOME 0.0001 0.0001
(0.00001) (0.00001)
1" 30.22" 29.74”
(13.52) (13.02)
At -5.74 -5.50
(4.20) (3.63)
X 10.74 -0.95
(12.79) (12.45)
Log (L) -1889.38 -1894.23
Corrected std. 32.6 31.9
error
Mean of LHS 26.1 26.1
Adjusted R2 0.03 0.03
significant at the 10% level
significant at the 5% level
significant at the 1% level
127
Because the survey did not solicit the amount of ambiguity associated with the
certified apple, the models use ambiguity reduction as an associated measure of the
ambiguity. This variable is not significantly different from zero in determining the
quantity of apples purchased in either model.
Although the explanatory variables do not explain very well the variation in the
quantity purchased, we use the results of model 1 to calculate the willingness to pay for
certification as discussed in the next section. Because the approached used to estimate
consumer surplus depends on the results from both the PROBIT and the second-stage
estimation, the WTP calculations are strongly influenced by the factors that affect the
choice to buy certified apples.
Willingness to Pay for Certification
The estimates of the willingness to pay for certification are shown in Table 5-12
through Table 5-15. Table 5-12 shows the welfare measures on a "per pound" basis,
Table 5-13 shows the measures on a "per year" basis. The measures of willingness to
pay are based on the coefficients from Model 1 (both stages). The "base" row is the
willingness to pay for certification using the observed values for all variables. Based on
the data from this survey, we estimate that peOple are willing to pay $0.31 (32%) more
per pound for the certified apple than what they pay for the regular apple. This is
within the range of the estimates obtained in the van Ravenswaay and Hoehn study
(1991) where they estimated the average added price per pound was between $0.24 and
$0.38, depending on the exact content of the certification.
128
Table 5-12: . Willingness to Pay for Certification and Sensitivity Analysis‘58
(5 per pound, per household)
MODEL 1
(ambiguity measured with e')
SCENARIO
Base Case $0.31
Scenario 1 $0.28 (-10% from base)
Scenario 2 $0.32 (+3% from base)
Scenario 3 $0.31 (+0% from base)
Table 5-13: Willingness to Pay for Certification ($ per year, per household)
MODEL 1
(ambiguity measured with e')
SCENARIO
Base Case $7.06
Scenario 4 $9.92 (+41% from base)
— E J
Table 5-14: Changes in WTP due to Changes in Baseline Risk
W
Willingness to Pay for Certification
(3 Per Pound)
Baseline Risk
1" S 0.00001 $0.31 (+0% from base)
0.00001 < 1" 5 0.001 $0.30 (-3% from base)
1' > 0.001 $0.33 (+6% from base)
— M
6’ The analysis is conducted at the following prices:
3098 for regular apples
$1.04 for certified apples
See text for discussion of price selection
129
Table 5-15: Changes in WTP for Certification due to Changes in Sureness
[mm =1
Sureness Level
Willingness to Pay for Certification
(5 Per Pound)
Very Sure $0.34 (+9% from base)
Somewhat Sure $0.31 (+0% from base)
Somewhat Unsure $0.28 (-10% from base)
Very Unsure $0.26 (-16% from base)
The tables also show the estimates of willingness to pay for certification under
various scenarios. We use these results to discuss the marginal willingness to pay for
ambiguity and risk reduction. The scenarios apply the coefficients from the base model
to the actual values of the variables, while changing the values of the risk or ambiguity
variables. The scenarios are as follows:
Scenario 1.
Scenario 2.
Scenario 3.
Scenario 4.
Scenarios
All consumers are either "very unsure" or "sOmewhat unsure" about their
risk estimate
All consumers are either "somewhat sure" or "very sure" about their risk
estimate
All consumers believe a federal program to test and certify food for
pesticide residues would be either "very effective” or "somewhat
effective."
The chance of having an adverse health outcome someday because of
pesticide residues in food is reduced by 0.000001.
130
The average willingness to pay for the certification on apples under scenario 1 is
$0.28. In this scenario everyone is either "very unsure" or "somewhat unsure." This is
$0.03 (or 10%) less than the under the base scenario in which 29% of the respondents
were "very unsure" or "somewhat unsure." Baseline risk was not changed in this
scenario, indicating that perhaps people simply want to know what they are getting in
terms of safety when they buy apples. They get that with the certified apple; they may
not be getting it with the regular apple. The results conform with the hypothesis that
people with higher initial ambiguity should be willing to pay less for certification.
If all respondents felt sure about their estimates (they were either "somewhat
sure" or "very sure" as in Scenario 2) willingness to pay for the certification increases
by $0.01 (+3%) from the base case and $0.04 (+14%) from scenario 1 in which all
respondents are "somewhat unsure" or "very unsure."
Scenario 3 shows that if everyone were convinced of the effectiveness of a
federal program to test and certify food for residue levels, willingness to pay for the
certified apple would be the same as in the base case ($0.31 per pound).
To estimate the marginal value of a change in mean risk, we create a scenario in
which the lifetime risk of having an adverse health effect from pesticide residues in food
is reduced an additional one in a million (Scenario 4). We reestimate the willingness to
pay assuming that the change in risk is increased in the amount that would reflect this
change. The results in Table 5-13 show that people would be willing to pay an
additional $2.94 (+41%) per year for pesticide-residue certification on apples to acquire
131
that level of risk reduction.69 It should be noted that this risk reduction is a reduction
in the lifetime risk of having a health problem someday because of pesticide residues in
all food, not just the risk from pesticide residues on apples.70
Table 5-14 and Table 5-15 show that, as hypothesized. the willingness to pay for
certification increases with baseline risk and decreases with baseline ambiguity.
Conclusions
The results presented in this chapter indicate that ambiguity may be an important
factor influencing the consumption of certified apples. Perceptions of high baseline
ambiguity probably drive the choice to buy the certified apples. but once that decision
has been made, neither baseline ambiguity nor changes in ambiguity help determine the
quantity purchased.
Respondents are willing to pay positive amounts (on average about $0.31 per
pound, or a 32% premium) for certification that the apples they buy either meet federal
standards for pesticide residues or have been produced without pesticides. The amount
they are willing to pay seems to be influenced by the ambiguity as evidenced by the
variation in the willingness to pay across perceived ambiguity levels. Unsureness about
6’ The surplus is calculated on a "per fall" basis. We multiply this figure by 4 to
get annual figures. Although the quantity of apples bought varies by seasons and is
probably lower in seasons other than the fall, this calculation gives an upper-bound
estimate.
7° If we assume that apples are a constant proportion of the total diet, this figure
can be interpreted as the amount consumers are willing to pay for risk reduction in
apples that contributes to a total risk reduction from all foods of one in a million.
132
risk estimates for regular apples may avert people from purchasing the certified apples
since they cannot be sure about the risk reduction they are getting.
Baseline risk perceptions, as they are quantitatively measured here, may not
adequately capture peoples’ concerns about the risks from pesticide residues. People
seem to be worried about the effects of pesticide residues in food for children, as
evidenced by the fact that households with children under the age of 18 are more likely
to buy certified apples than other households. Although this should be captured by the
risk estimate it is probably difficult for people to quantitatively estimate the risks they
face from pesticide residues in food.
Measures of both baseline ambiguity and baseline risk are imperfect at best.
When we use "trust that the standard is being enforced" as a proxy for ambiguity, for
example, we are probably confounding two opposing forces. The trust in the standard
variable probably captures the uncertainty about the risk estimate, but it probably also
captures some of the perception of the change in risk associated with the certified apple.
That is, when people feel that the standard is being enforced, they probably also feel
that their risks are reduced (both baseline ambiguity and changes in risk are influenced
by trust).
The measurement problem is probably the biggest obstacle reSearchers face in
terms of trying to understand how risk and ambiguity perceptions affect the willingness
to pay for certification and for reduced risk and ambiguity.
CHAPTER 6: POLICY IMPLICATIONS AND CONCLUSIONS
The goal of this research was to understand the effect of ambiguity on consumer
decision making under uncertainty. To this end, several steps were necessary. First,
we needed to define ambiguity. We then had to decide on a conceptual model of
decision making under uncertainty that incorporates ambiguity and use the model to
develop hypotheses about the effect of both risk and ambiguity on willingness to pay for
pesticide-residue certification. We also needed to deve10p methods for measuring
ambiguity. This chapter summarizes the conceptual and empirical findings of the
research and develops policy implications based on these findings. It also discusses the
issues related to survey design and accurate measurement of ambiguity. The chapter
concludes with the future research needs this study has identified.
Conceptual Findings
This research defines ambiguity as uncertainty about the probability of an
outcome. It can be characterized by the "spread," or the variance, of the second-order
probability distribution of an outcome;" the higher the level of ambiguity one feels
about his/her probability estimate, the wider the second-order probability distribution.
In traditional models of decision making under uncertainty (EU, for example),
ambiguity is behaviorally insignificant. This results because people are assumed to use
7' Higher moments of the probability distribution such as the skewness and kurtosis
may also characterize ambiguity. Here we assume the probability distribution around
the mean probability is normally-distributed and is therefore fully characterized by its
mean and variance.
133
134
the Reduction of Compound Lotteries Axiom (RCLA) when processing multiple layers
of risk. That is, if there is a probability distribution for the various probabilities that
might characterize an outcome, the decision maker "reduces" the compound
probabilities to a single probability by taking the "weighted average" of the possible
probabilities, where the weights are the probabilities of those probabilities. The
"spread" of the second-order probability distribution is then not important. However,
several empirical and experimental studies, starting with Ellsberg (1961), have shown
that peoples’ decisions are affected by the presence of multiple layers of risk.
Following Segal (1987), this research assumes that decision situations that
include both risk and ambiguity can be considered 2-stage lotteries in which the first
stage is a lottery over the probability, and the second stage is a lottery over the
outcome. Probability "certainty equivalents" (CEs), which give the probability that
would make a decision maker just indifferent between facing the first-stage lottery over
the probability and facing the second-stage lottery with the CE probability, are used to
convert a two-stage lottery to a one-stage lottery.
Ambiguity then enters peoples’ objective function in two ways: (1) through the
perceived amount of ambiguity and (2) through the individual’s attitude toward
ambiguity (the shape of the weighting function, I). Here we assume in that decision
makers are ambiguity averse (f is concave), i.e., they prefer known to unknown
probabilities. Ambiguity has the effect of increasing the overall riskiness of a prospect,
but its effect on decision making is distinct from standard risk aversion because of the
shape of the weighting function.
135
We use Segal’s method of incorporating ambiguity into a model of decision
making to develop the hypotheses that (1) the willingness to pay for certification about
pesticide residues is positive, (2) the willingness to pay for residue certification
increases with baseline risk and decreases with baseline ambiguity, and (3) the marginal
value of a reduction in either risk or ambiguity is positive.
Ambiguity affects consumers’ choices about certified and regular apples through
both baseline, or initial, ambiguity and through the perceived difference in the
ambiguity associated with the two types of apples. This study hypothesized that people
with higher initial ambiguity (higher ambiguity associated with the regular apples)
would be willing to pay more for certified apples than people with lower initial
ambiguity. Higher initial ambiguity reduces the likelihood that one is actually getting
significant risk reduction from the certified apple.
Consumer choice is also affected by the perception of the change in ambiguity
that is associated with the certified apple. A second hypothesis was that people who
perceived more ambiguity reduction from the certified apple would be willing to pay
more for the certification than people who do not believe the certification reduces
ambiguity.
Furthermore, both baseline risk and changes in risk should affect consumers’
choices about certified apples. People who perceive higher baseline risk should be
willing to pay more to for certification about pesticide residues. And people who
perceive that they get more risk reduction from the certification. should be willing to
pay more than people who perceive lower risk reduction.
136
Empirical Findings
The conceptual framework deve10ped in chapter 3 suggests that consumers make
decision about purchasing apples in two-stages. First they decide whether to buy the
certified or the regular apples. Once that decision has been made, they decide how
apples to buy. This research uses Heckman’s two stage model to allow the factors that
affect choice to differ from the factors that affect quantity. 1
To empirically test the hypotheses developed in chapter 3, this research used the
results of a contingent valuation survey of 1000 Michigan households’ attitudes about
pesticide residues and purchase intentions for apples with and without pesticide-residue
certification.
The empirical results of the two-stage model show that people are willing to pay
approximately 31 cents per pound above what they pay for regular apples to get the
pesticide-residue certification. The regular apples and the certified apples are identical
except for the certification indicating that the apple has been tested and certified to
either meet federal standards for pesticide residues or to have been produced without
pesticides.
As predicted by theory, willingness to pay for certification decreases with
baseline ambiguity. People who are "very unsure" about their risk estimates are willing
to pay $0.08 less per pound for the certified apples (a 24% discount) than the people
who are "very sure." We cannot reject the hypothesis that willingness to pay for
certification decreases with baseline ambiguity.
Although the survey did not directly measure the change in ambiguity associated
with the certified apple, the evidence presented in this research suggests that reducing
137
ambiguity has value. When all respondents are sure of their risk estimates (a decrease
in ambiguity) WTP increases by $0.04 (+13%) per pound of apples relative to the case
where all respondents are unsure. These findings hold constant the level of initial risk
and the amount of risk reduction acquired from the certified apples. This evidence
suggests that some of what people are willing to pay for the pesticide-residue
certification stems from the uncertainty they feel about the potential health risks from
pesticide residues in food.
As the third hypothesis predicted, people are willing to pay positive amounts to
reduce the risks from pesticide residues. The empirical results suggest that people are
willing to pay up to $2.92 (+41%) per year to reduce the lifetime risk of an adverse
health outcome from pesticide residues by 1 in a million.
The PROBIT analysis of the factors influencing the decision to buy certified
apples suggests that ambiguity is important in the decision to buy certified apples.
People with higher baseline ambiguity (as measured for the qualitative risk estimate)
were less likely to buy the certified apples than those with higher initial ambiguity.
However, other factors such as the risk reduction associated with certification, the
reduction in risks that one can obtain from one’s own actions, the prices of regular and
certified apples, household income, and the presence of children under the age of 18
also influence the decision to purchase the certified apples.
Policy Implications
The empirical findings suggest that ambiguity is a potentially important concern
to consumers. There is little doubt that consumers feel some uncertainty about the risks
138
they face from pesticide residues in apples: 29% of the survey respondents said they
were either "somewhat unsure" or "very unsure" about their risk estimate. Similarly, a
large percentage of respondents said they do not trust that the government sets the same
standards they would, nor do they trust that once the standards are set, all foods meet
those standards.
There are several sources of ambiguity about the risks from pesticide residues
that need to be distinguished before any policy recommendations about ambiguity can be
made. Some ambiguity stems from the current scientific uncertainty about the health
effects of pesticide residues in food. There are varying expert opinions about the health
risks from pesticide residues. If consumers are bombarded with information on the
varying opinions, it is unlikely that their ambiguity will be reduced. To diminish
ambiguity from this source, policy makers could develop policies to foster scientific
consensus (by holding conferences, for example).
Another source of ambiguity for consumers is the lack of information they have
about the current levels of pesticide residues in food. Although this information is
theoretically obtainable, it is costly to obtain. However. currently known information
about the levels of pesticide residues in food could potentially be offered to consumers
in a clear and informative way that would reduce ambiguity. Similarly, information
about the current standards for pesticide residues in food, information about whether
those standards are being met, and information about the ways people can reduce the
amount of residues they ingest, would all reduce consumers’ ambiguity and increase
their welfare.
139
Consumers often receive conflicting information from the media about the health
risks from pesticide residues. Consumers are reassured. for example. that domestically
produced goods currently meet the standards for pesticide residues. Yet they are also
informed of the "circle of poison" in which pesticides that are banned for use in the US
return to consumers via imported produce. This conflicting information increases the
ambiguity people feel about their risk estimates.
The government also needs some "public relations" work to convince the public
that it is taking their best interests into account when making policy decision about
pesticides. Consumers need to feel confident that the standards for pesticide residues in
food reflect safe levels and not levels that protect special interests.
This research suggests that people are willing to pay significant amounts to
receive the signal that something has been done about pesticide residues to protect their
health. It does not seem to matter what the exact content of the certification is ("warm
glow" effect). The difference in willingness to pay across ambiguity levels and across
risk levels suggests that a simple policy stating that a product meets the current
standards for pesticide residues could increase consumer welfare by reducing both
ambiguity and risks. Although testing and certifying products would be required for
such a policy, the cost of such a policy would probably be low since most foods already
meet the standards.
Survey Design Issues
The contingent valuation survey used in this research was carefully designed to
ensure valid results about the willingness to pay for pesticide—residue certification. It
140
created a shopping scenario that was within the respondents’ realm of experience,
questions were worded with careful attention to how they would be interpreted, and
extensive pretests were conducted to guarantee that the survey generated meaningful
data.
An important aspect of this survey was that substitution possibilities were
accounted for by allowing respondents to choose between regular and certified apples.
This is a realistic choice scenario that a consumer might actually face someday, and is
an improvement over most surveys which consider only one product at a time.
The survey was also designed to minimize the likelihood that people would have
difficulty making decisions about the products offered. Most pe0ple buy apples; it is
easy for them to accurately predict their purchasing behavior.
Although contingent valuation surveys commonly ask respondents to directly
state their willingness to pay for specific goods or services, this research did not use this
approach. Instead, the survey asked respondents how many apples they would likely
buy under different scenarios. It The information is then used to estimate willingness to
pay for risk reduction. Since consumers are more likely to have to make market
choices than to be asked to pay directly for reduced risks, the market scenario is more
understandable, plausible, and meaningful to respondents.
Measurement Issues
It is difficult to conceptualize measures of ambiguity that capture the actual
ambiguity people feel, since ambiguity is defined as the "spread" of the probability
distribution. The variance of the second-order probability distribution of an outcome
141
would be the most appropriate measure. However. without knowing the shape of the
respondents’ second-order probability distribution. it is difficult to measure ambiguity.
The proxies developed in this research serve as indicators of ambiguity levels that can
be used in the absence of more accurate measures.
This research developed three proxies for "baseline" ambiguity that are likely to
be highly correlated with the actual "baseline" ambiguity. The first of these proxies is
peoples’ "sureness" about their estimate of the risks associated with regular apples. The
survey asked people how sure they were of their estimate of the probability that
someone from a household like theirs would have a health problem someday because of
pesticide residues in food. We assume that people who are more "sure" about their
risk estimate have lower ambiguity. However, it is impossible to know the exact
relationship between actual ambiguity and its proxies.
The "sureness" proxy was also used to measure peoples’ ambiguity about their
qualitative perception of risk.
The third proxy developed was peoples’ trust that the standards for pesticide
residues for food were being met. Although this variable does not measure ambiguity,
it influences ambiguity - the less trust one has that the standards are being met, the
higher the ambiguity one will have about the risk estimate. This variable probably also
influences the mean probability: people who do not trust that the standard is being met
have a higher mean risk than those who do.
These proxies were developed for the baseline ambiguity associated with the
certified apples. The survey did not develop any explicit measures of the changes in
sureness or trust about foods meeting the standard when moving from the regular to the
142
certified apples. It did, however, construct an indicator of the change in ambiguity by
interacting sureness level with perceptions of the effectiveness of a federal program to
test and certify products for pesticide residue levels. This variable was significant in
determining the choice of apple (although it did not help explain the quantity
consumed). This finding highlights the need to further develop measures of consumers’
perceptions of the change in ambiguity.
When we use proxies for ambiguity. we are not guaranteed that we are
measuring ambiguity. .When we ask people about their level of trust that foods meet the
standards for pesticide residues, for example, it is difficult to know that we are not
getting a measure that captures both risk and ambiguity.
Future Research Needs
There is enough accumulated theoretical and empirical evidence to indicate that
ambiguity is an important input into consumer decision making under uncertainty. The
biggest challenge to researchers working in this area is to develop measures of
ambiguity that accurately reflect the variance of the second-order probability distribution
of an outcome. "Laboratory-type" experiments probably offer the most promising
means to explore methods for building a probability distribution for probabilities. It is
unlikely that direct questioning of consumers will yield satisfactory measures of
ambiguity.
It would be helpful _to understand more clearly the factors influencing
consumers’ ambiguity levels; we could then target trouble spots more effectively when
143
trying to reduce ambiguity. This hinges. however. on the development of accurate
measures of ambiguity.
This research has assumed a relationship between the proxies for ambiguity and
actual ambiguity. More research needs to be done to strengthen these assumptions.
This research was designed to examine whether ambiguity is a factor affecting
WTP for certification. The next step is to develop a measure of peoples’ perceptions of
the ambiguity associated with certified apples. The resources needed to develop this
information were not available at the time of the survey. However, the results of this
research suggest that it would be worthwhile to do so. It is necessary to have more
information on these perceptions in order to accurately estimate WTP to reduce
ambiguity.
All of the results of this research rest on the assumption that people are
ambiguity averse. There are some situations. however. where ambiguity might be
desirable. Methods for gauging peoples’ attitudes toward ambiguity need to be
developed in order to more effectively capture the effects of ambiguity on decision
making.
The results are also contingent upon the assumption that the RDEUWA model is
the appropriate representation of decision making. Future research needs to explore the
implications of introducing ambiguity into other generalized utility models to see if the
derived hypotheses are the same as the ones developed here.
Another approach to understanding the importance of ambiguity is to assume that
it affects the formulation of risk estimates. Bayesian analysis, for example, offers an
approach to thinking about how ambiguity enters the process of risk estimation (V iscusi
144
and O’Connor 1984, Eom 1994) Unlike the classical approach to uncertain events
where one tests hypotheses using the sampling distributions of a selected test statistic,
Bayesian analysis relies on posterior distributions. One starts with a prior probability
distribution over various states of nature. The decision maker then revises this
distribution in light of new observations. The revised distribution yields the posterior
distribution, which is then used in combination with a utility function as the basis for
decision making. Probability distributions over probabilities could be revised in light of
new information about risks to arrive at posterior probability distributions over
probabilities.
Ambiguity may also be a "dimension" of risk that is not captured by the mean
risk assessment, but is nonetheless an important input into the decision making process
under uncertainty. Dimensions of risk are characteristics of the risk scenario that affect
choice. Other than ambiguity, these might include the size of the severity of the
consequences, immediacy of effect, and voluntariness of risk (Slovic, et al 1990). This
should be explored as an approach to thinking about ambiguity.
This research has focused solely on the implications of ambiguity for consumer
choice. However, ambiguity also plays an important part in other areas concerned with
risk. Techniques for risk assessment, for example, are affected by how ambiguity is
handled. Often, each step of the assessment is fraught with uncertainty. It is not clear,
for example, exactly how to extrapolate findings in laboratory animals to what we
should expect in humans. Analysts have methods of handling uncertainty about risk
estimates. They use "safety factors" when assessing the risks. They might multiply the
assessed risk by 100, or 1000, just to be "safe." for example. However. all the
145
uncertainty about the risk assessment is obscured when scientists report a "crisp"
estimate of risk. There is no acknowledgement of the uncertainty about the risk
assessment. More research needs to be undertaken about how best to assess ambiguous
risks, and how to communicate ambiguity about risk estimates to the public.
Ambiguity promises to be an important area of research for gaining insights into
decision making under uncertainty. Only with further research on the best way to
conceptualize, measure, and incorporate it into a model of choice can we expect to more
thoroughly understand its importance in choice.
APPENDICES
146
.o>=_moq on EB 8:835 £5 6v 0.: new .9238 m_ 0885 Lo 5:3: 3535: 05 483088? ecu 038:3 m_ a. o:
>Um
e t u m... 80 3
$9-2.eéz ...5 ;
fi.@
..4 No.
C®NK€T$+¥VW+tweetetkvfiu...‘ + box». + thymus;
..Is st \FQ NI.. .9 an ..a HFQ on 1 o... N
esmSc seize nefhézso- 2 o o EYE: 5-: o o ..
nv—fllbm l mu cog—33:8 if.
v—xallbme t E uBmSSE 2a 03
:é
as.
csmSoorf .er or? bbmceéés s + Efmvmgééa ..s i Q
:9: Z— MUZ<=U 4. m9 HDJ<> x— new .0358 m_ 2.82: .3 .EE: 3535: 05 433088? was 98:8 fl ._ .:
>Um
u s .n fl ..9 050:3
§UIE.Q.Q.5>©
Amie
ewm 71.x 7‘ wm 7..
bwkc+ek+ Arumvkl bwmv\_bw_>+okvk.m_mws:.
swm .9 1 a... .o?‘ Tx 7.. .9 1 . <>
Ows}+nkv\©§UI V< QQQV>+—¢®NV\I ¢~va.\—Ow~>+©ukv\m_mwl A>UI I: QR»?
._ _
=05": >Um t
CO
07‘ _I\ 7 A <3
6% .. 1786 states: .. Sam ex gmvboi + tweuwso xterm. 5
an 52:3 B 83.5 <3:QO 2:
>P—DUEE Z— HUZ<=U < hO His—«5 Adi—9%: HIP ”urn «GA—2mg:
148
.9568 on EB 56853 £5 6v 3.; can .3338 fl 0585 he .EE: .2558 05 ..aoEoEzim can 028:8 m. .L .2
La: make 5283
be as N... .5 .
5.-..» . be. -3: bee ... . axioms. Biases; -
«chm 1km ~... e Nana . U
5... Theme perish EsoivsAoi. (newsmvsoiv ..:~-m obsess
”9.30:8 05 .3 can." on. 25 8 woo: oB .m_ 22,—.
_.>119mtm
he saw as new oB 62305.60 .8 9.5» .5 am: 8:82 5 women—.0 me Soto 05 e293 8.
ZO—B<.m—.—.¢MU 10,.— LE ZO Em: nil—4mg: kc POM—hum ”3:. unrn «Ga—Zak:
APPENDIX 5-1: SURVEY INSTRUMENT
[ Please note: all skip patterns and split-sample variations have been removed for better
readability of survey ]
Hello, is this (confirm phone number)
My name is and I am calling from the Center for Survey
Research at Michigan State University.
We are conducting a study on behalf of the Department of Agricultural
Economics at Michigan State University regarding pesticide residues in food.
According to our sampling design, 1 need to speak to the person in the
household, who is at least 18 years of age. who does the most grocery shopping.
Would that be you?
Before we begin, let me tell you that any information you give me will be kept
strictly confidential. Let me also tell you that this interview is completely voluntary.
Should we come to any question that you don’t want to answer, just let me know and
we’ll go on to the next question.
Throughout the study, we will be asking for your opinions in terms of the food
you buy for your household. Your household includes yourself, your dependents, and
persons with whom you share income and household living expenses. We will also be
talking about pesticide residues in food. Pesticides are used to control insects, diseases,
and other pests that spoil food. To protect consumers’ health, the federal government
sets standards that limit the amount of pesticide residues that may be in food sold in the
US.
Q1 In terms of pesticide residues, how confident are you that the food your
household eats is safe?
Would you say you are completely confident. mostly confident. somewhat
confident, or not confident at all? .
< 1> COMPLETELY CONFIDENT
<2> MOSTLY CONFIDENT
<3> SOMEWHAT CONFIDENT
<4> NOT CONFIDENT AT ALL
< 8) DON’T KNOW/NO OPINION
<9> REFUSED/NO ANSWER
149
Q2
3a
150
Suppose someone from a household like yours did nothing at all to reduce or
avoid pesticide residues in food. What do you think the chances would be that
someone from that household will have a health problem someday because of
pesticide residues in their food?
Would you say there is no chance. it is very unlikely. somewhat unlikely,
somewhat likely, very likely, or certain to happen?
< 1> NO CHANCE
<2> VERY UNLIKELY
<3> SOMEWHAT UNLIKELY
<4> SOMEWHAT LIKELY
<5> VERY LIKELY
<6> CERTAIN TO HAPPEN
<8> DON’T KNOW/NO OPINION
<9> REFUSED/NO ANSWER
How sure are you that there is of a health problem because of pesticide
residues in food? (blank is filled with respondents answer from Q2)
Would you say you are very sure, somewhat sure. somewhat unsure. or very
unsure?
< 1> VERY SURE
<2> SOMEWHAT SURE
<3> SOMEWHAT UNSURE
<4> VERY UNSURE
< 8) DON’T KNOW/NO OPINION
<9 > REFUSED/NO ANSWER
Now I would like to get a better idea about what you mean by (blank is
filled with respondents answer from 02). Suppose there were a million people
from households like yours who did nothing to reduce or avoid pesticide residues
in food. What do you think the chances are that a person from one of these
households would have a health problem someday because of pesticide residues
in food?
Would you say 1 person in a million. 1 in 100.000. 1 in 10.000. 1 in 1.000. 1
in 100, or 1 in 10?
<1) 1 PERSON IN A MILLION
<2> 1 IN 100,000
<3> 1 IN 10,000
Q5
Q7
Q8
151
<4> 11N1,000
<5) 1 IN 100
<6> 1 IN 10
< 8> DON’T KNOW/NO OPINION
<9> REFUSED/NO ANSWER
How sure are you that the chances are (blank is filled with respondents
answer from Q4).
Would you say you are very sure, somewhat sure, somewhat unsure, or very
unsure?
< l > VERY SURE
<2> SOMEWHAT SURE
<3> SOMEWHAT UNSURE
<4> VERY UNSURE
< 8) DON’T KNOW/NO OPINION
<9> REFUSED/NO ANSWER
Is there anything you usually do to reduce or avoid pesticide residues in your
food? (open-ended, field coded)
<1> YES:specify
< 5 > NO
<98> DON’T KNOW
<99> REFUSED
Suppose someone did the same things you usually do to reduce or avoid
pesticide residues in food.
What percent do you think that would reduce the chances of a health problem
happening some day?
<0-1005 ENTER EXACT PERCENT
<998> DON’T KNOW/NO OPINION
<999> REFUSED/NO ANSWER
Next, I am going to read you two statements, please tell me to what extent you
agree or disagree with each of them.
QlO
Q11
152
I trust the federal government to set the same standards that I would set in
limiting the amount of pesticide residues allowed in food.
Would you say you strongly agree, somewhat agree, somewhat disagree, or
strongly disagree?
<1> STRONGLY AGREE
<2> SOMEWHAT AGREE
<3 > SOMEWHAT DISAGREE
<4> STRONGLY DISAGREE
< 8 > DON’T KNOW/NO OPINION
<9> REFUSED/NO ANSWER
I trust that once the federal standards are set, all the food I buy will meetthose
standards.
Would you say you strongly agree, somewhat agree, somewhat disagree, or
strongly disagree?
< l > STRONGLY AGREE
<2> SOMEWHAT AGREE
< 3 > SOMEWHAT DISAGREE
<4> STRONGLY DISAGREE
< 8) DON’T KNOW/NO OPINION
< 9 > REFUSED/NO ANSWER
Now suppose that all foods met the federal standard for pesticide residues [were
produced without pesticides]. What percent do you think that would reduce the
chances of a health problem happening someday to people who currently do
nothing to reduce or avoid pesticide residues in food?
<0-100> ENTER EXACT PERCENT
<998> DON’T KNOW/NO OPINION
<999> REFUSED/NO ANSWER
Suppose foods were tested and certified to meet federal standards for pesticide
residues [to have been produced without pesticides]. Which of the following
organizations do you feel would be the most effective in conducting the tests and
issuing certificates?
Would you say the federal government, the state government, a well known
consumer’s group, or some other organization?
Q12
153
<1> FEDERAL GOVERNMENT
<2> STATE GOVERNMENT
<3> A WELL KNOWN CONSUMER’S GROUP
<4> SOME OTHER ORGANIZATION:SPECIFY
<98> DON’T KNOW/NO OPINION
<99> REFUSED/NO ANSWER
Suppose the federal government did the testing and certifying. How effective do
you think such a program would be in ensuring that foods had no pesticide
residues above federal standards?
Would you say very effective, somewhat effective, somewhat ineffective, or
totally ineffective? ‘
< 1) VERY EFFECTIVE
<2> SOMEWHAT EFFECTIVE
< 3 > SOMEWHAT INEFFECT IVE
< 4 > TOTALLY IN EFFECTIVE
< 8> DON’T KNOW/NO OPINION
<9> REFUSED/NO ANSWER
[ Note: half the sample received question 10, 11, and 12 with "foods were produced
without pesticides" replacing "foods met federal standards."]
Q13
Q14a
Q14b
Suppose someone from a household like yours had a health problem someday
that resulted from the current levels of pesticide residues in food. In your
opinion, what would the health problem most likely be? (Open-ended, field
coded)
Next I would like to ask a few questions about where you get your information
about the health risks of pesticide residues.
In the past 6 months have you gotten information about the health risks of
pesticide residues from a television program?
<1> YES
<5> NO
< 8) DON’T KNOW/NO OPINION
<9> REFUSED/NO ANSWER
In the past 6 months have you gotten information about the health risks of
pesticide residues from your doctor or health specialist?
Q14c
Q14d
Ql4e
Ql4f
0142
154
<1> YES
<5> NO
< 8) DON’T KNOW/NO OPINION
<9> REFUSED/NO ANSWER
In the past 6 months have you gotten information about the
pesticide residues from an article in a magazine?
YES
<5> N0
<8> DON’T KNOW/NO OPINION
<9> REFUSED/NO ANSWER
In the past 6 months have you gotten information about the
pesticide residues from a newspaper?
YES
<5> NO
< 8) DON’T KNOW/NO OPINION
<9> REFUSED/NO ANSWER
In the past 6 months have you gotten information about the
pesticide residues from a health newsletter?
<1> YES
<5> NO
<8> DON’T KNOW/NO OPINION
<9> REFUSED/NO ANSWER
In the past 6 months have you gotten information about the
pesticide residues from a radio program?
(1) YES
<5> NO
<8> DON’T KNOW/NO OPINION
<9> REFUSED/NO ANSWER
In the past 6 months have you gotten information about the
pesticide residues from family, relatives, or friends?
health
health
health
health
health
risks of
risks of
risks of
risks of
risks of
Q14b
Q15
Ql6
Q17
155
<1> YES
(5) NO
< 8) DON’T KNOW/NO OPINION
<9> REFUSED/NO ANSWER
In the past 6 months have you gotten information about the health risks of
pesticide residues) from any other sources? (open ended, field coded)
< 1> YES: SPECIFY
Next, I am going to read you several statements. In these statements, the term
’plants and animals’ refers to plants and animals produced for food. Please tell
me to what extent you agree or disagree with each of them.
If plants and animals were not protected in any way from insects, diseases, or
other pests, the supply of food available to me would decrease.
Would you say you strongly agree. somewhat agree. somewhat disagree, or
strongly disagree?
< 1> STRONGLY AGREE
<2> SOMEWHAT AGREE
< 3 > SOMEWHAT DISAGREE
<4> STRONGLY DISAGREE
< 8> DON’T KNOW/NO OPINION
< 9 > REFUSED/NO ANSWER
If plants and animals were not protected in any way from insects. diseases, or
other pests, the food available to me would not look as good as it does now.
Would you say you strongly agree, somewhat agree, somewhat disagree, or
strongly disagree?
< l > STRONGLY AGREE
<2> SOMEWHAT AGREE
< 3 > SOMEWHAT DISAGREE
<4> STRONGLY DISAGREE
< 8> DON’T KNOW/NO OPINION
<9> REFUSED/NO ANSWER '
If plants and animals were not protected in any way from insects, diseases, or
other pests, the price of food available to me would increase.
QlS
Q19
Q20
156
Would you say you strongly agree. somewhat agree. somewhat disagree, or
strongly disagree?
(I) STRONGLY AGREE
<2> SOMEWHAT AGREE
<3> SOMEWHAT DISAGREE
<4> STRONGLY DISAGREE
< 8> DON’T KNOW/NO OPINION
<9> REFUSED/NO ANSWER
There are many equally effective ways other than using pesticides to protect
plants and animals from insects, diseases. or other pests.
Would you say you strongly agree. somewhat agree. somewhat disagree, or
strongly disagree?
< 1) STRONGLY AGREE
<2> SOMEWHAT AGREE
< 3 > SOMEWHAT DISAGREE
<4> STRONGLY DISAGREE
<8> DON’T KNOW/NO OPINION
<9> REFUSED/NO ANSWER
It is more expensive to use other ways of protecting plants and animals from
pests than it is to use pesticides.
Would you say you strongly agree, somewhat agree, somewhat disagree, or
strongly disagree?
< l > STRONGLY AGREE
< 2 > SOMEWHAT AGREE
< 3 > SOMEWHAT DISAGREE
<4> STRONGLY DISAGREE
<8> DON’T KNOW/NO OPINION
<9> REFUSED/NO ANSWER
The scientific community can be trusted to be truthful about what they know
about health risks from pesticide residues.
Would you say you strongly agree, somewhat agree. somewhat disagree, or
strongly disagree?
Q21
Q21a
Q2113
157
<1> STRONGLY AGREE
<2> SOMEWHAT AGREE
<3> SOMEWHAT DISAGREE
<4> STRONGLY DISAGREE
< 8 > DON’T KNOW/NO OPINION
<9> REFUSED/NO ANSWER
The health risks associated with current levels of pesticide residues in food are
well known and understood by the scientific community.
Would you say you strongly agree, somewhat agree, somewhat disagree, or
strongly disagree?
< 1> STRONGLY AGREE
<2> SOMEWHAT AGREE
<3> SOMEWHAT DISAGREE
<4> STRONGLY DISAGREE
<8> DON’T KNOW/NO OPINION
<9> REFUSED/NO ANSWER
Food labeled as organic means the food is grown without pesticides.
Would you say you strongly agree, somewhat agree, somewhat disagree, or
strongly disagree?)
< 1) STRONGLY AGREE
<2> SOMEWHAT AGREE
< 3 > SOMEWHAT DISAGREE
<4> STRONGLY DISAGREE
. <8> DON’T KNOW/NO OPINION
<9 > REFUSED/NO ANSWER
All food that is labeled as organic has been certified by a reputable laboratory to
have been organically grown. '
Would you say you strongly agree, somewhat agree, somewhat disagree, or
strongly disagree?
< 1> STRONGLY AGREE
<2> SOMEWHAT AGREE
< 3 > SOMEWHAT DISAGREE
<4> STRONGLY DISAGREE
Q22
Q23
Q23a
Q23b
024
Q2421
158
< 8> DON’T KNOW/NO OPINION
<9> REFUSED/NO ANSWER
The next few questions are about your food shopping routine.
How often is the grocery shopping done in your household? (open ended, field
coded)
In the past year, has your household bought any fresh apples?
<1> YES
<5) NO
<8> DON’T KNOW/NO OPINION
<9> REFUSED/NO ANSWER
When you buy fresh apples, do you usually buy them individually, by the
pound, by the bag, by the peck, or by the bushel? (open ended, field coded)
< l > INDIVIDUAL
< 2 > POUNDS
< 3 > BAGS
< 4 > PECK
< 5 > BUSHEL
<0> OTHER (SPECIFY)
<98> DON’T KNOW/NO OPINION
<99> REFUSED/NO ANSWER
How many individual apples or pounds are usually in a bag?
< 1-997 >
<998> DON’T KNOW/NO OPINION
<999> REFUSED/NO ANSWER
About how often does your household buy fresh apples in the fall? (open ended,
field coded)
When you buy fresh apples in the fall, on average, how many apples do you buy
each time?
<0-997 >
<998> DON’T KNOW/NO OPINION
159
(999 > REFUSED/NO ANSWER
[ Note: 024 and 024a were asked with "winter," "spring." and "summer" replacing
fall.]
Q28
029
029a
029C
Now, suppose it is next fall and you are planning to buy some fresh apples. The
quality of all fresh apples is what you normally expect. Apples sold loose and
prepackaged are all the same price per pound. The prices of all fresh fruits
other than apples are what you normally expect.
How many apples of your usual variety would you buy if all fresh apples were _
_ (blank filled with one of several prices) per pound?
Now suppose you could also buy apples of your usual variety that have been
tested and certified by the federal government to have no pesticide residues
above federal standards [to have been produced without pesticides]. Fresh fruits
other than apples are not certified. The certified apples are per pound
compared to (blanks filled with one of several price combinations) per
pound for the regular apples.
Would you buy certified apples, regular apples, some of both, or none at all?
< 1 > REGULAR APPLES
<2> CERTIFIED APPLES
<3> SOME OF BOTH
<4> NONE AT ALL
<8> DON’T KNOW/NO OPINION
<9> REFUSED/NO ANSWER
How many of the certified apples would you buy (blank filled with one of
many prices) per pound?
<0-997>
<998> DON’T KNOW/NO OPINION
<999> REFUSED/NO
How many of the regular apples would you buy (blank filled with one of
several prices) a pound?
[ Note: Split sample variation of 029, 029a, and 029c: "tested and certified to have
been produced without pesticides.]
031
031a
031b
031C
032
033
160
The last few questions are for statistical purposes only. We need the
information to compare your Opinions with the other households we are
interviewing across Michigan.
How many people in your household are under 5 years of age?
< 0-97 >
<98> DON’T KNOW/NO OPINION
<99> REFUSED/NO ANSWER
How many people in your household are between 5 and 18 years of age?
< 0-97 >
<98> DON’T KNOW/NO OPINION
<99> REFUSED/NO ANSWER
(Including yourself), how many people in your household are between 19 and 64
years of age?
<0-97 >
<98> DON’T KNOW/NO OPINION
<99> REFUSED/NO ANSWER
(Including yourself), how many people in your household are over 64 years of
age?
<0-97 >
<98> DON’T KNOW/NO OPINION
<99> REFUSED/NO ANSWER
Respondent’s gender:
< 1) MALE
< 5 > FEMALE
< 8) DON’T KNOW/NO OPINION
<9> REFUSED/NO ANSWER
What is your age?
<18-100>
161
<998> DON’T KNOW/NO OPINION
(999 > REFUSED/NO ANSWER
034 What is the highest grade of school you have completed?
035
<0> GRADE SCHOOL ONLY
< 1> DID NOT FINISH HIGH SCHOOL
<2> HIGH SCHOOL OR GED
<3> VOCATIONAL OR TECHNICAL SCHOOL
<4> SOME COLLEGE
<5> COLLEGE GRADUATE (BA, BS)
<6> SOME GRADUATE OR PROFESSIONAL SCHOOL
<7> GRADUATE DEGREE (PHD, MD. MA. MBA)
<98> DON’T KNOW/NO OPINION
<99> REFUSED/NO ANSWER
To get a picture of people’s financial situations, we need to know the general
range of incomes of all respondents we interview. Now. thinking about your
household’s total annual income before taxes from all sources (including your
job) in 1991, did your household receive $45,000 or more in 1991?
<0-7>
<3> NO
<4> YES
<98 > DON’T KNOW/NO OPINION
<99> REFUSED/NO ANSWER
Q35a Was it $30,000 or more?
<3> YES
<2> NO
<98> DON’T KNOW/NO OPINION
<99> REFUSED/NO ANSWER
035b Was it $20,000 or more?
<2> YES
<1) NO
<98> DON’T KNOW/NO OPINION
<99> REFUSED/NO ANSWER
035C
035d
162
Was it $10,000 or more?
<1> YES
<0) NO
<98> DON’T KNOW/NO OPINION
<99> REFUSED/NO ANSWER
Was it $50,000 or more?
<5> YES
<4> NO
<98> DON’T KNOW/NO OPINION
<99> REFUSED/NO ANSWER
035e Was it $60.000 or more?
<6> YES
<5> NO
<98> DON’T KNOW/NO OPINION
<99> REFUSED/NO ANSWER
035f Was it $70,000 or more?
<7> YES
<6> NO
<98> DON’T KNOW/NO OPINION
<99> REFUSED/NO ANSWER
APPENDIX 5-2: PROBIT RESULTS WITH DUMMY VARIABLE "CERT"
Table AS-l: PROBIT Results with Dummy Variable "CERT"
MODEL 1 MODEL 2 MODEL 3
Dep Var: Z Z Z
Coefficient Coefficient Coefficient
Variable (standard error) (standard error) (standard error)
—
or, 0.22 0.15 0.21
(0.51) (0.51) (0.51)
p, 2.228" 2.19“ 2.07“
(0.60) (0.60) (0.60)
p, ~2.66"' 32.62" -2.45"'
. (0.54) (0.54) (0.53)
SCHOOL 0.06“ 0.06“ 0.06"
(0.03) (0.03) (0.03)
INCOME 0.00001“ 0.00001“ 0.00001“
(0.000003) (0.000003) (0.000003)
UND18 0.42“" 0.40“ 0.39"
(0.12) (0.12) (0.12)
1" 0.31 0.29
(0.26) (0.26)
f(NCVU) -0.42"
(0.21)
f(VLCT) 0.12
(0.15)
At(%) 0.52” 0.54" 0.48”
(0.23) (0.23) (0.23)
Aa(%OWN) -0.54" -0.48" -0.43'
(0.24) (0.24) (0.24)
¢'(SURE) 0.04
(0.13)
¢'(QUAL) -0.24'
(0.14)
('(N'I'RSI’SI'D) 0.18
(0.12)
CERT -0.005 0.013 0.04
(0.12) (0. 12) (0. 12)
% of buying min 72% 72% 73%
Log (L) -291.06 -290.96 -295.83
Pseudo R’ .10 .10 .10
% correct 74 74 74
indictions
‘ significant at the 10% level
” significant at the 5% level
‘” significant at the 1% level
Note: Model 3 uses I'(SUSL) as the base and t'(NCVU) (dummy variabe that has a value of one if
respondent answered "no chance" or "very unlikely" to 02) as an explanatory variable. The models in the
text use f(NCVU) as the base. This does not change the insignificance of the variable CERT.
163
BIBLIOGRAPHY
BIBLIOGRAPHY
Andreoni, .l. "Impure Altruism and Donations to Public Goods. A Theory of Warm
Glow Giving" Eggngmigs Jgarnal 1100 (1990): 464-477.
Becker, Selwyn W., and Fred 0. Brownson. "What Price Ambiguity? Or the Role of
Ambiguity in Decision-Making." laugaaLgLfiofitjgaLL-ngmy 72 (February
1964):62-73.
Blaylock, James R., and Noel Blisard. "U. S. Cigarette ConSumption: The Case of Low-
Income Women." Ameflcaa 1911:1131 9f Aggigal anal Egongmjgs 74 (August
1988):698-705.
Carson, R. "Constructed Markets." In Maaaaring 111: Damand fgr Enyjrgnmanta]
Mint, eds. .l. Braden and C. Kolstad, 121-162. Amsterdam: Elsevier, 1991.
Camerer, Colin, and Martin Weber. "Recent Developments in Modeling Preferences:
Uncertainty and Ambiguity." MW 5 (1992):325-370.
Cragg, John G. "Some Statistical Models for Limited Dependent Variables with
Application to the Demand for Durable Goods." W 39 (September
197 l):829-84-4.
Curley, Shawn P., and J. Frank Yates. "The Center and Range of the Probability
Interval as Factors Affecting Ambiguity Preferences." r i n v
andflumamflecisinnhocessss 36 (1985): 273-287
Einhorn, Hillel 1., and Robin M. Hogarth. "Ambiguity and Uncertainty in Probabilistic
Inference." W 92 (1985): 433-460.
Einhorn, Hillel 1., and Robin M. Hogarth. "Decision Making Under Ambiguity."
W 59 (1986):S225-S250.
Ellsberg, Daniel, "Risk, Ambiguity, and the Savage Axioms," MW
Economics, 75 (1961):643-669.
164
165
Born, Young Sook. "Pesticide Residue Risk and Food Safety Valuation: A Random
Utility Approach." Juurna! 9f Agrigultural Ecunumig§ 76 (November 1994): 760-
771.
Fellner, William. "Distortion of Subjective Probabilities as a Reaction to Uncertainty."
Quanerly Juurna] 9f Egunumicg 75 (l96l):670—689.
Food Marketing Institute. END ' nSIm r Atitude an
Washington DC: Food Marketing Institute. 1990.
Freeman, A. Myrick. -u-. - -u- . M o .. . .. -_rc- . -° l-u
amLMetugds. Washington, DC: Resources for the Future. 1993.
Gardenfors, P., and NE. Sahlin, "Unreliable Probabilities, Risk Taking, and Decision
Making." Syuthess 53 (1982):361-386.
"Decision Making With Unreliable Probabilities," WW
We! 36 (1983): 240-241
. eds. WW Cambridge:
Cambridge University Press, 1988.
Gilboa, Itzhak. "Expected Utility with Purely Subjective Non-Additive Probabilities."
WWW 16(1987):65-88.
Greene. William H. WWW!)
, Bellport: Econometric Software, Inc., 1992.
Haines, Pamela S., David K. Guilkey, and Barry M. Popkin. "Modeling Food
consumption Decisions as a Two-Step Process." Ame:i_cau_luu_ma]_uf
AsflcultmLEmmmiss 70 (August 1988): 543-552 '
Hammitt, James. ’ ' Wi
81:33. Santa Monica: The RAND Corp. R-3447-EPA, 1986.
Handa, J. "Risks, Probability, and a New Theory of Cardinal Utility." 1951mm
29l'lti21192n2mx 35 (1977) 97- 122
Haneman, W. Michael. "Qlality and Demand Analysis." In Wu
W ed Gordon
Rausser. New York. North-Holland, 1982.
Hazen, Gordon B. "Subjectively Weighted Linear Utility." W 23
(1987): 261-282.
166
Heclonan, J.A. "The Common Structure of Statistical Models of Truncation, Sample
Selection and Limited Dependent Variables and a Simple Estimator for Such
Models." Annals uf Eggnumig and Smial Maaguramant 5 (1976):475-492.
Heckman, J.A. "Sample Selection Bias as a Specification Error." W 47
(January 1979): 153-161.
Hellerstein, Daniel. "Estimating Consumer Surplus in the Censored Linear Model."
LaugLEuuuurujgs 68 (February 1992):83-92.
Hirshleifer, Jack, and John G. Riley. The Analytigs 9f Ungartainty and lufurujatjun.
Cambridge: Cambridge University Press, 1992.
Hogarth, Robin M., and Hillel J. Einhorn. "Venture Theory: A Model of Decision
Weights." W 36 (July 1990):?80-803.
Hogarth, Robin M., and Howard Kunreuther. "Risk, Ambiguity, and Insurance."
l29rnal_2f_Bisls_ansl_ul199rtainm 2 (April 1989):5-35.
Jeffrey, R. C. MW 2d ed. Chicago: University of Chicago Press.
1983. Cited in P. Gardenfors and N. E. Sahlin. eds. W
Wm, 9. Cambridge: Cambridge University Press, 1988.
Jones-Lee, Michael. "The Value of Changes in the Probability of Death or Injury."
Womb: 82 (July/August 1974): 835-849
Kahn, Barbara E., and Rakesh K. Sarin. "Modeling Ambiguity in Decisions Under
Uncertainty." MW 15 (September 1988):265-272.
Kahneman, D., and J.L. Knetsch. "Valuing Public Goods: The Purchase of Moral
Satisfaction." n n m . 22
(l992):57-70.
Kahneman, D., and A. Tversky. "Prospect Theory: An Analysis of Decision Under
Risk." W 47 (March 1979):263-291.
Kmenta, Jan. W. New York: Macmillan Publishing Company,
1986.
Levi, Isaac, "On Indeterminate Probabilities." mm. 71 (July
l974):391-418.
. Isaac, "Ignorance, Probability and Rational Choice." 532112219 53 (1982):387-
417.
167
Loomes, Graham, and Robert Sugden. "Regret Theory: An Alternative Theory of
Rational Choice Under Uncertainty." Tha Egpnpmig Journal 92 (December
1982):805-824.
Luce, R.D., and H. Raiffa. "Individual Decision Making Under Uncertainty." In
Dagisipn, ngpapjlity, and Uiill'fl; Salagtad Rpadipgs, eds. P. Gardenfors and
NE. Sahlin. Cambridge: Cambridge University Press. 1988. First published in
W. John Wiley & Sons, Inc., 1957. PP. 275-306.
Machina, Mark. "Generalized Expected Utility Analysis and the Nature of Observed
Violations of the Independence Axiom." In pagisjpu. Egpb ability, an nu Utility;
W, eds. P. Gardenfors and N. E. Sahlin. Cambridge. Cambridge
University Press. 1988. First published in W
jljh ppry with pApIi lgatjpus, ed. B. P. Stigum and F. Wenstop. Dordecht: D.
Reidel Publishing Company. 1983, pp.‘263-293.
Marschak, Jacob. "Personal Probabilities of Probabilities." W 6
(1975):121-153.
McKenzie, George W. "Measuring Gains and Loses." MW 84
(1976): 641-646. Cited in Myrick A. Freeman. 1]th Maa§uggmaut pf
Euv iguuguautal and Bagpugga Valugsg :1];an an nu Mathous. 61. Washington,
D. C.: Resources for the Future. 1993.
__-_. WEB—5W Cambridge Cambridge
University Press, 1983. Cited in Myrick A. Freeman. Wampum
Wad—114.9222. 61 Washington
D. C: Resources for the Future, 1993.
McKenzie, George W., and I. F. Pearce. "Exact Measures of Welfare and the Cost of
Living." W 43 (1976): 465—468. Cited in Myrick A.
Freeman. l. u r II .I I .1V Ill‘ -1 I-.I. ’t' I r ' V-. In H
M21122. 61. Washington, D. C.: Resources for the Future, 1993.
_. "Welfare Measurement-A Synthesis." WIS—321121! 72
(1982): 669-682. Cited in Myrick A. Freeman. W
W. 61 Washington
D. C.: Resources for the Future, 1993.
National Research Council. W Washington
D.C.: National Academy Press, 1993.
Quiggin. J. "A Theory of Anticipated Utility." W9
Qrgaumuuu 3 (1982):323-343.
168
. Qauaralizag Expectad Utility Thaogy: The Rank-Dapandant Mpdgl. Boston:
Kluwer Academic Publishers, 1993.
Ramsey, F.P., "Truth and Probability." In The Foundation; of Mathamatiga aud cha:
Lpgical Esaays, ed. R.B. Braithwaite. New York: Harcourt Brace. 1931.
Reaume, David M. "Cost-Benefit Techniques and Consumer Surplus: A Clarifactory
Analysis." MW 28 (1973): 196- 211. Cited in Myrick A. Freeman.
1‘ - I nvirI-nm n 1 ..In Re I eValles: Th I .nI M hII ,
61. Washington, D. C.: Resources for the Future, 1993.
Satin, Rakesh K., and Robert L. Winkler. "Ambiguity and Decision Modeling: A
Preference-Based Approach." Jutlrnal pf Ri§k and Emerging 5 (1992):389-407.
Savage, L.J., W. New York: Wiley, 1954.
Segal, Uzi. "The Ellsberg Paradox and Risk Aversion: An Anticipated Utility
Approach" We 28 (February 1987): 175-202
Segal, Uzi. "Two-Stage Lotteries without the Reduction Axiom." W158
(March 1990):349-377.
Schmeidler, D. "Subjective Probability and Expected Utility without Additivity."
Ecuuumejflpa 57 (1989):571-587.
Slovic, P., B. Fischhoff, and S. Lichtenstein. "Rating the Risks." In M,
eds. Theodore S. Glickman and Michael Gough. Washington D.C.: Resources
for the Future, 1990.
Smith, Vernon L. "Measuring Nonmonetary Utilities in Uncertain Choices: the Ellsberg
Urn." WM 83 (1969):324-329.
Smith, V. Kerry, and William H. Desvousges. "An Empirical Analysis of the Economic
Value of Risk Changes." n I f 1i i l n 95 (1987): 89-114.
United States Department of Agriculture. WNW
B99211- Economic Research Service. November, 1993.
van Ravenswaay, Eileen. "Public Perceptions of Food Safety: Implications for
Emerging Agricultural Technologies." Staff Paper No. 92-96. Department of
Agricultural Economics. Michigan State University, 1992.
van Ravenswaay, Eileen, and John P. Hoehn. "Contingent Valuation and Food Safety:
The Case of Pesticide Residues in Food." Staff Paper No. 91-13. Department of
Agricultural Economics. Michigan State University. 1991.
169
van Ravenswaay, Eileen, and Jennifer Wohl. "Valuing Risk Reduction When Risks are
Ambiguous." Staff Paper No. 93-37. Department of Agricultural Economics.
Michigan State University, 1992. (Forthcoming in Valuing FQQd Safaty, ed.
Julie A. Casewell).
van Ravenswaay, Eileen, Jennifer Wohl. and John P. Hoehn. "Michigan Consumers’
Perceptions of Pesticide Residues in Food." Staff Paper No. 92-56. Department
of Agricultural Economics. Michigan State University, 1992.
Viscusi, W. Kip, and Wesley A. Magat. "Bayesian Decisions with Ambiguous Belief
Aversion".129ma1_9f_Risk_aNJJNsLtai_nw 5 (1992): 371 -387
Viscusi, W. Kip, Wesley A. Magat, and Joel Huber. "An Investigation of the
Rationality of Consumer Valuations of Multiple Health Risks." AND qulgua] pf
Eguupmicu. 18(4): 465-479. 1987.
Viscusi, W. K., and J..C O’Connor. "Adaptive Responses to Chemical Labeling: Are
Workers Bayesian Decision Makers?" W 74
(December 1984). 942-956.
von Neumann, John, and Oskar Morgenstern. Thapg pf Qamag and Egpnpmig
Bahavipr. 2nd ed. Princeton: Princeton University Press, 1947.
Weber, M., and C. Camerer. "Recent Developments in Modelling Preferences Under
Risk." Qfljppkmuu 9 (1987):129-151.
Willig, Robert. "’Consumers Surplus Without Apology." Amgr iggn Eggngmig 891/1911!
69 (1976). Cited in Myrick A. Freeman. W
W 61 Washington D C: Resources
for the Future, 1993.
Winkler, Robert L. "Ambiguity, Probability. Preference. and Decision Analysis."
MW 4 (1990:285-297.
Yen, Steven T. "Working Wives and Food Away from Home: The Box-Cox Double
Hurdle Model " Amed2n_l2umal_.2LAgri9lllnlLal_E9oN__i9§ 75 (November
1993): 884-895.
;. "gt ‘
. we
‘7 J"
“at”! 8'5
.1. ~ .
'1
51:5“. . ‘
.131173' ‘13”:
attire}? ""1"
.I. . "'1"
1' “‘2, e 1211:
. n gg-Tfi“ "
.1:
__ , .-
.~.-'1“' V
‘ . V. '
«a. t\ 7 3". -
. 'v
:, -
I
26% r
E’Ifiggur-Fi‘fl .
‘ yw‘wé ’ «,le
.
..I.-- ..
I- 1
xx.
W .m .3321)",- ‘15.? '1
#53:; 11737;“ 17.5. a ..‘r"
‘ .'V. 1 A'xyucy‘k
34.1 .1393.“
.9932?
I
s
«'3