Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

FOOD VALUES

JAYSON L. LUSK AND BRIAN C. BRIGGEMAN

Because of the potential malleability of stated and revealed preferences for new food technologies
and attributes, this research sought to determine consumers’ food value systems by utilizing recent
advances in best-worst scaling. Based on previous literature related to human values and food prefer-
ences, a list of eleven food values was compiled. Results reveal that on average the values of safety,
nutrition, taste, and price were among the most important to consumers, whereas the values of fairness,
tradition, and origin were among the least important; however, there was significant heterogeneity in
the relative importance placed on food values. Results indicate that food values are significantly related
to consumers’ stated and revealed preferences for organic food.

Downloaded from http://ajae.oxfordjournals.org/ at University of Western Ontario on May 28, 2014


Key words: best-worst scaling, maximum-difference scaling, organic, preferences.

Each time a new food technology is developed, text of the decision task (e.g., Fishhoff 1993;
a new brand advertised, or a food policy de- Kahneman and Tversky 2000) and that peo-
bated, there is an associated spike in research ple use trial-and-error learning to “discover”
eliciting consumer preferences for the partic- preferences (e.g., Shogren, List, and Hayes
ular product or attribute in question. Exam- 2000).
ples include numerous papers on genetically In demand theory it is assumed that when
modified food, hormone use in meat produc- a consumer evaluates two consumption bun-
tion, food irradiation, nutrition, ecofriendly dles, x and y, they can identify which bundle is
products, etc. When studying future food poli- preferred to the other or whether they are in-
cies and technologies, an important question is different. The consumer then takes this prefer-
whether researchers must begin anew and treat ence ordering, expressed via a utility function,
each new product or product attribute as inde- and maximizes utility subject to a budget con-
pendent. Do consumers have some stable set of straint. As described by Varian (1982, 1983),
meta-preferences or food-specific values that if a consumer prefers x to y in one time pe-
drive preferences for food product attributes? riod but then in another prefers y to x, then
Economists have traditionally refrained the consumer has violated the axioms of re-
from drawing a sharp distinction between val- vealed preference theory (assuming prices are
ues and preferences. The neoclassical view is unchanged). Such a violation implies behav-
that consumers make choices based on their ior inconsistent with maximization of a stable
preferences. In turn, observed choices can be utility function. This means, for example, that
used to infer people’s preferences. As pointed preferences measured in one time period are
out by McFadden (2001) and others, such con- no longer of use in predicting demand or calcu-
sumer theory is virtually tautological: pref- lating welfare changes in another time period.
erences explain choices and given price and The importance of such preference instability
income constraints, choices explain prefer- can be noted, for example, by the significant
ences. A critical underlying assumption in this amount of attention devoted to determining
framework is that preferences are complete whether there was a structural change in meat
and stable. Research, however, suggests that demand and whether the weak axiom of reveal
preferences, the construct that defines choice, preferences (WARP) had been violated (e.g.,
may not be as stable as posited by this theory. Chalfant and Alson 1988; Eales and Unnevehr
It appears that people’s choices and thus pref- 1993; Moschini and Meilke 1989). When in-
erences can be affected by the frame and con- terest lies in determining preferences for new
or unfamiliar goods, it is quite possible to ob-
Lusk is professor and Willard Sparks Endowed Chair and Brigge- serve the kinds of preference reversals dis-
man is assistant professor, both in the Department of Agricultural cussed above.
Economics, Oklahoma State University.
The authors thank Jaebong Chang for assistance with survey
These observations have led some social sci-
administration and data entry and three anonymous reviewers for entists (e.g., Hechter 1993), including some no-
helpful comments on the article. table economists (e.g., Becker 1976), to draw
Amer. J. Agr. Econ. 91(1) (February 2009): 184–196
Copyright 2008 Agricultural and Applied Economics Association
DOI: 10.1111/j.1467-8276.2008.01175.x
Lusk and Briggeman Food Values 185

a distinction between potentially malleable investigates the validity of measured values by


preferences (the construct defining choices be- investigating their relationship to stated and
tween market goods) and more stable val- revealed preferences for organic food.
ues. Becker (1976) refers to these values as
“underlying preferences.” He argues, “The
preferences that are assumed to be stable do Background and Development
not refer to market goods and services . . . but of Food Values
to underlying objects of choice. . . . These un-
derlying preferences are defined over funda- In his seminal work on the issue, Rokeach
mental aspects of life, such as health, prestige, (1973) defined a value as “an enduring belief
sensual pleasure, benevolence, or envy, that do that a specific mode of conduct or end-state of
not always bear a stable relation to market existence is personally or socially preferable
goods and services.” For example, the assump- to an opposite or converse mode of conduct
tion is not that people’s preference ranking of or end-state of existence. A value system is

Downloaded from http://ajae.oxfordjournals.org/ at University of Western Ontario on May 28, 2014


nongenetically modified food over genetically an enduring organization of beliefs concerning
modified food is stable, but rather the more sta- preferable modes of conduct or end-states of
ble preferences are for the outcomes that are existence along a continuum of relative im-
expected to disseminate from eating nongenet- portance.” He developed a set of eighteen
ically modified instead of genetically modified terminal values—values such as “a comfort-
food. able life,” “inner harmony,” “pleasure,” and
Although values are generally conceptual- “wisdom” that relate to desirable end states of
ized as relating to abstract end states of ex- existence and another set of eighteen instru-
istence, a prudent question is whether there mental values—values such as “ambitious,”
might exist a set of intermediary values that “honest,” and “self-control” that relate to de-
relate specifically to people’s food choices, val- sirable modes of conduct that lead to partic-
ues that are perhaps more stable than people’s ular end states. Rokeach proposed a method
preference ranking of a specific set of foods or to measure the relative importance of these
food attributes. The concept of such intermedi- values that involved a straightforward rank-
ary values is, in fact, well ingrained in the con- ing task. He and others have shown that
ceptual framework related to means-end chain such values have considerable power in ex-
theory (Gutman 1982), a framework used by plaining variations in wide range of individual
psychologists and marketers that links people’s behaviors.
beliefs about concrete product attributes with More recent work, such as that conducted
the abstract values. by Schwartz (1992), has focused on other
A primary purpose of this article is to iden- value classification systems. Schwartz classi-
tify consumers’ food value systems. That is, we fied values into the following ten categories:
aim to identify people’s beliefs regarding the achievement, benevolence, conformity, hedo-
preferability of competing outcomes resulting nism, power, security, self-direction, stimula-
from food purchase and consumption along tion, tradition, and universalism. Some of these
a continuum of relative importance. One of value categories do not have a direct relation to
the primary impediments in empirically study- food (i.e., power and achievement), but many
ing an issue like food values is measurement. are of direct relevance. For example, people
Economists have historically shied away from for whom benevolence is an important value
directly measuring constructs that are not di- might be concerned about how their food con-
rectly observable (e.g., Colander 2007). A pri- sumption affects others in the supply chain;
mary concern is that people do not always people for whom hedonism is important may
use measurement scales identically, causing care a great deal about taste and visual ap-
difficulties in interpersonal and cross-cultural peal; people concerned about security will not
comparisons (e.g., Steenkamp and Baumgart- likely want to consume foods that will cause
ner 1998). Thus, a second primary purpose of illness and financial harm; and people who de-
this article is to apply recent advances in the sire stimulation (i.e., excitement and novelty)
field of best-worst scaling (sometimes referred will not likely be as concerned about new tech-
to as maximum-difference scaling) to value nologies, etc.
measurement. This approach permits the mea- Values research has gained widespread
surement of food values on a ratio scale by use in marketing research due to the de-
observing people’s choices of which values are velopment of means-end chain theory (e.g.,
most and least important. Finally, this article Gutman 1982), which argues that a set of core
186 February 2009 Amer. J. Agr. Econ.

underlying values motivate consumers’ pur- ucts are bought for private-safety reasons, gov-
chasing decisions. Buying a particular prod- ernment intervention in the market may be un-
uct or product attribute is conceptualized necessary, but if bought to help eliminate en-
as a means to obtain some desirable end vironmental externalities, intervention may be
state. Means-end chain theory, along with the warranted. Of course, firm advertising would
empirical approach of measuring means-end benefit substantially from knowing why con-
chains called laddering, is widely used by firms sumers buy certain products, such that adver-
for market segmentation and in developing tisements and promotional campaigns can be
advertising and promotional strategies (e.g., tailored to the value being served.
Reynolds and Gutman 1988; Gengler, Mul- Although means-end chain analyses typi-
vey, and Oglethorpe 1999). With the ladder- cally focus on linking product attributes to
ing approach consumers are shown a variety of end states of existence, in many cases these
competing products and asked to identify why are too abstract to be of practical relevance
they might choose one product over another. for food policy or advertising. However, the

Downloaded from http://ajae.oxfordjournals.org/ at University of Western Ontario on May 28, 2014


The typical answer involves the identification typical laddering approach elicits a number of
of a product attribute such as package size, nu- intermediary values along the chain to the fi-
tritional content, etc. Then, a monitor asks a nal end states of existence. It is here that we
question such as, “Why is that important to argue for the development of a set of food-
you?” Once an answer is provided, consumers specific intermediary values. Selecting the par-
are further encouraged to state more abstract ticular items to include as a potential set of
reasons for why an attribute is important to food-specific values is no easy task; many pre-
them, until they respond with a terminal value vious articles have asked consumers to rank
of the sort described by Rokeach (1973), or un- their concern for a variety of food technolo-
til the consumer cannot state a more abstract gies (e.g., Hwang, Roe, and Teisl 2005), food
reason. Common answers across consumers safety issues (e.g., Finn and Louviere 1992), or
are classified into means-end chains. Examples food attributes (e.g., Baker 1999). These ap-
of the use of means-end chain analysis related proaches, while useful in their own right, are
to food product marketing and food policy can too specific to be broadly applied when ex-
be found in the work of Bredahl (1999, 2001) plaining consumers’ choices across a variety
on genetically modified food and Makatouni of foods. Further, such studies often measure
(2002) and Padel and Foster (2005) on organic people’s preferences for technologies or con-
food. cepts that are relatively novel or unfamiliar
Measuring consumers’ values and means- and for which people do not likely have well-
end chains represents an explicit attempt to formed preferences.
identify why consumers prefer a particular The approach taken in this article was to re-
product over another. Some have argued that view the extant literature on consumers’ will-
economists should refrain from attempting to ingness to pay for food products and stud-
determine why consumers choose one product ies conducting food-related means-end chain
over another (e.g., Gul and Pesendorfer 2005). analyses to identify a set of food values that are
However, regulatory agencies must often de- likely to be relatively stable across time. The
termine the criteria or standards that foods idea is not to identify food attributes per se, but
must meet to be called organic, grass-fed, natu- rather more abstract attributes, consequences
ral, etc. Such decisions implicitly involve judg- and “end states” of food consumption that are
ment calls about the underlying values that potentially applicable in explaining choices be-
each of these claims aim to support. For ex- tween a wide range of food products. The list
ample, one set of criteria might be optimal if of food values that we compiled along with de-
consumers primarily buy organic because they scriptions shown to consumers in our empirical
believe it safer than traditional food. A differ- analysis is shown in table 1.
ent set of criteria would likely be optimal if Some of the items in table 1 could be con-
consumers desired organics because they be- sidered product attributes (e.g., convenience,
lieve such foods improve the environment. As nutrition, taste, etc.). However, in each case,
argued by Lusk et al. (2006), in the context of the value in table 1 is a more abstract con-
country of origin labeling, knowing why con- struct potentially representing numerous phys-
sumers prefer a particular attribute is critical ical product attributes. For example, several
in knowing whether a market failure exists and studies have investigated people’s preferences
thus whether public policy is needed. In the for fat content and vitamins in food; however,
case of organic food, for example, if such prod- all of these studies are related to the concept of
Lusk and Briggeman Food Values 187

Table 1. Food Values and Descriptions same as Rokeach’s (1973): “The overall pro-
cedure employed in selecting the two lists is
Value
admittedly an intuitive one, and there is no rea-
Naturalness (extent to which food is produced son to think that others working independently
without modern technologies) would have come up with precisely the same
Taste (extent to which consumption of the food is list of . . . values. It would be interesting to see
appealing to the senses) which values others might produce working in-
Price (the price that is paid for the food) dependently.”
Safety (extent to which consumption of food will
In the following we describe the approach
not cause illness)
Convenience (ease with which food is cooked we use to measure the relative importance
and/or consumed) consumers place on the values in table 1 and
Nutrition (amount and type of fat, protein, identify whether the measured values have ex-
vitamins, etc.) planatory power in explaining food choice.
Tradition (preserving traditional consumption

Downloaded from http://ajae.oxfordjournals.org/ at University of Western Ontario on May 28, 2014


patterns)
Origin (where the agricultural commodities were Data and Methods
grown)
Fairness (the extent to which all parties involved in To determine the relative importance con-
the production of the food equally benefit)
sumers place on the values in table 1, we uti-
Appearance (extent to which food looks
appealing) lized recent advances in best-worst scaling. The
Environmental Impact (effect of food production method was originally introduced by Finn and
on the environment) Louviere (1992), and the theoretical properties
of probabilistic, best-worst choice models were
recently elucidated by Marley and Louviere
(2005). The method is rapidly gaining popular-
nutrition. Although consumers may not have ity in business-marketing research (e.g., Saw-
well-defined preferences over Vitamin A con- tooth Software 2007) and has been recently
tent versus content of Vitamin B12, they are been applied to health care issues (Flynn et al.
much more likely to know whether nutrition, 2007) and to measure ethical beliefs in an inter-
generally, is more or less important than, say, national context (Auger, Devinney, and Lou-
price. That is, people’s relative value for “nutri- viere 2007). Lee, Soutar, and Louviere (2007)
tion” is likely to be much more stable and well- have even applied the best-worst approach to
reasoned than people’s relative preferences for measuring people’s overall life values. Their
Vitamin A versus Vitamin B12 or Omega 3 results suggested that the best-worst approach
fatty acid versus conjugated linoleic acid. We provides a better measure of people’s values
purposely refrained from selecting terms such than some frequently used rating approaches.
as “food quality,” which were too abstract to The best-worst method works as follows:
be operationally meaningful. Some of the val- consumers are shown a set of items and are
ues, such as origin, do relate to specific food at- asked to indicate which is best (or most impor-
tributes, but also have broader consequences tant) and which is worst (or least important).
with outcomes related to the community and Consumers make several repeated choices
with values related to ethnocentrism that are where the set of items varies across questions.
not likely captured by other values. As with Responses to the questions can be used to
Rokeach’s (1973) terminal values, the values in measure each item’s position on a continuum
table 1 can be classified as either personal (i.e., of the construct of interest—“importance” in
self-centered) or social (i.e., society-centered). this case. Figure 1 provides an example of
The values of tradition, origin, fairness, and en- one of the best-worst questions used in our
vironmental impact would fall into the latter application.
category. Best-worst scaling has several advantages
The values shown in table 1 are reasonably over other methods of measurement. A major
comprehensive in covering the breadth of is- difficulty with rating-based methods (where a
sues likely to motivate consumer food choice, person responds on a scale of 1 to 5, with 1 be-
and they undoubtedly capture the major val- ing not important and 5 being very important)
ues describing food choice. Still, we are under is that people are not forced to make trade-
no illusion that other researchers might have offs between the relative importance of issues.
generated a slightly different list had they un- Indeed, it is common for people to say all is-
dertaken the same task. Our response is the sues are “important.” Another problem with
188 February 2009 Amer. J. Agr. Econ.

Which of the following issues is most important and which is least important when you
purchase food? (check only one issue as the most important and one as the least important)
Most Least
Important Important
Taste
(extent to which consumption of the food is appealing to the senses)
Price
(the price that is paid for the food)
Safety
(extent to which consumption of food will not cause illness)
Convenience
(ease with which food is cooked and/or consumed)
Nutrition
(amount and type of fat, protein, vitamins, etc.)
Environmental Impact
(effect of food production on the environment)

Figure 1. Example of best-worst question

Downloaded from http://ajae.oxfordjournals.org/ at University of Western Ontario on May 28, 2014


rating-based methods is that different peo- values, and three contained eight values. Each
ple use the scale differently, with a 5 for one of the food values appeared six times across
person possibly representing a 4 for another. the twelve choice sets. In summary, each sur-
At best these measures provide measurement vey contained all twelve choice sets, and con-
on an ordinal scale. Best-worst scaling avoids sumers were asked to indicate the most impor-
these pitfalls. By having people choose the best tant and least important issue for each set as
and worst options, people are forced to de- shown in figure 1.2
cide which issues are more or less important,
and unlike rating scales, there is only one way
for people to respond to the question (with a Survey
choice). To implement the best-worst scaling approach,
Best-worst scaling is an extension of Thur- we created a mail survey that was sent to a ran-
stone’s (1927) paired comparison method, dom sample of 2,000 households in the spring
which has a long history in psychological re- of 2007. A postcard reminder was sent approxi-
search. In the paired comparison method peo- mately two weeks after the initial mailing. Two-
ple are simply presented two items and are hundred and twenty people responded to the
asked which is best (or more important). Best- survey, which after accounting for undeliver-
worst scaling provides much more information able addresses implies a response rate of about
than paired-comparisons, so much so, that it 11.4%. Some people did not answer all survey
is possible to construct individual-level scales questions or answered some questions incor-
and provide a measurement of the underlying rectly, leaving a total of 176 surveys available
construct on a ratio scale. for analysis. Table 2 provides summary statis-
As shown in table 1, there are eleven food tics for the respondents included in our analy-
values of interest. The challenge was how to al- sis. Most of the respondents were male (65%)
locate these eleven values to different choice and had a college degree (61%). Average age
sets in a survey. We used a 211 main-effects or- was about fifty-six years, and average house-
thogonal experimental design to assign each of hold income was about $74,000. As shown in
the eleven values to choice sets.1 The resulting the last four rows of the table, respondents
design consisted of twelve choice sets. Thus, were fairly well dispersed across the United
each respondent in our study answered twelve States.3
best-worst questions. Of the twelve choice sets,
six contained four values, three contained six

2
Given the literature on hypothetical bias, there may be some
1 concern with people’s hypothetical choices in the best-worst scaling
The experimental design we used is the same as that shown in
the appendix of Finn and Louviere (1992), with the food values in exercise. However, previous research has shown that in discrete
table 1 substituted in place of the societal issues studied by Finn choice conjoint applications, there is little or no hypothetical bias
and Louviere. This design, while having many advantageous prop- in marginal willingness to pay (e.g., Carlsson and Martinsson 2001;
erties, produces choice sets of unequal size (e.g., in one question Lusk and Schroeder 2004). Thus, the hypothetical nature of our
a respondent has to choose the best and worst out of four items, decision task may not be terribly problematic as all choices involve
whereas in a subsequent question a respondent has to choose the trade-offs between competing food values.
3
best and worst out of six items). Other experimental designs, such Despite the fact that our sample averages for education and
as balanced incomplete block designs, have the potential to over- income are higher than the averages for the U.S. population, we find
come this relative weakness such that the choice set sizes, while that demographic characteristics have little influence on people’s
varying across surveys, are constant within a survey. food values.
Lusk and Briggeman Food Values 189

Table 2. Characteristics of Survey Respondents


Standard
Variable Definition Mean Deviation
Gender 1 if female; 0 if male 0.351 0.479
Age Age in years 55.604 15.263
Degree 1 if obtained college degree; 0 otherwise 0.611 0.489
Child 1 if child under age of 12 living in household; 0 otherwise 0.166 0.373
Income Annual household income in $1,000s 74.044 37.567
West 1 if resides in West U.S. census region; 0 otherwise 0.201 0.402
Midwest 1 if resides in Midwest U.S. census region; 0 otherwise 0.339 0.475
South 1 if resides in South U.S. census region; 0 otherwise 0.282 0.451
Northeast 1 if resides in Northeast U.S. census region; 0 otherwise 0.178 0.384
Note: Number of observations is 176.

Downloaded from http://ajae.oxfordjournals.org/ at University of Western Ontario on May 28, 2014


Econometric Analysis (1)
When responding to each best-worst question, Prob( j is chosen best and k chosen worst)
consumers can be conceptualized as choos-
ing the two items that maximize the differ- e j −k
= J J .
ence between two items on an underlying scale l −m − J
l=1 m=1 e
of importance. If a choice set has J items, or
values in this case, then there are J(J − 1) The parameters j can be estimated by maxi-
possible best-worst combinations a person mization of the log-likelihood function based
could choose. The particular pair of items cho- on the probability statement in (1). That is, the
sen by the consumer as best and worst, then, dependent variable takes the value of 1 for the
represents a choice out of all J(J − 1) pos- pair of values chosen by the consumer as best
sible pairs that maximizes the difference in and worst, and a 0 for the reaming J(J − 1) −
importance.4 1 pairs of items in the choice set that were not
Formally, let j represent the location of choicen as best and worst. The estimated j
value j on the underlying scale of importance, represents the importance of value j relative
and let the true or latent unobserved level of to some value that was normalized to zero to
importance for individual i be given by I ij = prevent the “dummy variable trap.”
j + ε ij , where ε ij is a random error term. One of the primary objectives of this article
The probability that the consumer chooses, say, is to determine how values affect preferences
item j and item k, as the best and worst, re- for food, and as such, we need information
spectively, out of a choice set with J items, is on each individual’s values. Unfortunately, the
the probability that the difference in I ij and MNL described in equation (1) assumes that
I ik is greater than all other J(J − 1) − 1 pos- all individuals in the sample place the same
sible differences in the choice set. If the ε ij level of importance on each value (i.e., there is
are distributed i.i.d. type I extreme value, then no i subscript on j ). To overcome this weak-
this probability takes the familiar multinomial- ness of the MNL, we also estimated a random
logit (MNL) form: parameters model (RPL). In particular, let the
importance parameter for individual i and is-
sue j be specified as ˜ i j = ¯ j +  j i j , where
¯ j and  j are the mean and standard devia-
4
Although the approach described in his section is appealing
tion of j in the population, and i is a random
from a conceptual standpoint, some may find it difficult to im- term normally distributed with mean zero and
plement. Fortunately, there is a much simpler and straightforward unit standard deviation. Such a specification
(though somewhat less conceptually appealing) approach to deter-
mine the relative importance of food values. In particular, a log- implies that the importance of food value j is
interval scale of importance can be constructed simply by counting assumed to be distributed according to a nor-
the number of times a person chooses a particular value as most mal distribution with mean ¯ j and standard
important and subtracting it from the number of times a person
chooses the value as least important across the 12 choice sets. We deviation  j . Substituting ˜ i j = ¯ j +  j i j into
have created a technical appendix (Lusk and Briggeman 2008) to equation (1) yields a probability statement that
show the results from such a “counting” analysis and to illustrate
how cluster analysis can be used to group individuals based on depends on the random term in ij . Rather
these count-based value estimates. than attempting to explicitly integrate over
190 February 2009 Amer. J. Agr. Econ.

these random terms, following Train (2003), that an issue is picked as more important than
the model was estimated via simulation. In par- another.
ticular, the parameters were estimated by max- As shown by Train (2003) and Huber
imizing a simulated log-likelihood function, and Train (2001), once the parameters from
evaluated at 200 pseudorandom Halton draws the RPL are estimated, so-called individual-
for ij . The random draws are individual- specific estimates can be obtained by using the
specific, which takes into consideration the estimated parameters as a prior and using each
fact that each person answered twelve best- person’s actual choices to form an individual-
worst questions. Train (2003) provides more specific posterior estimate. These posterior es-
information on computational details for the timates of food values for each individual can
RPL. be compared with each person’s stated and re-
One of the underlying assumptions of the vealed preferences for organic food to deter-
RPL is that the variance of ε ij is assumed to mine if relationships among the variables exist.
equal one. It is possible, however, for the scale It is important to note that these Bayesian cal-

Downloaded from http://ajae.oxfordjournals.org/ at University of Western Ontario on May 28, 2014


to differ by people or alternatives (Louviere culations do not produce each person’s value
2001). The RPL model outlined above can be parameters per se, but rather as discussed thor-
also interpreted as an error-component model oughly by Train (2003), they are the means
where each alternative (or food value) has a of the conditional distribution (conditional on
separate variance (e.g., Train 2003), and as such the person’s actual choice), which are not nec-
the model accommodates differences in po- essarily the same as person’s actual coeffi-
tential scale over alternatives. Given the nor- cients. As Train (2003) shows, however, the dif-
malization of the variance of ε ij to one, how- ference between these two statistics becomes
ever, it is possible that the mean estimates small when people face even ten choice situa-
of j remain confounded with differences in tions and he further shows that such individual-
scale, and as such, it is important to recognize level Bayesian estimates improve the ability to
that the estimated population parameters re- predict the choices that each person actually
flect both differences in mean and scale. Fortu- makes. Thus, while we use the term individual-
nately, McFadden and Train (2000) have shown specific values, the estimates are, more pre-
that the RPL can approximate any underlying cisely, the mean of the parameter distribu-
random utility model, and as such, the RPL can tion conditioned on each individual’s actual
produce goods estimate of predicted probabil- choices.
ities, even if the there is some misspecification
in the choice model. Preferences for Organic Food
Given the potential confound with scale,
we calculate a “share of preference” for each To investigate the extent to which differences
food value, which is the forecasted probabil- in food values explain differences in prefer-
ity that each food value is picked as most ences for specific foods, the survey contained
important: both stated and revealed preference ques-
tions. In the stated preference question respon-
(2) share of preference for value j dents were asked to state the largest premium
that they were willing to pay for an organic
eˆ j loaf of bread over a loaf of bread produced
= J .
k=1 eˆ l through conventional, nonorganic means as-
suming both loaves were the same brand name.
These shares of preferences must sum to one A brief definition of organic food was provided
across all eleven food values. Equation (2) re- in case people were unaware of the mean-
ports the importance of the value j on a ratio ing of organic food. A payment-card method
scale, meaning that if one value has a share was used to elicit people’s willingness to pay.
value twice that of another value, it can accu- That is, people responded to the question by
rately be said that the former value is twice checking a box that contained several dol-
as important as the latter. It is important to lar amounts associated with the largest pre-
note that the calculated share of preference mium the consumer was willing to pay (e.g., $0,
for a food value will reflect both true impor- $0.01 to $0.49, $0.50 to $0.99, etc.). We investi-
tance of the value as well as relative uncer- gate whether stated willingness to pay for or-
tainty in the importance people place on the ganic food was related to food values by calcu-
value; however, the share of preference calcu- lating the correlations between each person’s
lations convey a key message—the probability stated willingness-to-pay premium and her/his
Lusk and Briggeman Food Values 191

individual-specific value shares of preferences zero), implying that there is significant hetero-
obtained from the RPL. geneity in the sample with regard to relative
In addition to this stated preference ques- importance of food attributes. Indeed, despite
tion, a revealed preference question was the fact that tradition and origin were the least
also asked. In particular, people were asked important food values, on average, there were
whether they had ever purchased an organic some people in the sample who picked these
food. Response categories were “yes,” “no,” issues as most important every time the is-
and “I don’t know.” For subsequent analysis sue appeared in one of the twelve best-worst
people who responded “yes” were classified questions.
as purchasers of organic food, and responses One interesting observation is that origin
of “no” or “I don’t know” were classified was found to rank last in average importance,
as nonpurchasers of organic food.5 To deter- but some previous studies related to beef have
mine whether food values relate to organic found origin of production to be very impor-
purchases, we compared the means of the tant to consumers (e.g., Menneke et al. 2007;

Downloaded from http://ajae.oxfordjournals.org/ at University of Western Ontario on May 28, 2014


individual-specific value shares of preferences Loureiro and Umberger 2007). Given recent
obtained from the RPL estimates for pur- attention to the issue of origin in policy cir-
chasers to the same statistic for nonpurchasers cles, a brief discussion on the discrepancy in
of organic food. these findings is in order. Previous research
We chose organic food because it is one for such as that conducted by Loureiro and Um-
which there are many potential reasons why a berger (2005) showed that U.S. consumers be-
person may buy organic food. Some have hy- lieve that U.S. beef is safer than foreign beef.
pothesized that food safety concerns are the As such, origin is likely to serve as a proxy
primary motivator behind organic food sales, for food safety and may even serve as a proxy
whereas others argue that the primary moti- for other attributes such as fairness, tradition,
vators are concerns for the environment or and taste. In previous country-of-origin stud-
inequity in the agricultural supply chain. By ies conducted by Menneke et al. (2007) and
linking food values to preferences for organic Loureiro and Umberger (2007), it is unclear
food, some of these competing hypotheses can whether the measured preferences for origin
be tested. It should be noted, however, that were a result of uncontrolled beliefs about dif-
there is nothing particularly special about or- ferences in food safety, tradition, fairness, en-
ganic food. In principle people’s food values vironmental impact, etc. across origins.
might have power in explaining preferences for Our approach explicitly measures prefer-
any type of food. ences for food safety relative to origin. Because
our study shows food safety as the most im-
Results portant food value, it is possible that if origin
is a proxy for food safety it too would also be
Table 3 shows the relative importance of each ranked highly on a list of importance if safety
of the eleven food values as estimated by the preferences were not fully disentangled from
MNL and the RPL models. The importance of preferences from origin. There are also differ-
each food value was estimated relative to ori- ences in the way importance is calculated in
gin (the least important value according to the previous conjoint studies and in the present
estimated models). Results reveal that safety, best-worst scaling approach. In the conjoint
on average, is the most important food value studies importance is measured by calculat-
and significantly more important than origin. ing the utility difference in the best and worst
Nutrition, taste, and price are the next most levels within an attribute, but in our study
important food values on average, whereas en- no specific attribute levels are specified, and
vironment, fairness, and tradition are the least- the explicit attribute levels used in the previ-
important food values. On average natural, ous conjoint studies may or may not be the
convenience and appearance were of interme- levels that endogenously come to mind for
diary importance but were significantly more the consumer. Finally, these previous studies
important, on average, than origin. Despite the dealt specifically with beef steaks, whereas our
averages, results reveal large standard devia- study deals with food in general and meta-
tions for the importance measures (note: the preferences.
MNL restricts the standard deviations to equal One of the difficulties in evaluating the im-
portance of each value that results from the
MNL and RPL models is that the estimates
5
Only 7% of respondents responded with “I don’t know.” themselves have no natural interpretation. To
192 February 2009 Amer. J. Agr. Econ.

Table 3. Relative Importance of Food Values


Econometric Estimates Shares of Preference
Value MNL RPL MNL RPL
∗a ∗
Safety 2.564 3.785 0.268 0.342
(0.088)b (0.157) [0.000] [0.308]
[0.000]c [2.106]
Nutrition 2.176∗ 3.192∗ 0.182 0.139
(0.087) (0.115) [0.000] [0.116]
[0.000] [0.428]
Taste 2.156∗ 3.104∗ 0.178 0.172
(0.086) (0.127) [0.000] [0.185]
[0.000] [1.251]
Price 2.021∗ 3.104∗ 0.156 0.215

Downloaded from http://ajae.oxfordjournals.org/ at University of Western Ontario on May 28, 2014


(0.087) (0.154) [0.000] [0.245]
[0.000] [1.834]
Natural 0.721∗ 0.983∗ 0.042 0.026
(0.082) (0.114) [0.000] [0.045]
[0.000] [1.182]
Convenience 0.595∗ 0.847∗ 0.037 0.017
(0.082) (0.108) [0.000] [0.022]
[0.000] [0.804]
Appearance 0.592∗ 0.909∗ 0.037 0.019
(0.079) (0.104) [0.000] [0.025]
[0.000] [0.836]
Environment 0.502∗ 0.709∗ 0.034 0.029
(0.080) (0.119) [0.000] [0.061]
[0.000] [1.544]
Fairness 0.090 0.294∗ 0.023 0.015
(0.078) (0.112) [0.000] [0.029]
[0.000] [1.254]
Tradition 0.071 0.168 0.022 0.021
(0.078) (0.127) [0.000] [0.052]
[0.000] [1.695]
Origin 0.000 0.000 0.021 0.005
[0.000] [0.004]
N individuals 176 176
N choices 2076 2076
Log likelihood −5129.4 −4405.4
Pseudo R2 0.383 0.471
a Oneasterisk (∗) implies mean importance of the value is statistically different from origin at p = 0.05 level.
b Numbers in parentheses ( ) are standard errors.
c Numbers in brackets [ ] are standard deviations.

remedy this problem, the last two columns in or origin) as most important. Nevertheless, it
table 3 report the shares of preference for appears that natural and the environment are
each value as calculated using equation (2). one-and-a-half to two times as important as
According to the RPL model, about 34.2% convenience and fairness, and more than five
of people would rate safety as the most im- times as important as origin.
portant food value. Interestingly, price has the Table 4 reports the correlation between
next highest share of preference, with 21.5% people’s food values, as determined by the
of people on average believing price to be the individual-specific estimates stemming from
most important value. Although nutrition and the RPL. None of the value estimates exhibit
taste are important to consumers, food safety correlations above 0.5, indicating that each
is about twice as important as these two values. of the values represents a unique construct.
Less than 3% of people would pick any of the Still, there are groups of values that appear
remaining food values (natural, convenience, to be related. For example, safety and taste
appearance, environment, fairness, tradition, exhibit negative correlations, meaning people
Lusk and Briggeman Food Values 193

Table 4. Pearson Correlations between Food Values from Individual-Specific RPL Estimates
Value 1 2 3 4 5 6 7 8 9 10
Safety (1) 1.00
Nutrition (2) 0.02 1.00
Taste (3) −0.10 −0.02 1.00
Price (4) 0.09 0.17 0.24 1.00
Natural (5) −0.04 −0.32 −0.11 −0.16 1.00
Convenience (6) −0.06 −0.04 0.10 0.12 −0.05 1.00
Appearance (7) −0.01 0.10 0.10 0.09 −0.18 0.08 1.00
Environment (8) 0.24 0.09 −0.18 −0.12 0.04 −0.06 −0.15 1.00
Fairness (9) 0.03 0.01 −0.23 −0.19 0.08 −0.29 −0.26 0.23 1.00
Tradition (10) −0.09 0.01 0.29 −0.03 −0.10 0.10 0.23 −0.24 −0.11 1.00

Downloaded from http://ajae.oxfordjournals.org/ at University of Western Ontario on May 28, 2014


who believe safety is important are less likely Table 5. Correlations between Shares of
to believe taste is important. One interesting Preferences for Food Values and Stated
finding is that the correlation between safety Willingness-to-Pay Premium for Organic
and naturalness is small and negative. Many Breada
have hypothesized that concern for modern
Pearson Spearman Rank
food technologies is a result of food safety con- Value Correlations Correlations
cerns, but the low correlation between these
two food values suggests otherwise. Interest- Safety 0.029 0.119
ingly, people who place a high importance Nutrition 0.167∗ 0.162∗
on fairness tend to place low values on the Taste −0.102 −0.070
more self-centered values of taste, price, con- Price −0.246∗ −0.306∗
venience, and appearance. Natural 0.234∗ 0.297∗
Convenience 0.020 0.031
The preceding results are useful in identi- Appearance 0.026 0.127
fying the relative importance of food values Environment 0.248∗ 0.187∗
in this somewhat small sample of individuals, Fairness 0.243∗ 0.108
but an important question is whether these Tradition −0.013 0.073
values exhibit explanatory power over these Origin 0.183∗ 0.147∗
people’s choices. To address this issue, con-
Number of observations = 176.
sumers were asked to state how much more a People were assigned a willingness to pay for organic bread of $0, $0.25,
they were willing to pay for organic bread $0.75, $1.25, $1.75, or $2.25 based on their response to an interval-censored,
over nonorganic bread. Table 5 reports Pear- payment card question.
b One asterisk (∗) implies the correlation between willingness to pay a
son and Spearman-rank correlation coeffi- premium for organic bread and the food value is significantly different from
cients between each person’s measured food 0 at p = 0.05 level or lower.
value as determined by the RPL individual-
specific shares of preference and the per-
son’s willingness-to-pay premium for organic and willingness to pay for organic bread, sug-
bread.6 Overall, results are consistent with a gesting that naturalness (defined as extent to
priori expectations. People for whom price is which food is produced without modern tech-
more important stated lower willingness to pay nologies) is a key motivating factor behind
for organic food than people for whom price people’s preferences for organic food. Inter-
is less important. There is a relatively large estingly, one of the factors often attributed
correlation between importance of naturalness to organics—safety—was not significantly re-
lated to stated willingness to pay a premium for
organic food, suggesting safety consideration
6
The willingness to pay for each person was simply determined may not be a driving force behind growth of
by assigning each person the average dollar amount in the range the organic food market. As expected, table 5
checked on the payment card amount (e.g., if the checked premium
range was $1.50–$1.99, the person was assigned a willingness-to- shows that the relative importance of the en-
pay value of $1.75). The Spearman-rank correlation coefficient ex- vironment was positively correlated with will-
plicitly accounts for the fact that this coding scheme only provides ingness to pay a premium for the organic food.
ordinal indications of willingness to pay. We have also estimated
interval-censored regressions, where interval-censored willingness Because responses to contingent valuation
to pay is the dependent variable and the importance of the food questions can be strongly influenced by the
values are independent variables. Overall, findings from these re-
gressions are qualitatively similar to the correlations reported in
frame of the question, including the hypothet-
table. ical nature of the inquiry, we also asked a
194 February 2009 Amer. J. Agr. Econ.

Table 6. Revealed Preferences for Organic fortunately, this means that the measured
Food and Shares of Preferences for Food preferences may be less stable than is often
Values postulated in theoretical models of consumer
decision-making. This article sought to iden-
Shares of Preference
tify a set of food values or meta-attributes
Have Not for which people may have more well-defined
Previously Previously preferences. Implicit in the approach is an at-
Purchased Purchased tempt to crack open the consumer’s “black-
Value Organic Organic box” and provide some insight into why
Safety 0.337 0.368 she/he chooses one product or attribute over
Nutrition 0.130 0.149 another.
Taste 0.169 0.163 Based on previous literature on human val-
Price 0.262∗ 0.177∗ ues and food preferences, a list of eleven food
Natural 0.018∗ 0.049∗ values was compiled. The relative importance

Downloaded from http://ajae.oxfordjournals.org/ at University of Western Ontario on May 28, 2014


Convenience 0.019 0.016 that people placed on each of the food val-
Appearance 0.021 0.017 ues was measured using the relatively new ap-
Environment 0.015∗ 0.024∗ proach of best-worst scaling, which involves
Fairness 0.009 0.015
Tradition 0.016 0.016
people stating the most and least important
Origin 0.005 0.006 issue out of a set of competing issues. Results
reveal that the values of safety, nutrition, taste,
Number of 64 112
observations
and price were among the most important to
consumers on average; however, we also found
Number of observations = 176. significant heterogeneity across consumers in
One asterisk (∗) implies the hypothesis that the mean values are the same for
people who have and who have not previously purchased organic is rejected
terms of the relative importance placed on
at the p = 0.05 level of significance or lower according to a two-tailed t-test. food values.
We found that measured values were signif-
icantly related to stated and revealed prefer-
revealed preference question by asking peo- ences for organic food. People for whom nat-
ple whether they had previously purchased or- uralness, fairness and the environment were
ganic food. Across the entire sample, 64% of more important were more likely to have pre-
respondents indicated they had previously pur- viously bought organic food and to state a
chased organic food. Table 6 reveals signifi- higher willingness to pay for organic food. The
cant differences in mean shares of preferences opposite was true for people that placed a rela-
of those people who have and have not pre- tively high level of importance on price. Over-
viously purchased organic food. Overall, re- all, the relative importance people placed on
sults are very similar to the correlations with naturalness, a value that was defined as the ex-
stated preferences reported in table 5. For ex- tent to which food is produced without modern
ample the mean share of preference for price technologies, seemed to have the largest influ-
was 26.2% for nonpurchasers but 17.7% for ence on stated and revealed preferences for
purchasers. That is, price was a less important organic food.
attribute for those people who purchased or- The research reported in this article rep-
ganic food than for those who had not pre- resents a first attempt at measuring people’s
viously purchased organic food. Purchases of values for food. More work is needed. First,
organic believed naturalness and the environ- if interest lies in identifying the relative im-
ment to be significantly more important than portance of food values in the population,
nonpurchasers of organic. work such as that presented here should be
repeated with a larger and more representa-
tive sample of consumers. Second, this study
Conclusions only compared the relationship between food
values and preferences for organic food. One
Agricultural economists have devoted a sig- benefit of measuring food values is that they
nificant amount of attention in recent years to should have significant explanatory power in
estimating people’s preferences for one food explaining choices between a wide range of
product attribute over another. Such stud- food products, and future research will reveal
ies often focus on measuring preferences for whether this is indeed the case. Third, addi-
goods and attributes for which people have tional work along the lines of that outlined
little prior knowledge or experience and un- above could help predict consumers’ responses
Lusk and Briggeman Food Values 195

to food policy and marketing changes. Small- Chalfant, J.A., and J.M. Alston. 1988. “Account-
scale laboratory research can be used to de- ing for Changes in Tastes.” Journal of Political
termine the link consumers make between Economy 96:391–410.
specific food attributes, such as use of biotech- Colander, D. 2007. “Edgeworth’s Hedonimeter and
nology, fat content, meat tenderness, etc. and the Quest to Measure Utility.” Journal of Eco-
food values. Then, knowledge of the distri- nomic Perspectives 21:215–25.
bution of food values in the population can Eales, J.S., and L.J. Unnevehr. 1993. “Simultaneity
aid in identifying how consumers, in aggre- and Structural Change in U.S. Meat Demand.”
gate, will respond to changes in food composi- American Journal of Agricultural Economics
tion. Fourth, a benefit of measuring food values 75:259–68.
is that these constructs should be more sta- Finn, A., and J.J. Louviere. 1992. “Determining the
ble over time; however, even concepts such Appropriate Response to Evidence of Public
as “culture,” which are thought to be rela- Concern: The Case of Food Safety.” Journal of
tively impermeable are known to change. This Public Policy and Marketing 11:12–25.

Downloaded from http://ajae.oxfordjournals.org/ at University of Western Ontario on May 28, 2014


suggests that there may be some benefit in Fischhoff, B. 1993. “Value Elicitation: Is There Any-
tracking a panel of consumers over time, mea- thing There?” In M. Hetchter, L. Nadel, and
suring how and when food values change. Fi- R.E. Michod, eds. The Origin of Values. New
nally, measuring differences in food values York: Aldine de Gruyter, pp. 187–214.
across countries may have significant utility Flynn, T.N., J.J. Louviere, T.J. Peters, and J. Coast.
in explaining international differences in food 2007. “Best-Worst Scaling: What It Can Do for
consumption patterns and responses to new Health Care Research and How to Do It.” Jour-
food technologies. nal of Health Economics 26:171–89.
Gengler, C.E., M.S. Mulvey, and J.E. Oglethorpe.
[Received July 2007; 1999. “A Means-End Analysis of Mother’s In-
accepted April 2008.] fant Feeding Choices.” Journal of Public Policy
and Marketing 18:172–88.
Gul, F., and W. Pesendorfer. 2005. “A Case for
Mindless Economics.” Working Paper, Dept.
References
of Econ., Princeton University. Available at:
Auger, P., T.M. Devinney, and J.J. Louviere. 2007. http://www.princeton.edu/∼pesendor/mindless.
“Using Best-Worst Scaling Methodology to pdf
Investigate Consumer Ethical Beliefs across Gutman, J. 1982. “A Means-End Chain Model
Countries.” Journal of Business Ethics 70:299– Based on Consumer Categorization Pro-
326. cesses.” Journal of Marketing 46:60–72.
Baker, G.A. 1999. “Consumer Preferences for Food Hechter, M. 1993. “Values Research in the So-
Safety Attributes in Fresh Apples: Market Seg- cial and Behavioral Sciences.” Hechter, M., L.
ments, Consumer Characteristics, and Market- Nadel, and R.E Michod (Eds.) The Origin of
ing Opportunities.” Journal of Agricultural and Values. New York: Aldine de Gruyter, pp. 1–
Resource Economics 24:80–97. 30.
Becker, G.S. 1976. The Economic Approach to Hu- Huber, J., and K. Train. 2001. “On the Similarity of
man Behavior. Chicago: University of Chicago Classical and Bayesian Estimates of Individual
Press. Mean Partworths.” Marketing Letters 12:259–
Bredahl, L. 1999. “Consumers’ Cognitions with Re- 69.
gard to Genetically Modified Foods. Results of Hwang, Y.-J., B. Roe, and M.F. Teisl. 2005. “An Em-
a Qualitative Study in Four Countries.” Ap- pirical Analysis of United States Consumers’
petite 33:343–60. Concerns about Eight Food Production and
——. 2001. “Consumer Attitudes and Decision Processing Technologies.” AgBioForum 8:40–
Making with Regard to Genetically Engi- 49.
neered Food Products: A Review of the Liter- Kahnemann, D., and A. Tversky. 2000. Choices, Val-
ature and a Presentation of Models for Future ues, and Frames. Cambridge, MA: Cambridge
Research.” Journal of Consumer Policy 21:23– University Press.
61. Lee, J.A., G.N. Soutar, and J. Louviere. 2007. “Mea-
Carlsson, F., and P. Martinsson. 2001. “Do Hypo- suring Values Using Best-Worst Scaling: The
thetical and Actual Marginal Willingness to LOV Example.” Psychology and Marketing
Pay Differ in Choice Experiments?” Journal 24:1043–58.
of Environmental Economics and Management Loureiro, M.L., and W.J. Umberger. 2005. “As-
41:179–92. sessing Consumer Preferences for Country-of-
196 February 2009 Amer. J. Agr. Econ.

Origin Labeled Products.” Journal of Agricul- Moschini, G., and K.D. Meilke. 1989. “Modeling the
tural and Applied Economics 37:49–63. Pattern of Structural Change in U.S. Meat De-
——. 2007. “A Choice Experiment Model for mand.” American Journal of Agricultural Eco-
Beef: What US Consumer Responses Tell Us nomics 71:253–61.
about Relative Preferences for Food Safety, Padel, S., and C. Foster. 2005. “Exploring the
Country-of-Origin Labeling and Traceability.” Gap between Attitudes and Behaviour: Under-
Food Policy 32:496–514. standing Why Consumers Buy or Do Not Buy
Louviere, J.J. 2001. “What If Consumer Experi- Organic Food.” British Food Journal 107:606–
ments Impact Variances as Well as Means? 25.
Response Variability as a Behavioral Phe- Reynolds, T.J., and J. Gutman. 1988. “Ladder-
nomenon.” Journal of Consumer Research ing Theory, Method, Analysis and Interpreta-
28:506–11. tion.” Journal of Advertising Research 28:11–
Lusk, J.L., and B.C. Briggeman. 2008. “AJAE Ap- 31.
pendix: Food Values.” Unpublished. Available Rokeach, M. 1973. The Nature of Human Values.

Downloaded from http://ajae.oxfordjournals.org/ at University of Western Ontario on May 28, 2014


at: http://agecon.lib.umn.edu/. New York: Free Press.
Lusk, J.L., and T.C. Schroeder. 2004. “Are Choice Sawtooth Software. 2007. “The MaxDiff/Web Sys-
Experiments Incentive Compatible? A Test tem Technical Paper.” Available at: http://www.
with Quality Differentiated Beef Steaks.” sawtoothsoftware.com/download/techpap/max
American Journal of Agricultural Economics difftech.pdf
86:467–82. Schwartz, S.H. 1992. “Universals in the Content
Lusk, J.L., J. Brown, T. Mark, I. Proseku, R. Thom- and Structure of Values: Theoretical Advances
son, and J. Welsh. 2006. “Consumer Behavior and Empirical Tests in 20 Countries.” Ad-
and Country of Origin Labeling.” Review of vances in Experimental Social Psychology 25:1–
Agricultural Economics 28:284–92. 65.
Makatouni, A. 2002. “What Motivates Consumers Shogren, J.F., J.A. List, and D.J. Hayes. 2000. “Pref-
to Buy Organic Food in the UK? Results erence Learning in Consecutive Experimental
from a Qualitative Study.” British Food Journal Auctions.” American Journal of Agricultural
104:345–52. Economics 82:1016–21.
Marley, A.A.J., and J.J. Louviere. 2005. “Some Prob- Steenkamp, J.-B.E.M., and H. Baumgartner. 1998.
abilistic Models of Best, Worst, and Best-Worst “Assessing Measurement Invariance in Cross-
Choices.” Journal of Mathematical Psychology National Consumer Research.” Journal of
49:464–80. Consumer Research 25:78–90.
McFadden, D. 2001. “Economic Choices.” Ameri- Thurstone, L.L. 1927. “A Law of Comparative Judg-
can Economic Review 91:351–78. ment.” Psychological Review 34:273–86.
McFadden, D., and K. Train. 2000. “Mixed MNL ——. 1983. “Non-Parametric Tests of Consumer Be-
Models of Discrete Response.” Journal of Ap- haviour.” Review of Economic Studies 50:99–
plied Econometrics 15:447–70. 110.
Mennecke, B.E., A.M. Townsend, D.J. Hayes, and Train, K.E. 2003. Discrete Choice Methods with Sim-
S.M. Lonergan. 2007. “A Study of the Fac- ulation. Cambridge, UK: Cambridge Univer-
tors that Influence Consumer Attitudes to- sity Press.
ward Beef Products Using the Conjoint Mar- Varian, H.R. 1982. “The Nonparametric Approach
ket Analysis Tool.” Journal of Animal Science to Demand Analysis.” Econometrica 50:945–
85:2639–59. 73.

You might also like