Lusk 2016

Download as pdf or txt
Download as pdf or txt
You are on page 1of 39

Applied Economic Perspectives and Policy Advance Access published January 11, 2016

Applied Economic Perspectives and Policy (2016) volume 0, number 0, pp. 1– 39.
doi:10.1093/aepp/ppv050

Submitted Article
How Meat Demand Elasticities Vary with Price,
Income, and Product Category
Jayson L. Lusk* and Glynn T. Tonsor

Jayson L. Lusk is Regents Professor and Willard Sparks Endowed Chair in the
Department of Agricultural Economics at Oklahoma State University and Glynn

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


T. Tonsor is associate professor in the Department of Agricultural Economics at
Kansas State University.
*Correspondence may be sent to: Jayson Lusk, 411 Ag Hall, OSU, Stillwater,
OK 74078; (405) 744 –7465; jayson.lusk@okstate.edu

Submitted 12 January 2015; accepted 3 September 2015.

Abstract As US beef and pork prices approached record high levels in 2014, indus-
try analysts expressed surprised at consumer response. Because the relative price
swings have occurred only recently, traditional approaches to demand analysis that
rely on historical data series may be less useful than is typically the case. Employing
one of the largest and longest-running choice experiments, we analyze data on
110,295 choices made by 12,255 consumers observed over a year-long time period co-
inciding with historically high meat prices. Our findings reveal nonlinear demands
for meat products, with demand being more inelastic at higher prices. Ground beef,
steak, and pork chop demands are more sensitive to changes in chicken breast price
than the reverse. Moreover, cross-price elasticities between disaggregate meat pro-
ducts shrink as prices rise. Consumers’ incomes significantly affect demand inter-
relationships. Higher income consumers are more likely to choose steak and chicken
breasts and are less likely to choose ground beef, chicken wings, and deli ham than are
lower income consumers. High-income consumers tend to be less responsive to own-
price changes and more responsive to cross-price changes than lower income consu-
mers. This analysis provides estimates of structural demand parameters that help
explain current meat expenditure patterns, and the results have implications for the
assumption of linearity often invoked in policy analyses.
Key words: beef, chicken, choice experiment, heterogeneous consumers,
meat demand, mixed logit, nonlinear elasticities, pork.
JEL Classification codes: C83, D12, Q11, Q18.

A confluence of events related to drought, high feed prices, disease, and


growing global demand led to unprecedented rises in the prices of beef and
pork in the United States. Figure 1 provides some perspective on the recent
changes. The Bureau of Labor Statistics reports ground beef prices were 54%
higher (an annual increase of over 10%) and steak prices were 43% higher

# The Author 2016. Published by Oxford University Press on behalf of the Agricultural and Applied
Economics Association. All rights reserved. For permissions, please e-mail:
journals.permissions@oup.com

1
Applied Economic Perspectives and Policy

Figure 1 Change in retail prices for six meat products from January 2010 to January 2015

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


Source: Bureau of Labor Statistics.

(an annual increase of 8.6%) in January 2015 as compared to January 2010.


Pork chop and ham prices were 30% and 39% higher than they were in 2010
despite recent declines. By contrast, in January 2015, chicken prices were only
about 10% higher than five previous (experiencing annual increases of only
1%–2%). Although meat demand is one of the most studied issues in the
agricultural economics literature (e.g., see the review in Unnevehr et al. 2010),
prior demand estimates were generally obtained using historical disappear-
ance data on aggregated products for “representative households” that lie
outside the range of prices recently witnessed.
Industry observers have expressed surprise about how consumers have
responded to recent price changes (Ishmael 2014). In particular, expenditures
for beef and pork have not fallen as much as some people expected given the
high prices. Industry analysts have asked “where is the tipping point” when
consumers will stop buying beef and pork (Rutherford 2014), but it may be that
demand elasticities are more nonlinear than previously realized. Moreover, rela-
tive price swings would have seemed to have favored chicken over beef and
pork, and yet there does not seem to be a high degree of substitution in the
current market environment. Such observations raise the possibility that cross-
price elasticities have changed or are lower at higher price levels. Finally, these
meat price swings have occurred during and following a serious and prolonged
economic recession. Amid contemporary discussions of income inequality have
come questions about which segments of the population have been most
affected by price increases and how substitution patterns might differ for the
rich and the poor. Questions of income and price-sensitivities in light of current
market outcomes diverging from expectations motivate this study.
One challenge faced in meat demand analysis is data quality and availabil-
ity (Brester and Wohlgenant 1993). Aggregate disappearance data are often
used, but these data may not be perfectly correlated with actual consumption,
are typically limited to aggregate categories, do not yield much information
on heterogeneity in preferences, and are necessarily constrained by historical
price ranges (Piggott and Marsh 2004; Tonsor, Mintert, and Schroeder 2010).

2
Meat Demand Elasticities Vary with Price, Income, and Product Category

Scanner data can provide more information on differential responses across


different consumer types in a more timely manner, but for fresh meat, these
data are typically “random weight” items, meaning one only observes the
number of units sold without knowing the products’ weights (Taylor and
Tonsor 2013). Either type of these data also pose challenges associated with
endogeneity, aggregation, unobserved promotions and quality variation,
measurement error, and frequency of observation (Lensing and Purcell 2006).
Many of these problems can be overcome with carefully designed choice
experiments. Choice experiments have their own drawbacks, but coupling
insights from timely choice experiment surveys with pre-existing insights
from secondary data might provide a deeper understanding of the current

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


meat demand environment.1
The overall purpose of this article is to determine consumer demand for
cut-specific meat products in light of recent market conditions. Specifically,
we aim to determine how: (1) own- and cross-price elasticities vary with the
market price, (2) consumers substitute between different cuts of meat at dif-
ferent price levels, (3) price sensitivities and substitution patterns vary across
consumers with different incomes, and (4) models assuming linearity in price
effects or restrictive substitution patterns mis-state consumer responses to
meat price changes. To address these issues, we bring to bear what is perhaps
one of the largest and longest-running choice experiments. We utilize data on
110,295 choices made by 12,255 consumers who responded to choice experi-
ments conducted over a year-long time period concurrent with the historical-
ly high beef and pork prices.
The next section briefly reviews the research on meat demand and situ-
ates our work in that body of literature. Then we describe our data and
econometric methods. The penultimate section discusses results, and the
last section concludes.

Synthesis of Existing Literature


Meat demand is complex and evolving as demand drivers emerge and
change over time. Multiple factors work to collectively shape meat demand
including traditional economic determinants such as relative prices and
income as well as nontraditional determinants such as health, nutrition,
and food safety information; changing product characteristics and new
product developments; and shifts in consumer demographics and lifestyles
(Tonsor, Mintert, and Schroder 2010). Over time, new dimensions of
demand may arise and the relative importance of previously examined
determinants may adjust in response to new information and economic con-
ditions. For example, the rise of high-protein diets, discovery of new health
risks, or shifts in household work-home time allocation may alter structural
demand parameters. Similarly, as meat prices change, so may the accuracy
of elasticity estimates derived from past market situations. Hence, ongoing

1
Possible drawbacks of choice experiments include the possibility for hypothetical bias, particularly in the
propensity to choose the “none” option, framing effects resulting from the need to specify particular
options and attribute levels and display them in a particular format, and selection effects that might arise
if the survey sample doesn’t match the population of interest. Overviews of the method and discussions of
advantages and disadvantages of choice experiments can be found in Louviere, Hensher, and Swait
(2000), Holmes and Adamowicz (2003), and Carlsson (2011).

3
Applied Economic Perspectives and Policy

demand estimation is important for informed decision making by industry


stakeholders and policy makers.
A large body of prior research has sought to investigate meat demand
determinants (e.g., see Bryant and Davis [2008] or Gallett [2010] for integra-
tive quantitative reviews of prior demand estimates). Unnevehr et al. (2010)
and Piggott and Marsh (2011) provide some historical context for this work,
much of which arose during a period in which researchers debated whether
the fall (rise) in per-capita beef ( poultry) consumption resulted from a
change in relative prices or a shift in preferences (i.e., a structural shift).
Prior research has identified a variety of potential demand shifters includ-
ing food safety and product recall news (Burton and Young 1996; Marsh,

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


Schroeder, and Mintert 2004; Piggott and Marsh 2004; Mazzocchi 2006),
health and dietary information (Brown and Schrader 1990; Chang and
Kinnucan 1991; Rickertsen, Kristofersson, and Lothe 2003), and advertising
(Brester and Schroeder 1995; Piggott et al. 1996; Kinnucan et al. 1997; Park
and Capps Jr. 2002).
More broadly, there are large literatures that have touched on some of the
same themes of the present analysis, albeit using traditional demand
systems with historical time series data. For example, researchers have
studied separability between disaggregate meat products and aggregate cat-
egories (Eales and Unnevehr 1988; Moschini, Moro, and Green 1994; Nayga
and Capps 1994), structural change (Eales and Unnevehr 1988, 1993;
Moschini and Meilke 1989; Rickertsen 1996; Davis 1997), product aggrega-
tion (Brester and Wohlgenant 1991; Coffey, Schroeder, and Marsh 2011;
Schulz, Schroeder, and Xia 2012), and functional form (Kastens and Brester
1996; Banks, Blundell, and Lewbel 1997; Beach and Holt 2001; Golan,
Perloff, and Shen 2001), among others.
More recently, the literature has expanded to include data from consumer
experiments and surveys, often choice experiments. Beyond examining
topical issues associated with credence attributes (e.g., Alfnes and Rickertsen
2003; Lusk, Roosen, and Fox 2003; Lusk, Norwood, and Pruitt, 2006), meth-
odological inquiries applied to meat demand issues have focused on hypo-
thetical bias (Lusk and Schroeder 2004), beliefs and consumer inferences
(Lusk et al. 2014), and the effects of question format (Carlsson et al. 2007; Gao
and Schroeder 2009; Tonsor 2011), just to name a few. It is important to note
that most of these studies provide “snap-shot” assessments, as they are con-
ducted only at one point in time with a given set of study participants.
While this existing work provides a wealth of insights into meat demand,
we are unaware of research conducted in the current era of high meat prices,
focusing on own- and cross-price elasticities, disaggregated meat products,
and income effects.

Methods
The data utilized in this study come from the Food Demand Survey (FooDS)
project that began in May 2013. FooDS tracks consumer preferences, food
expenditures, price expectations, and awareness and concern for a variety of
food issues. FooDS is an on-line survey with a sample size of at least 1,000 indi-
viduals each month. The survey is delivered on the 10th of the month unless
that date falls on a weekend, in which case it is moved to the next closest
Monday. The requisite sample size is typically acquired within three days. The

4
Meat Demand Elasticities Vary with Price, Income, and Product Category

Figure 2 Example choice experiment question

data collected do not constitute a panel because a new sample is drawn each
month; however, precisely the same questions are posed each month. The

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


survey is administered to an opt-in panel maintained by Survey Sampling
International, and participants receive points worth about $1.50 for participat-
ing. Participants can redeem accumulated points for cash or various goods like
frequent flyer miles, electronics, or sports equipment.
We make use of the data collected from the year-long time period from
June 2013 through June 2014. We omitted the May 2013 data because the
order of appearance of the products used in the choice experiment were not
randomized in that initial survey as they were in all the remaining issu-
ances, and we stopped in June 2014 so that we could begin to estimate the
models reported here. The final data set used for this analysis consists of
110,290 choices made by 12,255 people. The appendix shows the demo-
graphic make-up of the sample and compares it to the US population. The
levels of income, age, gender, ethnicity, region of residence, household size,
and food stamp participation in our FooDS sample is very similar to that for
the general US population; the largest divergence is for education as our
sample has more college graduates (45%) than the US population (28%).
The data that serve as the focal point of this analysis come from a choice
experiment delivered each month. As noted earlier, choice experiments are
widely used in the meat demand literature, but their applications in the eco-
nomics literature often focus on estimating willingness-to-pay for specific
quality attributes rather than estimating demand elasticities or substitution
patterns per se. Choice experiments have been found to yield results that
are predictive of and broadly consistent with actual retail purchasing pat-
terns (Louviere, Hensher, and Swait 2000; Chang, Lusk, and Norwood 2009;
Brooks and Lusk 2010), though as with all surveys there are concerns with
the extent to which people will actually behave as they say they will—a
problem in choice experiments that seems to be most related to the decision
of whether to buy or not rather than which product to buy conditional on a
choice (Lusk and Schroeder 2004).
In our choice experiment, each subject answered nine choice questions
like the one illustrated in figure 2. Preceding the set of questions was the
verbiage: “Imagine you are at the grocery store buying the ingredients to
prepare a meal for you or your household. For each of the nine questions
that follow, please indicate which meal you would be most likely to buy.”
Each of the choice questions was identical except the prices varied across
each question. Each question had nine options (two beef, two pork, two
chicken, two nonmeat, and one “no purchase” alternative) and the price of
each option was varied at three levels. For each meat type, we chose a
higher value product (steak, pork chop, chicken breast) and a lower value
product (ground beef, deli ham, and chicken wing) to investigate substitu-
tion not only across meat species but also within species. We purposely

5
Applied Economic Perspectives and Policy

included two nonmeat products in the choice experiment to allow for shifts
out of meat to not be necessarily forced into the no-purchase option as this
more accurately reflects grocery store experiences.
Rather than presenting respondents with choices involving all possible
price combinations that could exist, we selected a subset of choices
(or a fractional factorial design) such that the prices of each choice alterna-
tive were uncorrelated with each other alternative (i.e., there is no multicolli-
nearity). This so-called main effects orthogonal fractional factorial design
(see Louviere, Hensher, and Swait 2000; Lusk and Norwood 2005) required
27 choices. The 27 choices were blocked into three sets of nine, and each
person was randomly assigned to one of the three blocks. The design is able

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


to identify nonlinearity in the price effect (because there are three levels),
and it allows the price effects to vary for each alternative. When implement-
ing the survey, the order in which the alternatives were presented was ran-
domly varied across respondents (e.g., some people saw ground beef as the
first in the set, others saw chicken breast first in the set, etc.). The midpoint
of the price levels for each product were chosen to be roughly similar to
market prices when the project was initiated, and the higher and lower
levels were $1.50 above and below the midpoint.2 The appendix shows the
precise price levels used for each alternative, compares them to actual retail
prices over this time period, and shows the full experimental design.

Econometric Methods
We analyze the choice data through the lenses of a random utility model
(McFadden 1973). In particular, consumer i derives the following utility
from choice option j: Uij = Vij + 1ij . If the 1ij follow a Type I extreme value
distribution and are independently and identically distributed across i and
j, then the conventional multinomial logit model (MNL) results:

eVij
Prob(i chooses j) = 9 . (1)
k=1 eVik

The MNL imposes restrictive substitutions pattern across alternatives that


are particularly undesirable in this application (all cross-price elasticities for
an alternative are identical due to the independence of irrelevant alterna-
tives [IIA] property of the MNL). Moreover, many applications of random
utility models utilize functional forms for Vij (the deterministic component
of utility function) that assume that the marginal utility of a price change is
constant (i.e., the utility function is linear in price) and is identical for all
alternatives. Fortunately, one can rely on the simple structure in equation
(1) and relax the restrictive assumptions by modifying Vij .
There are a variety of ways to relax the assumptions of the MNL, and we
consider several approaches. It is useful to begin by considering the most
basic utility specification:

Vij = bj + a pj , (2)

2
We chose to vary the difference in prices for each alternative by an equal amount ($1.50) which could
have produced a different result than varying prices by equal percentage amounts. We varied the prices
using the same levels (not same percentages) because in the discrete choices models we consider, choices
are driven by the differences in levels of attributes (not proportional differences). See Hanley et al. (2005)
for evidence that choice of price levels has no influence on resulting implications.

6
Meat Demand Elasticities Vary with Price, Income, and Product Category

where pj is the price of alternative j, a is the marginal utility of a price


change, and bj is an alternative specific constant indicating the utility of
option j relative to the utility of the “no purchase” option which is normalized
to zero for identification purposes. Some simple modifications to the utility
function can allow more flexible price response patterns. For example, price
responsive can be allowed to vary across alternatives by adding a j subscript
to a, and by adding quadratic terms:

Vij = bj + aj pj + dj p2j . (3)

While adding a quadratic term (versus say a logarithmic term) might

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


produce undesirable predictions (i.e., at some point the estimates will suggest
people prefer paying higher prices), over the price ranges we consider here,
we do not observe such anomalies, and as such we stick with this approxima-
tion. Although equation (3) is less restrictive than equation (2), the IIA prop-
erty is still imposed.
We consider three different approaches to relax the IIA property. The first is
the universal or mother logit (McFadden, Tye, and Train 1977; Timmermans,
Borgers, and van de Waerden 1991) with quadratic price effects. The mother
logit not only includes the price of alternative j in alternative j’s utility function
but the price of all other alternatives in the choice set. The mother logit has the
advantage of being easy to estimate while allowing highly flexible substitution
patterns. It is specified as:


8
Vij = bj + a jt pt + dj p2j . (4)
t=1

Despite its flexibility and ease of estimation, the mother logit may yield esti-
mates that are inconsistent with utility maximization, and it results in a prolif-
eration of parameters (in our case 56 more parameters than equation [3]).
As a result, we also consider an error-component model (ECM) (DeJong
et al. 2003; Scarpa, Ferrini, and Willis 2005), which has the advantage of
relaxing IIA in a more parsimonious manner while explicitly modeling the
panel nature of the data (i.e., each person answered nine choice questions,
and it is assumed that their preferences are correlated over the choices). In
general, the ECM can be written as:

Vij = bj + aj pj + dj p2j + eij , (5)

where eij is a mean-zero Normally distributed error component for person i


and alternative j. We parsimoniously summarize the error component in equa-
tion (5) with a single term, eij , but in practice we consider specifications in
which this term is made up of several additive error components that enter
the utility function for multiple alternatives. In particular, we consider error
components for: (1) all food products vs. the “none” alternative, (2) all meat
products vs. non meat products, (3) the two beef product alternatives, (4) the
two pork product alternatives, (5) the two chicken product alternatives, (6)
higher value meat products (steak, pork chop, chicken breast), and (7) lower
value meat products (ground beef, deli ham, chicken wing). With the addition
of just seven additional parameters (relative to equation [3]), the ECM allows a
much richer possibility of substitution possibilities. Still, the ECM requires the

7
Applied Economic Perspectives and Policy

analyst to make judgment calls about how the error components enter the
utility functions, which may not properly reflect consumers’ preferences.
The final specification we consider is the mixed logit model, also known
as the random parameter logit (RPL). The most general form of the RPL we
consider is:

 

8
Vij = bj + v jt dit + aj pj + dj p2j , (6)
t=1

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


where dit  N(0, 1) and v jt are the elements of the lower-triangle of the
Cholesky decomposition associated with the covariance matrix of the
random parameters. Equation (6) posits a specification in which the eight al-
ternative specific constants are distributed multivariate normal and in which
correlations between each alternative specific constant are estimated. If the
off-diagonals of the Cholesky matrix are assumed zero, then each alternative
specific constant, bj , is independently distributed. The advantage of the RPL
is that, according to McFadden and Train (2000), it can approximate any
random utility model to any degree of accuracy. However, the fully correlated
version of the model requires 60 parameters and, given our large sample size,
is quite slow to estimate (requiring several days to reach convergence).
Given the robust set of models considered, we took the following approach
for model selection. First, using the entire data set, we estimated the MNL,
mother logit, ECM, and RPL models for utility specifications that included
and excluded quadratic price effects, and except for the mother logit, models
that assumed alternative-specific and alternative-independent price effects.
The MNL and mother logit were estimated using conventional maximum
likelihood estimation, whereas the ECM and RPL were estimated using
simulated maximum likelihood techniques using 200 individual-specific
quasi-random Halton draws (see Train [2009] for discussion on discrete
choice estimation using simulation). Likelihood ratio tests are used to test the
appropriateness of each utility specification within each model approach (i.e.,
whether aj = a ∀ j and whether dj = 0). In addition, a likelihood ratio test is
used to determine whether the RPL model should include independent or
correlated alternative-specific constants (i.e., whether v jt = 0 ∀ j = t). The
Akaike Information Criteria (AIC) is used to select across the preferred MNL,
mother logit, ECM, and RPL models. Finally, for the preferred specification,
we split the sample into three income categories: low income (,$40,000 in
annual household income), middle income (between $40,000 and $99,999 in
annual household income), and high income (≥$100,000 in annual house-
hold income), and re-estimate the preferred specification for each income
category.3 Likelihood ratio tests are used to test the hypothesis that the model
parameters are identical for each income category.
3
In addition to income, it is possible to include other demographics like age, gender, and household size
which might affect preferences. We do not do this here for several reasons. First, the RPL models allow for
taste heterogeneity across households. So, heterogeneity is not ignored (it is a key aspect of the RPL
model), rather it is not linked to observable demographics. Second, the addition of demographic variables
quickly leads to parameter proliferation. For example, assume five additional demographic variables were
considered. There are eight alternative specific constants ( plus one normalized to zero), eight linear price
effects, and eight quadratic price effects. This would mean an additional 5*(8+8+8)¼120 coefficients
would be estimated. While our sample size is sufficient to undertake this activity, succinctly reporting
and summarize these findings would be a challenge.

8
Meat Demand Elasticities Vary with Price, Income, and Product Category

Results
For the MNL, mother logit, ECM, and RPL models, likelihood ratio tests
strongly reject the assumption of constant marginal utility in favor of includ-
ing quadratic effects. Likelihood ratio tests also reject the assumption of
alternative-independent price effects in favor of alternative-specific price
effects. Thus, no matter the model specification (MNL, mother logit, ECM, or
RPL), the data support nonlinear price effects that vary with the type of
product. Across specifications, the AIC suggests that the preferred specifica-
tion is the RPL model with quadratic price effects and correlated alternative-
specific constants. Moreover, within the RPL specification, likelihood ratio

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


tests reject the assumptions of uncorrelated alternative specific constants and
zero quadratic effects. The appendix shows the log-likelihood function and
AIC values for each model as well as details on the likelihood ratio tests.
Table 1 reports the results for the preferred specification for the aggregate
pooled model and for the three income classes. All parameters are of
expected sign, and the results indicate significant differences in price re-
sponsiveness across different meat products, statistically significant quad-
ratic price effects, and large covariances between random parameters (the
appendix reports the correlation coefficients between random parameters
implied by the variance-covariance estimates in table 1). To provide some
insight into interpretation of estimates, consider the pooled model. The first
coefficient in the model, 6.295, implies that the “average” consumer receives
6.295 more utils from ground beef than from buying nothing (the “none”
option). Likewise, the “average” consumer receives 7.534 more utils from
steak than “none.” Thus, steak provides 7.53226.295 ¼ 1.239 more utils
than ground beef (ignoring price differences). The price coefficients provide
the marginal utility of a price change. A $1 increase in the price of ground
beef, for example, causes a 21.437 + 0.094*pground beef change in the utility
derived from the ground beef option. Although the mean preference for
ground beef over “none” is 6.295, the estimated variance is 11.122, meaning
there is significant heterogeneity across consumers. We would expect
68% of consumers
√ to have a preference for ground beef that is within
+ 11.122 = 3.335 of the mean, 6.295.
Comparing the sum of the log likelihood function values from the three
income category models to the pooled model in table 1 yields a chi-squared
value of 1256.95 with 120 degrees of freedom, which yields a p-value of less
than .001. Thus, the hypothesis that the model parameters are the same for
all income categories is firmly rejected. Preferences differ by income.
Comparing across the linear price effect estimates, high-income consumers
are less responsive to own-price changes for every type of food than are low-
income consumers, with middle-income consumers falling between the two.
Recognizing that comparisons of individual parameters across models could
be confounded with differences in error variance, table 2 reports the esti-
mated market shares for each income class of consumers at the median prices
for each product used in the choice experiment (shares are calculated by
plugging the estimated coefficients associated with equation [6] into equation
[1]; see the appendix for more detail). Higher income consumers are almost
twice as likely to choose steak (the most expensive product in the set) than
are lower income consumers. In addition, the estimated market share
of chicken breast is 6 percentage points higher for high- versus low-income
consumers. By contrast, low-income consumers are more likely to choose

9
Applied Economic Perspectives and Policy
Table 1 Estimates of Random Parameter Models by Income

Variables Pooled Low income <$40,000 Middle income $40,000 to $99,999 High income >$100,000

Means of Alternative-Specific constants


Ground beef (GB) 6.295*a (0.120)b 6.379* (0.190) 6.353* (0.179) 5.730* (0.291)
Steak (SK) 7.534* (0.496) 7.446* (0.883) 8.107* (0.737) 6.874* (1.045)
Pork chop (PC) 5.537* (0.158) 5.239* (0.260) 5.654* (0.235) 5.626* (0.354)
Deli ham (DH) 3.943* (0.091) 4.051* (0.141) 3.781* (0.136) 3.752* (0.231)
Chicken breast (CB) 7.612* (0.097) 7.555* (0.155) 7.627* (0.142) 7.534* (0.227)
Chicken wing (CW) 3.478* (0.081) 3.444* (0.123) 3.419* (0.120) 3.440* (0.198)
Rice and beans (RB) 2.847* (0.069) 2.903* (0.106) 2.695* (0.101) 2.957* (0.171)
Pasta (PA) 5.898* (0.209) 6.012* (0.340) 5.856* (0.315) 5.674* (0.481)
Linear Price Effects
Ground beef (GB) 21.437* (0.070) 21.531* (0.113) 21.512* (0.105) 21.040* (0.169)
10

Steak (SK) 21.394* (0.158) 21.534* (0.283) 21.555* (0.236) 20.958* (0.332)
Pork chop (PC) 21.271* (0.090) 21.203* (0.150) 21.324* (0.135) 21.165* (0.197)
Deli ham (DH) 21.115* (0.068) 21.256* (0.109) 21.062* (0.104) 21.049* (0.169)
Chicken breast (CB) 22.021* (0.054) 22.172* (0.090) 22.039* (0.080) 21.683* (0.122)
Chicken wing (CW) 21.131* (0.068) 21.073* (0.105) 21.256* (0.106) 20.994* (0.166)
Rice and beans (RB) 20.907* (0.048) 20.960* (0.076) 20.865* (0.073) 20.916* (0.118)
Pasta (PA) 21.894* (0.113) 22.008* (0.185) 21.894* (0.171) 21.640* (0.258)
Quadratic Price Effects
Ground beef (GB) 0.094* (0.001) 0.098* (0.017) 0.107* (0.016) 0.050* (0.025)
Steak (SK) 0.069* (0.012) 0.084* (0.022) 0.079* (0.018) 0.034 (0.026)
Pork chop (PC) 0.083* (0.012) 0.071* (0.021) 0.088* (0.019) 0.079* (0.027)
Deli ham (DH) 0.098* (0.013) 0.112* (0.022) 0.086* (0.021) 0.108* (0.033)
Chicken breast (CB) 0.177* (0.008) 0.196* (0.014) 0.180* (0.013) 0.133* (0.019)
Chicken wing (CW) 0.086* (0.017) 0.054* (0.027) 0.121* (0.027) 0.075 (0.042)
Rice and beans (RB) 0.089* (0.012) 0.092* (0.020) 0.082* (0.019) 0.105* (0.030)
Pasta (PA) 0.177* (0.014) 0.192* (0.024) 0.178* (0.022) 0.144* (0.033)

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


Variances of Alternative-Specific Constants
Ground beef (GB) 11.122* (0.351) 11.183* (0.562) 9.892* (0.484) 10.769* (0.780)
Steak (SK) 15.117* (0.491) 14.277* (0.719) 15.257* (0.671) 18.081* (1.105)
Pork chop (PC) 11.614* (0.378) 12.563* (0.622) 9.909* (0.500) 10.693* (0.793)
Deli ham (DH) 10.556* (0.355) 9.810* (0.528) 9.235* (0.482) 13.173* (0.876)
Chicken breast (CB) 11.338* (0.351) 11.912* (0.612) 10.021* (0.506) 12.453* (0.831)
Chicken wing (CW) 12.784* (0.424) 12.702* (0.585) 11.015* (0.535) 12.919* (0.907)
Rice and beans (RB) 9.108* (0.603) 9.732* (0.492) 8.086* (0.408) 10.490* (0.787)
Pasta (PA) 8.220* (0.398) 9.174* (0.526) 7.310* (0.417) 9.058* (0.656)
Covariances of Alternative-Specific Constants
SK, GB 11.619* (0.383) 11.430* (0.597) 11.118* (0.528) 12.493* (0.828)
PC, GB 10.790* (0.351) 11.548* (0.569) 9.327* (0.467) 9.895* (0.700)
PC, SK 10.359* (0.395) 12.798* (0.625) 11.201* (0.529) 12.567* (0.855)
DH, GB 10.292* (0.337) 9.743* (0.520) 8.981* (0.453) 11.107* (0.731)
DH, SK 11.071* (0.382) 10.116* (0.580) 9.813* (0.521) 12.243* (0.864)

Meat Demand Elasticities Vary with Price, Income, and Product Category
11

DH, PC 10.498* (0.343) 10.613* (0.544) 8.988* (0.456) 11.029* (0.759)


CB, GB 10.130* (0.337) 10.528* (0.540) 8.968* (0.437) 10.100* (0.672)
CB, SK 10.677* (0.376) 11.007* (0.590) 9.602* (0.504) 11.620* (0.811)
CB, PC 10.435* (0.347) 11.325* (0.567) 9.002* (0.441) 10.035* (0.706)
CB, DH 9.334* (0.329) 9.242* (0.518) 7.984* (0.438) 10.619* (0.726)
CW, GB 10.048* (0.364) 10.093* (0.547) 8.641* (0.456) 9.722* (0.676)
CW, SK 9.830* (0.405) 10.848* (0.598) 10.213* (0.524) 11.431* (0.824)
CW, PC 10.275* (0.375) 10.869* (0.569) 8.670* (0.458) 9.414* (0.707)
CW, DH 10.479* (0.361) 9.725* (0.527) 8.944* (0.457) 11.399* (0.757)
CW, CB 9.816* (0.355) 10.162* (0.546) 8.318* (0.441) 9.552* (0.694)
RB, GB 7.373* (0.305) 7.783* (0.484) 5.898* (0.394) 7.415* (0.574)
RB, SK 8.047* (0.346) 8.484* (0.541) 6.239* (0.466) 8.219* (0.719)
RB, PC 7.445* (0.304) 8.645* (0.506) 6.055* (0.390) 6.661* (0.589)
RB, DH 7.558* (0.301) 7.662* (0.482) 5.896* (0.389) 9.385* (0.652)
Continued

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


Applied Economic Perspectives and Policy
Table 1 Continued

Variables Pooled Low income <$40,000 Middle income $40,000 to $99,999 High income >$100,000

RB, CB 7.197* (0.300) 8.002* (0.494) 6.041* (0.387) 7.845* (0.587)


RB, CW 8.369* (0.337) 8.667* (0.497) 6.739* (0.405) 9.950* (0.707)
PA, GB 7.887* (0.302) 8.608* (0.499) 6.795* (0.412) 7.172* (0.578)
PA, SK 7.962* (0.341) 9.797* (0.567) 7.517* (0.486) 9.067* (0.703)
PA, PC 7.943* (0.303) 9.392* (0.523) 6.651* (0.413) 6.708* (0.588)
PA, DH 7.345* (0.301) 7.672* (0.494) 6.094* (0.400) 8.280* (0.634)
PA, CB 7.523* (0.298) 8.557* (0.519) 6.634* (0.402) 7.798* (0.591)
PA, CW 7.588* (0.331) 8.347* (0.505) 6.347* (0.413) 9.921* (0.654)
PA, RB 5.305* (0.366) 8.691* (0.493) 6.923* (0.385) 9.105* (0.620)
N choices (people) 110295 (12255) 41913 (4657) 47952 (5328) 20430 (2270)
Log Likelihood 2180574.4 268654.4 278669.7 232621.8
12

a
One asterisk represents statistical significance at the 0.01 level or lower.
b
Numbers in parentheses are standard errors.

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


Meat Demand Elasticities Vary with Price, Income, and Product Category

Table 2 Estimated Market Shares by Income

Product Aggregatea Low income Middle income High income

Ground beef (GB) 12.97% 13.83% 12.91% 11.33%


Steak (SK) 10.47% 7.64% 11.20% 14.55%
Pork chop (PC) 8.34% 7.89% 8.46% 8.99%
Deli ham (DH) 7.11% 7.68% 7.07% 6.03%
Chicken breast (CB) 24.57% 22.17% 25.11% 28.22%
Chicken wing (CW) 11.19% 13.35% 10.22% 9.04%
Rice and beans (RB) 10.35% 10.82% 10.61% 8.75%
Pasta (PA) 5.85% 5.83% 6.02% 5.49%

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


No purchase 9.16% 10.79% 8.39% 7.61%

Note: Market shares estimated using mid-points of prices employed in the experimental design; shares for
each income category are calculated by averaging over 25,000 draws from the respective estimated
parameter distributions.
a
Aggregate market share is the weighted average share across income category (the sample has 38% low
income, 43.5% middle income, and 18.5% high income).

lower-priced products such as ground beef, chicken wings, and rice and
beans than are richer consumers, suggesting these products are inferior
goods. Although there is nothing in the model that would require such an
outcome, the market share of middle-income consumers falls between that of
low and high income for every alternative (except pasta). This finding,
coupled with the results of the likelihood ratio test, suggests income is an
important and predictable driver of meat preferences.
Also of interest in table 2 is the result that low-income consumers are
more likely to choose the “no purchase” alternative than are higher income
consumers. This finding suggests some caution in utilizing conditional
demand systems (which simply re-allocate a fixed level of expenditure
among products in the assumed separability set when prices rise) to
compare consumption patterns of richer and poorer consumers. This result
is also interesting in light of rising prices and amidst the recent recession.
Specifically, the lower-income segment is likely representing a declining
share of total consumption for meat products. Given our finding of different
elasticities by income segment, this is important for understanding the
current, aggregate meat demand landscape.
Figure 3 utilizes the estimates in table 1 to construct demand curves for
four meat products for each income category over the range of prices used
in the experimental design. The first thing to note about the demand curve
for each product and income class is that they are convex (as implied by the
positive quadratic effects in table 1). This implies that demand becomes
more inelastic as prices rise. Price changes in the upper-end of the price dis-
tribution do not change the quantity demanded as much as price changes in
the lower-end of the price distribution. This phenomenon helps explain the
surprise expressed by some analysts at the lack of reduction in beef and
pork expenditures despite higher prices.
For some products, like pork chops, there are not marked differences in
the demand curves for low-, middle-, and high-income consumers (except
at high price levels). For other products like steak, income changes result in
large shifts in demand. In all cases, but perhaps most easily seen for ground
beef, the demand curves for high-income consumers are more inelastic than
that for low-income consumers.

13
Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016
14

Figure 3 Implied demand curves for four meat products, by income

Applied Economic Perspectives and Policy


Meat Demand Elasticities Vary with Price, Income, and Product Category

Figure 4 Change in demand for steak and chicken breast as prices of other meats change,
middle-income consumers

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


Whereas figure 3 illustrates relative responsiveness to own-price changes,
figure 4 uses the estimated model for middle-income consumers (representing
the largest share of consumers at 43.5%) to illustrate cross-price demand
responses for steak and chicken breast. The top panel reveals that a drop in
price for chicken breast has a much larger impact on steak demand than an
equivalent increase in ground beef price. This demand inter-relationship is
consistent with the observed reduction in the share of meat purchases that
have come from beef over the past thirty to forty years as the price of chicken
has substantively fallen in relative terms. However, as the bottom panel of
table 5 shows, demand for chicken breast is virtually unaffected by large price
increases for steak or ground beef. This result is consistent with observations
in the current market environment suggesting that consumers have not substi-
tuted in large numbers toward chicken despite much higher beef prices.
It is useful to illustrate the potential bias inherent in more restrictive
models that assume linear price effects or uncorrelated alternative-specific
constants. Figure 5 shows the implied demand curves for ground beef asso-
ciated with the preferred specification for middle-income consumers (the
third column of results in table 1) and compares it to RPL models that
assume linear price effects or uncorrelated random parameters (the appen-
dix shows the estimates for these rejected models). When all prices are at

15
Applied Economic Perspectives and Policy

Figure 5 Ground beef demand implied by alternative model specifications, middle-income


consumers

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


Note: the RPL, correlated, quadratic model is the preferred specification.

the median level used in the experimental design (which is $3.50 for ground
beef ), there is not much difference in the implied market shares across the
competing models. However, as prices increase or decrease, the bias grows.
The models assuming linear price effects under-state the quantity
demanded (or market share) at both low and high prices. It is important to
note that even the specifications that assume linear price effects still imply
demand curves with some degree of curvature because of the nonlinear
form of the logit formula in equation (1). Not only does the assumption of
constant marginal price effects yield biased estimates of the slope of the
demand curve, as figure 5 also shows, the assumption also has implications
for understanding cross-price demand responses. In particular, when the
price of chicken breast falls, the RPL model with uncorrelated random para-
meters suggests a more dampened demand shift than actually results.
Allowing random parameter correlation yields a more accurate depiction of
the cross-price demand response, but the lack of curvature in the price
effects yields biased estimates—this time at the midpoint of the price for
ground beef rather than at the extremes.
The own- and cross-price demand responses can be more systematically
investigated by investigating elasticities. Because of the nonlinear nature of
the demand curves, table 3 reports elasticities implied by a price increase,
and table 4 reports elasticities implied by a price decrease. The reported
elasticities are arc elasticities calculated by determining how a constant per-
centage price increase for each product changes the estimated market share
(determined via equation [1]), starting from the mid-points of the prices for
each product utilized in the experimental design (the appendix provides
more detail on how elasticities are calculated and shows qualitatively
similar results associated with a constant dollar increase for each product).
Before discussing the results, we note that it is a bit of a challenge to dir-
ectly compare our elasticities with those present in the extant literature
because prior analyses typically provide elasticities aggregated over all cuts
within a species and over all markets. We have product-level demands
observed from a choice question format that places people in a single, store
market environment (where demand is likely to be more elastic). That said,

16
Table 3 Arc Elasticities for a Price Increase by Income

Change in Price

Change in Quantity GB SK PC DH CB CW RB PA

Low income
Ground beef (GB) 21.959 0.184 0.224 0.141 0.544 0.152 0.071 0.075
Steak (SK) 0.375 21.738 0.269 0.097 0.425 0.141 0.065 0.091
Pork chop (PC) 0.470 0.277 21.927 0.157 0.566 0.151 0.078 0.076
Deli ham (DH) 0.421 0.141 0.225 21.306 0.422 0.194 0.103 0.064
Chicken breast (CB) 0.339 0.131 0.169 0.089 21.503 0.145 0.075 0.064
Chicken wing (CW) 0.276 0.126 0.132 0.119 0.419 20.939 0.099 0.054
Rice and beans (RB) 0.211 0.096 0.111 0.103 0.355 0.162 20.744 0.152
Pasta (PA) 0.310 0.188 0.151 0.089 0.425 0.123 0.212 21.252

Meat Demand Elasticities Vary with Price, Income, and Product Category
17

No Purchase 0.186 0.069 0.065 0.083 0.272 0.091 0.087 0.065


Middle Income
Ground beef (GB) 21.834 0.373 0.197 0.135 0.603 0.107 0.060 0.078
Steak (SK) 0.344 21.836 0.221 0.075 0.365 0.108 0.039 0.056
Pork chop (PC) 0.325 0.398 21.844 0.136 0.623 0.109 0.068 0.070
Deli ham (DH) 0.398 0.239 0.243 21.224 0.449 0.172 0.077 0.064
Chicken breast (CB) 0.286 0.189 0.179 0.073 21.394 0.101 0.067 0.071
Chicken wing (CW) 0.231 0.253 0.142 0.127 0.457 20.949 0.094 0.057
Rice and beans (RB) 0.168 0.117 0.116 0.074 0.394 0.121 20.658 0.171
Pasta (PA) 0.258 0.199 0.141 0.072 0.492 0.087 0.202 21.253
No purchase 0.166 0.109 0.101 0.068 0.341 0.073 0.093 0.083
Continued

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


Applied Economic Perspectives and Policy
Table 3 Continued

Change in Price

Change in Quantity GB SK PC DH CB CW RB PA

High Income
Ground beef (GB) 21.703 0.459 0.185 0.094 0.580 0.090 0.051 0.053
Steak (SK) 0.253 21.674 0.187 0.041 0.378 0.066 0.028 0.052
18

Pork chop (PC) 0.271 0.496 21.524 0.097 0.599 0.086 0.036 0.045
Deli ham (DH) 0.342 0.265 0.242 20.940 0.479 0.128 0.086 0.050
Chicken breast (CB) 0.207 0.244 0.147 0.048 21.151 0.063 0.047 0.058
Chicken wing (CW) 0.203 0.268 0.133 0.080 0.392 20.869 0.101 0.144
Rice and beans (RB) 0.157 0.152 0.076 0.072 0.401 0.137 20.632 0.191
Pasta (PA) 0.148 0.262 0.087 0.039 0.454 0.180 0.176 21.327
No purchase 0.163 0.132 0.122 0.028 0.339 0.050 0.063 0.082

Note: Arc elasticities are calculated by determining the proportionate change in market shares resulting from a 10% increase in the price of each respective food divided by 0.1; shares for each
income category are calculated by averaging over 25,000 draws from the respective estimated parameter distributions.

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


Table 4 Arc Elasticities for a Price Decrease by Income

Change in Price

Change in Quantity GB SK PC DH CB CW RB PA

Low income
Ground beef (GB) 22.511 0.280 0.292 0.173 0.681 0.167 0.081 0.119
Steak (SK) 0.478 22.652 0.347 0.120 0.546 0.155 0.075 0.145
Pork chop (PC) 0.587 0.408 22.526 0.192 0.705 0.166 0.088 0.121
Deli ham (DH) 0.530 0.217 0.292 21.614 0.542 0.211 0.116 0.102
Chicken breast (CB) 0.435 0.202 0.223 0.111 21.916 0.159 0.085 0.103
Chicken wing (CW) 0.358 0.195 0.175 0.146 0.537 21.031 0.112 0.087
Rice and beans (RB) 0.278 0.149 0.148 0.128 0.460 0.177 20.839 0.238
Pasta (PA) 0.398 0.282 0.199 0.111 0.544 0.136 0.234 21.995

Meat Demand Elasticities Vary with Price, Income, and Product Category
19

No purchase 0.244 0.108 0.088 0.103 0.357 0.101 0.098 0.104


Middle income
Ground beef (GB) 22.377 0.514 0.260 0.161 0.735 0.122 0.068 0.121
Steak (SK) 0.441 22.606 0.290 0.092 0.467 0.123 0.045 0.087
Pork chop (PC) 0.420 0.545 22.437 0.163 0.758 0.124 0.077 0.109
Deli ham (DH) 0.504 0.343 0.316 21.480 0.566 0.193 0.087 0.099
Chicken breast (CB) 0.370 0.274 0.236 0.089 21.737 0.115 0.076 0.110
Chicken wing (CW) 0.304 0.358 0.190 0.152 0.572 21.079 0.104 0.089
Rice and beans (RB) 0.224 0.175 0.155 0.090 0.499 0.137 20.735 0.259
Pasta (PA) 0.336 0.286 0.188 0.087 0.611 0.099 0.221 21.935
No purchase 0.220 0.163 0.136 0.083 0.435 0.084 0.104 0.129
Continued

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


Applied Economic Perspectives and Policy
Table 4 Continued

Change in Price

Change in Quantity GB SK PC DH CB CW RB PA

High income
Ground beef (GB) 22.075 0.548 0.236 0.114 0.673 0.100 0.059 0.078
Steak (SK) 0.307 22.061 0.238 0.051 0.455 0.074 0.033 0.076
20

Pork chop (PC) 0.328 0.588 21.947 0.118 0.693 0.095 0.042 0.067
Deli ham (DH) 0.406 0.332 0.303 21.157 0.568 0.141 0.098 0.074
Chicken breast (CB) 0.254 0.307 0.188 0.059 21.360 0.070 0.055 0.085
Chicken wing (CW) 0.249 0.333 0.171 0.098 0.471 20.967 0.115 0.207
Rice and beans (RB) 0.194 0.196 0.100 0.089 0.479 0.151 20.724 0.272
Pasta (PA) 0.184 0.325 0.114 0.049 0.535 0.196 0.198 21.931
No purchase 0.200 0.171 0.157 0.035 0.409 0.057 0.072 0.119

Note: Arc elasticities are calculated by determining the proportionate change in market shares resulting from a 10% decrease in the price of each respective food divided by 20.1; shares for each
income category are calculated by averaging over 25,000 draws from the respective estimated parameter distributions.

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


Meat Demand Elasticities Vary with Price, Income, and Product Category

the estimates in table 3 are not outside the realm of that reported in prior
literature. For example, the own-price elasticities of demand for steak we
estimate are between 21.67 and 21.84, depending on income class, which
can be compared with the estimate of 21.88 in Taylor and Tonsor (2013)
obtained using scanner data.
We find ground beef demand is more elastic than steak, which is consistent
with the findings of Brester and Wohlgenant (1991) and Coffey, Schroeder,
and Marsh (2011) but opposite of that found by Capps (1989), Nayga and
Capps (1994), and Taylor and Tonsor (2013). Table 3 indicates pork chop
demand to be more elastic than ham. This is consistent with the findings of
Capps (1989) and Coffey, Schroeder, and Marsh (2011) but differs from the

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


findings of Nayga and Capps (1994).4 Our estimates suggesting chicken breast
demand is less elastic than ground beef, steak, and pork chop demand is con-
sistent with Capps (1989).
Comparing table 3 to table 4, one can see that own-price arc elasticities are
more inelastic for a price increase (for higher prices) than they are for a price
decrease (for lower prices). For example, the own-price elasticity of demand
for ground beef for low-income consumers is 21.96 for a price increase and
22.51 for a price decrease. Moreover, the magnitude of the cross-price elasti-
cities is generally higher for price decreases than price increases. Stated differ-
ently, there is less substitutability at higher prices than lower prices. This
result, while somewhat counterintuitive at first blush, seems to be consistent
with outcomes in the current market environment. This finding may be
related to shifts in the relative purchasing of higher income versus lower
income households in the presence of higher prices. That is, if higher income
households are less price sensitive, they would comprise a larger share of
total purchases in a higher meat price environment.
Indeed, within either table 3 or 4, the results show that higher income con-
sumers tend to have more inelastic own-price elasticities and smaller cross-
price elasticities than lower income consumers, again with middle-income
consumers falling somewhere in between. As figure 4 revealed, the cross-price
elasticities for ground beef and steak with respect to chicken breast are larger
than the cross-price elasticities for chicken breast with respect to ground beef
and steak. The results in table 3 show that for low-income consumers, a 1% in-
crease in the price in chicken breast results in a 0.54% increase in the market
share for ground beef; however, a 1% increase in the price of ground beef only
results in a 0.34% increase in the market share for chicken breast.

Discussion and Conclusion


Banks, Blundell, and Lewbel (1997, 527) argue that “for many commod-
ities, standard empirical demand models do not provide an accurate picture
of observed behavior.” This study sought to help resolve a similar discon-
nect between the current meat demand environment and analyst expecta-
tions. Utilizing a novel data set generated by an ongoing series of nationally
distributed choice experiments, we reveal nonlinear demands for meat pro-
ducts, with demand being more inelastic at higher prices. We also identify a
series of insights into how cross-price relationships vary across meat prices,
meat products, and consumers with different incomes.
4
Note, however, the insignificant own-price elasticity estimated by Capps (1989) of 0.36 is inconsistent
with curvature expectations. Moreover, the Coffey, Schroeder, and Marsh (2011) study actually provides
a ‘Roast/Ham’ estimate of 20.17, which is also not statistically different from zero.

21
Applied Economic Perspectives and Policy

A number of implications result from this analysis. The often-imposed as-


sumption of linearity in empirical work and policy analyses using constant
elasticities warrants further scrutiny. It is common to use econometrically
estimated elasticities of demand in general or partial equilibrium models
investigating various policy or market shocks (e.g., Balagtas and Kim 2007;
Pendell et al. 2010; Weaber and Lusk 2010; Okrent and Alston 2012).
Similarly, elasticities published in the literature (typically based on econo-
metric models utilizing time series data spanning multiple decades) are fre-
quently inputs for models providing forecasts of longer-term prices, net
farm incomes, government expenditures, and other statistics of key interest
to stakeholders throughout the entire US agricultural system (Rosegrant,

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


Tokgoz, and Bhandary 2013; FAPRI 2014; USDA ERS 2014). In many appli-
cations, analysts assume constant elasticities over all price ranges and, as a
result, may over- or underestimate economic impacts if elasticities are actu-
ally nonlinear. Our findings reinforce the assertion that equilibrium dis-
placement models should be limited to situations characterized by small
changes (Wohlgenant 2011). Research that seeks to project effects of larger
shocks might do well to utilize approaches with widely varying prices or
market environments, perhaps by applying split sample survey or experi-
ment designs (Tonsor 2011). Furthermore, our finding of demand patterns
differing by income not only adds to the large body of research on Engel
curves but suggests that such effects need further consideration in the dis-
crete choice demand literature. These differential consumption patterns
across income categories buttress Gustavsen (2015) conclusion that “con-
sumption patterns are usually different in different cohorts.” This highlights
the potential importance of conducting research separately for consumer seg-
ments distinguished by income group.
The empirical demand estimates might also be useful in helping forecast
future developments in the meat market. As the old adage goes, “the cure for
high prices is high prices.” High beef and pork prices are incentivizing pro-
ducers to hold back breeding stock. If expansion occurs as expected in the
coming years, then real retail prices will decline, assuming no countervailing
demand or supply shocks. The estimates presented here suggest that the cur-
rently observed “limited substitution” may not hold as beef and pork prices
fall. Moreover, because the hog cycle is shorter than the cattle cycle, we are
likely to witness falling pork prices before beef, which is likely to have impli-
cations for substitution patterns within and across species categories.
One of the advantages of the ongoing choice experiment is that we may ul-
timately be able to more cleanly determine whether and why structural
change has occurred than is feasible using more conventional approaches.
While the present study focused on the effects of relative price changes, future
research utilizing this sort of data can study how consumers’ choices vary over
time to the same set of unchanging questions. The advantage of the survey-
based approach used here is that more clearly delineates consumer demand
from supply-side effects, and the experimental design ensures there are no
confounds with unmeasured characteristics. As previously indicated, prior
research provides a mixed bag of demand elasticities for disaggregate meat
products, and it is difficult to ascertain whether differences arise because of
differences in assumed functional forms, data sets, or time periods. Another
future benefit worthy of additional attention is the potential ability to forecast
future demand patterns utilizing less aggregated product-level elasticities
derived from ongoing choice experiments. While there currently are limited

22
Meat Demand Elasticities Vary with Price, Income, and Product Category

bases of comparison, either in the form of product-level quantity data or


previously published elasticities, that constraint may ease over time making
out-of-sample forecasting assessments feasible.
Of course, survey-based methods like choice experiments are not a panacea.
One major concern is that people may answer survey questions differently
than they actually shop. The potential for this sort of hypothetical bias appears
less problematic for choice experiments than other methods, and are a variety
of studies that have found choice experiments yield preference consistent with
that from actual market transactions, but the concern exists nonetheless.
Moreover, choice experiments require a number of design choices be made
such as the experimental design, the way option are presented, and the choice

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


of attributes and levels, any of which may influence the ultimate results in a
way that limits external validity. For example, we chose to present pictures for
each option. Would a different set of pictures yield a different pattern of
results? There is no way to know except to conduct a different study with dif-
ferent pictures. There is a long literature on conjoint analysis which provides
guidance when making such design choices, but potentially consequential
design decisions must ultimately be made by the researcher.
While all empirical analyses have their limitations, the possible demand
insights offered by consistently conducting the same set of choice experi-
ments with a large set of consumers over time are illustrated by this study. It
wasn’t too long ago that scanner data analysis of meat demand was in its
infancy (e.g., Capps 1989). Given the ongoing challenge of analyzing fresh
meat demand with this sort of data, given the inability to identify unit
weights, new approaches will be needed. There have emerged efforts to use
stated preference surveys, sometimes merged with revealed preferences
(e.g., Brooks and Lusk 2010), to extend our understanding of demand pat-
terns. The next development—larger and longer running choice experi-
ments—may yield further insights and help researchers better disentangle
demand determinants.

Funding
The project was partially supported by the Oklahoma Agricultural Experiment
Station, the Willard Sparks Endowment at Oklahoma State, and the Agricultural
and Food Research Initiative Competitive Program of the USDA National Institute
of Food and Agriculture (NIFA), grant number 2015-67023-23134.

References
Alfnes, F., and K. Rickertsen. 2003. European Consumers’ Willingness to Pay for US
Beef in Experimental Auction Markets. American Journal of Agricultural Economics
85 (2): 396 –405.
Balagtas, J. V., and S. Kim. 2007. Measuring the Effects of Generic Dairy Advertising
in a Multimarket Equilibrium. American Journal of Agricultural Economics 89 (4):
932 –46.
Banks, J., R. Blundell, and A. Lewbel. 1997. Quadratic Engle Curves and Consumer
Demand. Review of Economics and Statistics 79: 527– 39.
Beach, R. H., and M. T. Holt. 2001. Incorporating Quadratic Scale Curves in Inverse
Demand Systems. American Journal of Agricultural Economics 83: 230–45.

23
Applied Economic Perspectives and Policy

Brester, G. W., and T. C. Schroeder. 1995. The Impacts of Brand and Generic
Advertising on Meat Demand. American Journal of Agricultural Economics 77: 969–79.
Brester, G. W., and M. K. Wohlgenant. 1991. Estimating Interrelated Demand for
Meats Using New Measures for Ground and Table Cut Beef. American Journal of
Agricultural Economics 73: 1182–94.
———. 1993. Correcting for Measurement Error in Food Demand Estimation. Review of
Economics and Statistics 75: 352 –56.
Brooks, K., and J. L. Lusk. 2010. Stated and Revealed Preferences for Organic and
Cloned Milk: Combining Choice Experiment and Scanner Data. American Journal
of Agricultural Economics 92 (4): 1229–41.
Brown, D. J., and L. F. Schrader. 1990. Cholesterol Information and Shell Egg
Consumption. American Journal of Agricultural Economics 72: 548–55.

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


Bryant, H. L., and G. C. Davis. 2008. Revisiting Aggregate U.S. Meat Demand with a
Bayesian Averaging of Classical Estimates Approach: Do We Need a More
General Theory? American Journal of Agricultural Economics 90: 103–16.
Burton, M., and T. Young. 1996. The Impact of BSE on the Demand for Beef and
other Meats in Great Britain. Applied Economics 28: 687– 93.
Capps, O. 1989. Utilizing Scanner Data to Estimate Retail Demand Functions for
Meat Products. American Journal of Agricultural Economics 71: 750– 60.
Carlsson, F. 2011. “Non-Market Valuation: Stated Preference Methods.” In Oxford
Handbook of the Economics of Food Consumption and Policy, ed. J. L. Lusk, J. Roosen,
and J. Shogren, 181– 214. Oxford: Oxford University Press.
Carlsson, F., P. Frykblom, and C.J. Lagerkvist. 2007. Consumer Willingness to Pay
for Farm Animal Welfare: Mobile Abattoirs versus Transportation to Slaughter.
European Review of Agricultural Economics 34 (3): 321–44.
Chang, H., and H. W. Kinnucan. 1991. Advertising, Information, and Product Quality:
The Case of Butter. American Journal of Agricultural Economics 73: 1195–203.
Chang, J. B., J. L. Lusk, and F. B. Norwood. 2009. How Closely Do Hypothetical
Surveys and Laboratory Experiments Predict Field Behavior? American Journal of
Agricultural Economics 91 (2): 518– 34.
Coffey, B. K., T. C. Schroeder, and T. L. Marsh. 2011. Disaggregated Household Meat
Demand with Censored Data. Applied Economics 43 (18): 2343 –63.
Davis, G. C. 1997. The Logic of Testing Structural Change in Meat Demand: A
Methodological Analysis and Appraisal. American Journal of Agricultural Economics
79: 1186–92.
De Jong, G., A. Daly, M. Pieters, C. Vellay, M. Bradley, and F. Hofman. 2003. A
Model for Time of Day and Mode Choice Using Error Components Logit.
Transportation Research Part E: Logistics and Transportation Review 39 (3): 245–68.
Eales, J. S., and L. J. Unnevehr. 1988. Demand for Beef and Chicken Products: Separability
and Structural Change. American Journal of Agricultural Economics 70: 521–32.
———. 1993. Simultaneity and Structural Change in US Meat Demand. American
Journal of Agricultural Economics 75: 260– 68.
Food and Agricultural Policy Research Institute (FAPRI 2014). U.S. Baseline Briefing
Book: Projections for Agricultural and Biofuel Markets. FAPRI-MU Report 2–14.
Accessed online: http://www.fapri.missouri.edu/outreach/publications/2014/
FAPRI_MU_Report_02_14.pdf.
Gallet, C. A. 2010. Meat Meets Meta: A Quantitative Review of the Price Elasticity Of
Meat. American Journal of Agricultural Economics 92: 258– 72.
Gao, Z., and T. C. Schroeder. 2009. Effects of Additional Quality Attributes on
Consumer Willingness-to-Pay for Food Labels. American Journal of Agricultural
Economics 91: 795 –809.
Golan, A., J. M. Perloff, and E. Z. Shen. 2001. Estimating a Demand System with
Nonnegativity Constraints: Mexican Meat Demand. Review of Economics and
Statistics 83: 541 –50.
Gustavsen, G.W. 2015. Consumer Cohorts and Demand Elasticities. European Review
of Agricultural Economics. 42: 217 –37.

24
Meat Demand Elasticities Vary with Price, Income, and Product Category

Hanley, N., W. Adamowicz, and R. E. Wright. 2005. Price Vector Effects in Choice
Experiments: An Empirical Test. Resource and Energy Economics 27 (3): 227–34.
Holmes, T. P., and W. L. Adamowicz. 2003. Attribute-Based Methods. In A Primer on
Nonmarket Valuation, ed. P. A. Champ, K. J. Boyle, and T. C. Brown, 171–219.
Dordrecht: Kluwer.
Ishmael, W. 2014. Why Is Beef Demand Growing as Per-Capita Income Shrinks?
BEEF Editor’s Blog. January 27, 2014. Available at: http://beefmagazine.com/
blog/why-beef-demand-growing-capita-income-shrinks.
Kastens, T., and G. Brester. 1996. Model Selection and Forecasting Ability of Theory-
Constrained Food Demand Systems. American Journal of Agricultural Economics 78: 301–12.
Kinnucan, H. W., H. Xiao, C. Hsia, and J. D. Jackson. 1997. Effects of Health
Information and Generic Advertising on US Meat Demand. American Journal of

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


Agricultural Economics 79: 13 –23.
Lensing, C., and W. D. Purcell. 2006. Impact of Mandatory Price Reporting
Requirements on Level, Variability, and Elasticity Parameter Estimations for Retail
Beef Prices. Review of Agricultural Economics 28: 229–39.
Louviere, J. J., D. A. Hensher, and J. D. Swait. 2000. Stated Choice Methods: Analysis
and Applications. Cambridge: Cambridge University Press.
Lusk, J. L., and F. B. Norwood. 2005. Effect of Experimental Design on Choice-Based
Conjoint Valuation Estimates. American Journal of Agricultural Economics 87: 771–85.
Lusk, J. L., F. B. Norwood, and J. R. Pruitt. 2006. Consumer Demand for a Ban on
Antibiotic Drug Use in Pork Production. American Journal of Agricultural Economics
88: 1015–33.
Lusk, J. L., J. Roosen, and J. A. Fox. 2003. Demand for Beef from Cattle Administered
Growth Hormones or Fed Genetically Modified Corn: A Comparison of Consumers
in France, Germany, the United Kingdom, and the United States. American Journal of
Agricultural Economics 85: 16–29.
Lusk, J. L., and T. C. Schroeder. 2004. Are Choice Experiments Incentive Compatible?
A Test with Quality Differentiated Beef Steaks. American Journal of Agricultural
Economics 86: 467–82.
Lusk, J. L., T. C. Schroeder, and G. T. Tonsor. 2014. Distinguishing Beliefs from
Preferences in Food Choice. European Journal of Agricultural Economics 41 (4): 627–55.
Marsh, T. L., T. C. Schroeder, and J. Mintert. 2004. “Impacts of Meat Product Recalls
on Consumer Demand in the USA.” Applied Economics 36: 897–909.
Mazzocchi, M. (2006). No News Is Good News: Stochastic Parameters versus Media
Coverage Indices in Demand Models after Food Scares. American Journal of
Agricultural Economics 88 (3): 727– 41.
McFadden, D. 1973. Conditional Logit Analysis of Qualitative Choice Behavior. In
Frontiers in Econometrics, ed. P. Zarembka, 105–42. New York: Academic Press.
McFadden, D., and K. Train. 2000. Mixed MNL Models for Discrete Response.
Journal of Applied Econometrics 15 (5): 447 –70.
McFadden, D., W. Tye, and K. Train. 1977. An Application of Diagnostic Tests for the
Irrelevant Alternatives Property of the Multinomial Logit Model. Transportation
Research Record 637: 39 –46.
Moschini, G., and K. D. Meilke. 1989. Modeling the Pattern of Structural Change in
U.S. Meat Demand. American Journal of Agricultural Economics 71: 253–61.
Moschini, G., D. Moro, and R. D. Green. 1994. Maintaining and Testing Separability
in Demand Systems. American Journal of Agricultural Economics 76: 61–73.
Nayga, R. M., and O. Capps. 1994. Tests of Weak Separability in Disaggregated Meat
Products. American Journal of Agricultural Economics 76: 800– 808.
Okrent, A. M., and J. M. Alston. 2012. The Effects of Farm Commodity and Retail
Food Policies on Obesity and Economic Welfare in the United States. American
Journal of Agricultural Economics 94 (3): 611–46.
Park, J., and O. Capps, Jr. 2002. Impacts of Advertising, Attitudes, Lifestyles and
Health on the Demand for U.S. Pork: A Micro-level Analysis. Journal of Agricultural
and Applied Economics 34: 1–15.

25
Applied Economic Perspectives and Policy

Pendell, D., G. Brester, T. Schroeder, K. Dhuyvetter, and G. T. Tonsor. 2010. Animal


Identification and Tracing in the United States. American Journal of Agricultural
Economics 92: 927 –40.
Piggott, N., J. Chalfant, J. Alston, and G. Griffith. 1996. Demand Response to
Advertising in the Australian Meat Industry. American Journal of Agricultural
Economics 78: 226 –79.
Piggott, N. E., and T. L. Marsh. 2011. Constrained Utility Maximization and Demand
System Estimation. In Oxford Handbook of the Economics of Food Consumption and Policy,
ed. J. L. Lusk, J. Roosen, and J. F. Shogreneds, 7–34. Oxford: Oxford University Press.
———. 2004. Does Food Safety Information Impact US Meat Demand? American
Journal of Agricultural Economics 86: 154– 74.
Reed, A. J., J. W. Levedahl, and C. Hallahan. 2005. The Generalized Composite

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


Commodity Theorem and Food Demand Estimation. American Journal of Agricultural
Economics 87 (1): 28–37.
Rickertsen, K. 1996. Structural Change and the Demand for Meat and Fish in
Norway. European Review of Agricultural Economics 23: 316– 30.
Rickertsen, K., D. Kristofersson, and S. Lothe. 2003. Effects of Health Information on
Nordic Meat and Fish Demand. Empirical Economics 28: 249–73.
Rosegrant, M.W., S. Tokgoz, and P. Bhandary. 2013. The New Normal? A Tighter
Global Agricultural Supply and Demand Relation and Its Implications for Food
Security. American Journal of Agricultural Economics 95: 303– 9.
Rutherford, B. 2014. Will Beef Production Align with Demand in 2014? BEEF Editor’s
Blog. May 13, 2014. Available at: http://beefmagazine.com/blog/will-beef-
production-align-demand-2014.
Scarpa, R., S. Ferrini, and K. Willis. 2005. Performance of Error Component Models
for Status-Quo Effects in Choice Experiments. In Applications of Simulation Methods
in Environmental and Resource Economics, ed. Riccardo Scarpa, and Anna Alberini,
247 –73. New York: Springer.
Schulz, L. L., T. C. Schroeder, and T. Xia. 2012. Studying Composite Demand using
Scanner Data: The Case of Ground Beef in the US. Agricultural Economics 43:
49 –57.
Taylor, M., and G.T. Tonsor. 2013. Revealed Demand for Country of Origin Labeling
of Meat in the United States. Journal of Agricultural and Resource Economics 38:
235 –47.
Timmermans, H., A. Borgers, and P. van der Waerden. 1991. Mother Logit Analysis
of Substitution Effects in Consumer Shopping Destination Choice. Journal of
Business Research 24 (2): 177– 89.
Tonsor, G. T. 2011. Consumer Inferences of Food Safety and Quality. European Review
of Agricultural Economics 38: 213 –35.
Tonsor, G. T., and T. L. Marsh. 2007. Comparing Heterogeneous Consumption in US
and Japanese Meat and Fish Demand. Agricultural Economics 37: 81–91.
Tonsor, G. T., J. Mintert, and T. C. Schroeder. 2010. U.S. Meat Demand: Household
Dynamics and Media Information Impacts. Journal of Agricultural and Resource
Economics 35: 1–17.
Train, K. E. 2009. Discrete Choice Methods with Simulation. Cambridge: Cambridge
University Press.
Unnevehr, L., J. Eales, H. Jensen, J. L. Lusk, J. McCluskey, and J. Kinsey. 2010. Food
and Consumer Economics. American Journal of Agricultural Economics 92 (2): 506–21.
US Department of Agriculture, Economic Research Service (USDA ERS). 2014.
USDA Agricultural Projections to 2023.
Weaber, R. L., and J. L. Lusk. 2010. The Economic Value of Improvements in Meat
Tenderness by Genetic Marker Selection. American Journal of Agricultural Economics
92: 1456–71.
Wohlgenant, M.K. 2011. Consumer Demand and Welfare in Equilibrium
Displacement Models. In Oxford Handbook of the Economics of Food Consumption and
Policy, ed. J. L. Lusk, J. Roosen, and J. F. Shogren. Oxford: Oxford University Press.

26
Meat Demand Elasticities Vary with Price, Income, and Product Category

Appendix
Calculating Market Shares and Elasticities
Most of the results are based off the RPL with correlated random para-
meters. The deterministic portion of the utility function is given the esti-
mated parameters (indicated by the ^) is:
 
8
V̂ij = b̂j + v̂ jt dit + âj pj + d̂j p2j , (A1)
t=1

Plugging (A1) into the multinomial logit formula yields:

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


eV̂ ij
Sij |d i = 9 , (A2)
V̂ ik
k=1 e

where Sij is the probability of purchase (or market share) for individual i
and food option j and d i is a vector containing the terms dit . Equation (A2)
cannot be directly evaluated because it contains the random terms dit which
are each distributed according to a standard Normal distribution. Following
Train (2009), the mean share can be approximated through simulation. In
particular, for a set of N draws for dit , the unconditional share can be
written as:

1 N
eV̂ ij
Sj = 9 V̂ . (A3)
N i=1 k=1 e ik

In all our calculations, we use N¼25,000.


The demand curves reported in figures 3 – 5 are created simply by evaluat-
ing (A3) at successively higher price levels. For example, to construct the
steak demand curve, prices of all other products would be set at their mid-
points (see table A2), and equation (A3) would be calculated for psteak = 5,
psteak = 5.2, psteak = 5.2 , . . . , psteak = 7.9, psteak = 8. The resulting share
values are then plotted against the corresponding prices.
The arc elasticities in tables 3 and 5 are calculated as follows. First, equa-
tion (A3) is used to calculate the shares for each of the j ¼ 9 options at the
midpoint of the price levels (see table A3). Call these values S j . Then, to de-
termine arc elasticities associated with a change in the price of good k, the
price of good k is increased (or decreased) from its midpoint by 10% (i.e.,
p′k = 0.1pk , where pk is the midpoint price level for good k) while all other
prices are kept at their midpoints. Then, (A3) is recalculated for each j¼1 to
9. Let S′j be the new share values given the new price p′k . Now, the elasticity,
or percentage change in the share of good j that results from a 1% increase
in the price of good k is:

S′j − S
j 1
g jk = j 0.1 . (A4)
S

27
Applied Economic Perspectives and Policy

Table A1 Characteristics of Survey Respondents

Variable Sample (N512,257) Population/Census

Income , $40K 38.0% 34.4%


$40K , Income , $99K 44.5% 43.1%
Income . $100K 18.5% 22.5%
Age , 24 11.0% 12.9%
25 , Age , 34 20.9% 18.0%
35 , Age , 44 20.6% 17.2%
45 , Age , 54 17.2% 19.0%
55 , Age , 64 12.4% 16.0%
Age . 65 17.9% 17.0%

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


HS degree or less 19.0% 43.2%a
Some college 35.7% 28.6%a
College degree 45.3% 28.2%a
Female 49.6% 51.1%

White 72.4% 77.7%


Black 15.0% 13.2%
Hispanic 12.4% 17.1%

Northeast US 19.1% 18.3%


Midwest 22.3% 21.7%
South US 36.2% 36.8%
West 22.2% 23.2%

Household size 2.63 2.61


Food stamp participation 15.1% 15.0%

a
Education attainment statistics are for individuals 18 years and older; for 25 years and older census data
indicate, 39% have a HS degree or less, 31% have some college, and 30% have a BS degree or higher.

Table A2 Prices Used in Choice Experiment ($/lb)

Product Low Mid High

Ground beef (GB) $2.00 $3.50 $5.00


Steak (SK) $5.00 $6.50 $8.00
Pork chop (PC) $2.25 $3.75 $5.25
Deli ham (DH) $1.15 $2.65 $4.15
Chicken breast (CB) $1.75 $3.25 $4.75
Chicken wing (CW) $0.75 $1.75 $3.25
Rice and beans (RB) $0.50 $2.00 $3.50
Pasta (PA) $2.50 $4.00 $5.50

Table A3 Bureau of Labor Statistics (BLS) Retail Prices for Products Associated with Products
used in the Choice Experiment (2013 average)

BLS Product Descriptor Price ($/lb)

Ground beef, 100% beef $3.40


Steak, sirloin, USDA Choice, boneless $6.83
All pork chops $3.55
All ham (excluding canned ham and luncheon slices) $3.79
Chicken breast, boneless $3.45
Chicken legs, bone-in $1.61
Beans, dried, any type, all sizes $1.42
Rice, white, long grain, uncooked $0.72
Spaghetti and macaroni $1.30
Tomatoes, field grown $1.53

28
Table A4 Main Effects Orthogonal Design Used in Choice Experiment Survey

Ground Steak Pork Chop Deli Ham Chicken Chicken Rice and Beans
Question Beef (GB) (SK) (PC) (DH) Breast (CB) Wing (CW) (RB) Pasta (PA) Block

1 $2.00 $6.50 $3.75 $2.65 $3.25 $1.75 $0.50 $2.50 1


2 $5.00 $5.00 $2.25 $2.65 $4.75 $0.75 $2.00 $5.50 1
3 $3.50 $5.00 $2.25 $4.15 $3.25 $0.75 $3.50 $4.00 1
4 $2.00 $8.00 $5.25 $4.15 $4.75 $3.25 $0.50 $2.50 1
5 $5.00 $8.00 $5.25 $1.15 $3.25 $3.25 $2.00 $5.50 1
6 $3.50 $8.00 $5.25 $2.65 $1.75 $3.25 $3.50 $4.00 1
7 $5.00 $6.50 $3.75 $4.15 $1.75 $1.75 $2.00 $5.50 1
8 $2.00 $5.00 $2.25 $1.15 $1.75 $0.75 $0.50 $2.50 1
9 $3.50 $6.50 $3.75 $1.15 $4.75 $1.75 $3.50 $4.00 1
10 $2.00 $8.00 $3.75 $1.15 $1.75 $0.75 $2.00 $4.00 2

Meat Demand Elasticities Vary with Price, Income, and Product Category
29

11 $3.50 $6.50 $2.25 $2.65 $1.75 $3.25 $0.50 $5.50 2


12 $5.00 $8.00 $3.75 $2.65 $4.75 $0.75 $3.50 $2.50 2
13 $5.00 $6.50 $2.25 $1.15 $3.25 $3.25 $3.50 $2.50 2
14 $3.50 $8.00 $3.75 $4.15 $3.25 $0.75 $0.50 $5.50 2
15 $2.00 $6.50 $2.25 $4.15 $4.75 $3.25 $2.00 $4.00 2
16 $3.50 $5.00 $5.25 $1.15 $4.75 $1.75 $0.50 $5.50 2
17 $5.00 $5.00 $5.25 $4.15 $1.75 $1.75 $3.50 $2.50 2
18 $2.00 $5.00 $5.25 $2.65 $3.25 $1.75 $2.00 $4.00 2
19 $5.00 $6.50 $5.25 $2.65 $4.75 $0.75 $0.50 $4.00 3
20 $5.00 $5.00 $3.75 $1.15 $3.25 $3.25 $0.50 $4.00 3
21 $2.00 $6.50 $5.25 $1.15 $1.75 $0.75 $3.50 $5.50 3
22 $5.00 $8.00 $2.25 $4.15 $1.75 $1.75 $0.50 $4.00 3
23 $3.50 $6.50 $5.25 $4.15 $3.25 $0.75 $2.00 $2.50 3
24 $3.50 $8.00 $2.25 $1.15 $4.75 $1.75 $2.00 $2.50 3
25 $2.00 $5.00 $3.75 $4.15 $4.75 $3.25 $3.50 $5.50 3
26 $2.00 $8.00 $2.25 $2.65 $3.25 $1.75 $3.50 $5.50 3
27 $3.50 $5.00 $3.75 $2.65 $1.75 $3.25 $2.00 $2.50 3

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


Applied Economic Perspectives and Policy
Table A5 Alternative Model Specifications fit to the Choice Data

Model Description Utility Specification

1 MNL with single price effect Uij = bj + a pj + 1ij


2 MNL with alternative-specific price effects Uij = bj + aj pj + 1ij
3 MNL with alternative-specific & quadratic price effects Uij = bj + aj pj + dj p2j + 1ij
4 Mother logit with linear price effects 8
Uij = bj + a jk pk + 1ij
k=1
5 Mother logit with symmetric and quadratic price effects 8
Uij = bj + a jk pk + dj p2j + 1ij , a jk = akj
k=1
6 Mother logit with quadratic price effects 8
Uij = bj + a jk pk + dj p2j + 1ij
30

k=1
7 Error component model with linear price effects Uij = bj + aj pj + eij + 1ij a
8 Error component model with quadratic price effects Uij = bj + aj pj + dj p2j + eij + 1ij a
9 RPL with linear price effects, uncorrelated alternative-specific constants Uij = (bj + sj dij ) + aj pj + 1ij , dij  N(0, 1)
 
10 RPL with linear price effects, correlated alternative-specific constants 8
Uij = bj + v jk dik + aj pj + 1ij , dik  N(0, 1)b
k=1
 
11 RPL with quadratic price effects, correlated alternative-specific constants 8
Uij = bj + v jk dik + aj pj + dj p2j + 1ij , dik  N(0, 1)b
k=1

a
Our specification included 7 mean-zero Normally distributed error components that we parsimoniously summarize in this equation with a single term, eij . In particular, over-lapping error
components were included for: (1) all food products vs. the “none” alternative, (2) all meat products vs. non meat, (3) beef products, (4) pork products, (5) chicken products, (6) higher value meat
products (steak, pork chop, chicken breast), and (7) lower value meat products (ground beef, deli ham, chicken wing).
b
The terms, v jk , are the elements of the lower-triangle of the Cholesky decomposition associated with the covariance matrix of the random parameters.

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


Meat Demand Elasticities Vary with Price, Income, and Product Category

Table A6 Eleven Competing Models Fit to Choice Data (N¼110,295 choices made by
12,255 people)

#
Model Description Parms LLF AIC

1 MNL with single price effect 9 2210542 421102


2 MNL with alternative-specific price effects 16 2209847 419726
3 MNL with alternative-specific and quadratic 24 2209543 419134
price effects
4 Mother logit with linear price effects 72 2209458 419059
5 Mother logit with symmetric & quadratic price 52 2209278 418659

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


effects
6 Mother logit with quadratic price effects 80 2209170 418499
7 Error component model with linear price effects 23 2185449 370944
8 Error component model with quadratic price 31 2185057 370175
effects
9 RPL with linear price effects, uncorrelated 24 2188511 377070
alternative-specific constants
10 RPL with linear price effects, correlated 52 2181162 362428
alternative-specific constants
11 RPL with quadratic price effects, correlated 60 2180574 361269
alternative-specific constants

Table A7 Likelihood Ratio Tests associated with models in Table A6

Test x2 df P-value Explanation

Model 2 vs. 1 1389 7 ,.001 common price specification rejected in favor


of alternative-specific price effects
Model 3 vs. 1 1997 15 ,.001 common price & linear specification effect
rejected in favor of quadratic &
alternative-specific effects
Model 3 vs. 2 608 8 ,.001 linear price specification rejected in favor of
quadratic
Model 6 vs. 4 576 8 ,.001 linear price specification rejected in favor of
quadratic
Model 6 vs. 5 216 28 ,.001 linear price specification rejected; symmetry
rejected
Model 6 vs. 3 747 56 ,.001 MNL rejected in favor of mother logit
Model 8 vs. 7 784 8 ,.001 linear price specification rejected in favor of
quadratic
Model 8 vs. 3 48973 7 ,.001 MNL rejected in favor of ECM
Model 10 vs. 9 14698 28 ,.001 uncorrelated ascs rejected in favor of
correlated ascs
Model 11 vs. 9 15874 36 ,.001 linear price specification rejected in favor of
quadratic; uncorrelated ascs rejected in
favor of correlated ascs
Model 11 vs. 10 1176 8 ,.001 linear price specification rejected in favor of
quadratic
Model 11 vs. 3 57938 36 ,.001 MNL rejected in favor of RPL

31
Applied Economic Perspectives and Policy

Table A8 Correlation Matrices for Random Parameters Implied by the


Variance-Covariance Matrix Reported in Table 1 of the Main Text

Pooled Model
GB| 1.00000 .89604 .94934 .94986 .90205 .84265 .73259 .82490
SK| .89604 1.00000 .78181 .87644 .81557 .70713 .68577 .71424
PC| .94934 .78181 1.00000 .94812 .90936 .84320 .72387 .81291
DH| .94986 .87644 .94812 1.00000 .85318 .90208 .77086 .78847
CB| .90205 .81557 .90936 .85318 1.00000 .81537 .70825 .77927
CW| .84265 .70713 .84320 .90208 .81537 1.00000 .77554 .74024
RB| .73259 .68577 .72387 .77086 .70825 .77554 1.00000 .61305
PA| .82490 .71424 .81291 .78847 .77927 .74024 .61305 1.00000

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


Low Income
GB| 1.00000 .90460 .97424 .93016 .91213 .84683 .74603 .84980
SK| .90460 1.00000 .95565 .85482 .84400 .80553 .71976 .85608
PC| .97424 .95565 1.00000 .95601 .92573 .86040 .78184 .87486
DH| .93016 .85482 .95601 1.00000 .85496 .87121 .78419 .80874
CB| .91213 .84400 .92573 .85496 1.00000 .82613 .74324 .81853
CW| .84683 .80553 .86040 .87121 .82613 1.00000 .77957 .77322
ARB .74603 .71976 .78184 .78419 .74324 .77957 1.00000 .91983
PA| .84980 .85608 .87486 .80874 .81853 .77322 .91983 1.00000
Middle Income
GB| 1.00000 .90499 .94212 .93963 .90076 .82777 .65946 .79907
SK| .90499 1.00000 .91099 .82669 .77654 .78780 .56171 .71179
PC| .94212 .91099 1.00000 .93962 .90344 .82987 .67645 .78150
DH| .93963 .82669 .93962 1.00000 .82993 .88683 .68228 .74167
CB| .90076 .77654 .90344 .82993 1.00000 .79174 .67110 .77510
CW| .82777 .78780 .82987 .88683 .79174 1.00000 .71409 .70726
RB| .65946 .56171 .67645 .68228 .67110 .71409 1.00000 .90046
PA| .79907 .71179 .78150 .74167 .77510 .70726 .90046 1.00000
High Income
GB| 1.00000 .89529 .92203 .93254 .87216 .82419 .69761 .72619
SK| .89529 1.00000 .90379 .79332 .77438 .74790 .59680 .70848
PC| .92203 .90379 1.00000 .92925 .86956 .80092 .62891 .68162
DH| .93254 .79332 .92925 1.00000 .82913 .87381 .79834 .75802
CB| .87216 .77438 .86956 .82913 1.00000 .75307 .68637 .73419
CW| .82419 .74790 .80092 .87381 .75307 1.00000 .85472 .91710
RB| .69761 .59680 .62891 .79834 .68637 .85472 1.00000 .93406
PA| .72619 .70848 .68162 .75802 .73419 .91710 .93406 1.00000

32
Table A9 Arc Elasticities for a $1.50/lb Price Increase by Income

Change in Price

Change in Quantity GB SK PC DH CB CW RB PA

Low income
Ground beef (GB) 21.337 0.140 0.157 0.091 0.352 0.108 0.048 0.033
Steak (SK) 0.257 21.322 0.190 0.061 0.269 0.100 0.044 0.041
Pork chop (PC) 0.329 0.213 21.342 0.102 0.367 0.107 0.052 0.034
Deli ham (DH) 0.292 0.107 0.158 20.834 0.266 0.141 0.070 0.028
Chicken breast (CB) 0.231 0.099 0.118 0.057 20.957 0.102 0.050 0.028
Chicken wing (CW) 0.186 0.096 0.091 0.076 0.266 20.665 0.068 0.024
Rice and beans (RB) 0.140 0.072 0.077 0.066 0.222 0.116 20.508 0.068
Pasta (PA) 0.210 0.143 0.105 0.056 0.270 0.085 0.153 20.555

Meat Demand Elasticities Vary with Price, Income, and Product Category
33

No purchase 0.124 0.052 0.044 0.053 0.169 0.063 0.059 0.029


Middle income
Ground beef (GB) 21.237 0.302 0.135 0.091 0.404 0.068 0.041 0.038
Steak (SK) 0.234 21.468 0.153 0.050 0.234 0.069 0.027 0.027
Pork chop (PC) 0.220 0.323 21.268 0.092 0.420 0.069 0.047 0.034
Deli ham (DH) 0.275 0.190 0.169 20.818 0.291 0.114 0.054 0.031
Chicken breast (CB) 0.193 0.149 0.123 0.048 20.916 0.064 0.047 0.034
Chicken wing (CW) 0.154 0.202 0.097 0.085 0.299 20.604 0.066 0.027
Rice and beans (RB) 0.111 0.092 0.079 0.049 0.255 0.078 20.461 0.084
Pasta (PA) 0.174 0.158 0.096 0.047 0.324 0.054 0.148 20.612
No purchase 0.110 0.086 0.069 0.045 0.220 0.046 0.065 0.041
High income
Ground beef (GB) 21.251 0.407 0.132 0.059 0.429 0.061 0.032 0.030
Steak (SK) 0.187 21.463 0.134 0.025 0.268 0.044 0.017 0.030
Continued

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


Applied Economic Perspectives and Policy
Table A9 Continued

Change in Price

Change in Quantity GB SK PC DH CB CW RB PA

Pork chop (PC) 0.200 0.442 21.085 0.061 0.445 0.058 0.023 0.026
Deli ham (DH) 0.258 0.230 0.175 20.583 0.346 0.089 0.055 0.028
34

Chicken breast (CB) 0.152 0.211 0.104 0.029 20.833 0.042 0.030 0.033
Chicken wing (CW) 0.148 0.233 0.094 0.050 0.279 20.594 0.065 0.084
Rice and beans (RB) 0.113 0.130 0.053 0.045 0.288 0.095 20.404 0.112
Pasta (PA) 0.106 0.228 0.061 0.024 0.329 0.128 0.117 20.767
No purchase 0.119 0.112 0.087 0.017 0.240 0.033 0.040 0.047

Note: Arc elasticities are calculated by determining the percentage change in market shares divided by the percentage change in prices, when moving from the midpoints to the high points of prices
employed in the experimental design; shares for each income category are calculated by averaging over 25,000 draws from the respective estimated parameter distributions.

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


Table A10 Arc Elasticities for a $1.50/lb Price Decrease by Income

Change in Price

Change in Quantity GB SK PC DH CB CW RB PA

Low income
Ground beef (GB) 23.673 0.369 0.438 0.284 0.931 0.231 0.125 0.228
Steak (SK) 0.685 23.526 0.507 0.209 0.802 0.217 0.117 0.271
Pork chop (PC) 0.796 0.521 23.870 0.309 0.952 0.231 0.135 0.231
Deli ham (DH) 0.737 0.291 0.435 22.722 0.798 0.278 0.171 0.201
Chicken breast (CB) 0.634 0.272 0.343 0.194 22.771 0.221 0.131 0.200
Chicken wing (CW) 0.542 0.262 0.277 0.245 0.786 21.431 0.165 0.173
Rice and beans (RB) 0.439 0.204 0.236 0.216 0.698 0.242 21.247 0.426
Pasta (PA) 0.589 0.370 0.311 0.193 0.792 0.196 0.310 23.791

Meat Demand Elasticities Vary with Price, Income, and Product Category
35

No purchase 0.385 0.149 0.145 0.176 0.567 0.150 0.147 0.200


Middle income
Ground beef (GB) 23.594 0.628 0.402 0.249 0.956 0.202 0.102 0.221
Steak (SK) 0.644 23.280 0.438 0.153 0.695 0.201 0.069 0.162
Pork chop (PC) 0.624 0.661 23.782 0.251 0.974 0.204 0.113 0.203
Deli ham (DH) 0.712 0.435 0.471 22.370 0.803 0.290 0.126 0.187
Chicken breast (CB) 0.559 0.353 0.366 0.148 22.402 0.191 0.112 0.204
Chicken wing (CW) 0.478 0.451 0.304 0.237 0.801 21.761 0.148 0.169
Rice and beans (RB) 0.366 0.231 0.249 0.147 0.720 0.219 21.060 0.443
Pasta (PA) 0.513 0.366 0.298 0.144 0.837 0.170 0.287 23.529
No purchase 0.355 0.216 0.220 0.136 0.645 0.145 0.149 0.237
Continued

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


Applied Economic Perspectives and Policy
Table A10 Continued

Change in Price

Change in Quantity GB SK PC DH CB CW RB PA

High income
Ground beef (GB) 22.850 0.609 0.342 0.185 0.822 0.151 0.095 0.139
Steak (SK) 0.416 22.353 0.343 0.089 0.610 0.116 0.055 0.133
Pork chop (PC) 0.443 0.651 22.854 0.191 0.840 0.145 0.070 0.120
Deli ham (DH) 0.526 0.385 0.424 21.928 0.729 0.204 0.147 0.134
36

Chicken breast (CB) 0.352 0.356 0.277 0.101 21.746 0.111 0.088 0.149
Chicken wing (CW) 0.346 0.383 0.254 0.161 0.626 21.456 0.175 0.340
Rice and beans (RB) 0.276 0.233 0.156 0.146 0.629 0.219 21.124 0.436
Pasta (PA) 0.265 0.374 0.176 0.085 0.685 0.268 0.276 23.289
No purchase 0.281 0.205 0.232 0.063 0.555 0.093 0.113 0.204

Note: Arc elasticities are calculated by determining the percentage change in market shares divided by the percentage change in prices, when moving from the mid-points to the low points of prices
employed in the experimental design; shares for each income category are calculated by averaging over 25,000 draws from the respective estimated parameter distributions.

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


Table A11 Estimates from Competing Models for Middle Income Respondents Used to Construct Figure 4 in the Main Text

RPL with Correlated ASCs RPL with Correlated RPL with Uncorrelated MNL with MNL with Linear
Variables and Quadratic Prices ASCs and Linear Prices ASCs and Linear Prices Quadratic Prices Prices

Means of alternative-specific constants


Ground beef (GB) 6.353*a (0.179)b 5.323* (0.097) 3.182* (0.048) 3.831* (0.148) 3.024* (0.042)
Steak (SK) 8.107* (0.737) 5.015* (0.136) 2.873* (0.104) 4.829* (0.621) 2.844* (0.084)
Pork chop (PC) 5.654* (0.235) 4.670* (0.102) 2.419* (0.059) 3.223* (0.211) 2.443* (0.053)
Deli ham (DH) 3.781* (0.136) 3.411* (0.096) 1.263* (0.048) 1.759* (0.100) 1.454* (0.041)
Chicken breast (CB) 7.627* (0.142) 6.117* (0.095) 3.926* (0.044) 4.488* (0.097) 3.440* (0.034)
Chicken wing (CW) 3.419* (0.120) 3.115* (0.098) 0.937* (0.049) 1.575* (0.072) 1.388* (0.035)
Rice and beans (RB) 2.695* (0.101) 2.531* (0.094) 0.443* (0.049) 1.156* (0.045) 1.054* (0.031)
Pasta (PA) 5.856* (0.315) 3.456* (0.110) 1.343* (0.073) 3.597* (0.028) 1.613* (0.064)
Linear price effects

Meat Demand Elasticities Vary with Price, Income, and Product Category
37

Ground beef (GB) 21.512* (0.105) 20.800* (0.014) 20.829* (0.015) 21.277* (0.010) 20.718* (0.013)
Steak (SK) 21.555* (0.236) 20.550* (0.016) 20.549* (0.016) 21.037* (0.200) 20.392* (0.013)
Pork chop (PC) 21.324* (0.135) 20.698* (0.017) 20.715* (0.017) 21.133* (0.130) 20.638* (0.016)
Deli ham (DH) 21.062* (0.104) 20.647* (0.018) 20.665* (0.018) 20.934* (0.099) 20.597* (0.017)
Chicken breast (CB) 22.039* (0.080) 20.908* (0.012) 20.913* (0.012) 21.462* (0.069) 20.677* (0.010)
Chicken wing (CW) 21.256* (0.106) 20.799* (0.022) 20.818* (0.022) 20.915* (0.093) 20.636* (0.019)
Rice and beans (RB) 20.865* (0.073) 20.555* (0.017) 20.557* (0.017) 20.604* (0.063) 20.415* (0.014)
Pasta (PA) 21.894* (0.171) 20.515* (0.019) 20.513* (0.018) 21.592* (0.160) 20.452* (0.017)
Quadratic price effects
Ground beef (GB) 0.107* (0.016) 0 0 0.084* (0.015) 0
Steak (SK) 0.079* (0.018) 0 0 0.051* (0.016) 0
Pork chop (PC) 0.088* (0.019) 0 0 0.070* (0.018) 0
Deli ham (DH) 0.086* (0.021) 0 0 0.070* (0.020) 0
Chicken breast (CB) 0.180* (0.013) 0 0 0.126* (0.011) 0
Continued

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


Applied Economic Perspectives and Policy
Table A11 Continued

RPL with Correlated ASCs RPL with Correlated RPL with Uncorrelated MNL with MNL with Linear
Variables and Quadratic Prices ASCs and Linear Prices ASCs and Linear Prices Quadratic Prices Prices

Chicken wing (CW) 0.121* (0.027) 0 0 0.074* (0.024) 0


Rice and beans (RB) 0.082* (0.019) 0 0 0.051* (0.016) 0
Pasta (PA) 0.178* (0.022) 0 0 0.147* (0.020) 0
Variances of alternative-specific constants
Ground beef (GB) 9.892* (0.484) 9.854* (0.484) 0.952* (0.057) 0 0
Steak (SK) 15.257* (0.671) 15.237* (0.671) 3.660* (0.162) 0 0
Pork chop (PC) 9.909* (0.500) 9.888* (0.501) 0.956* (0.069) 0 0
Deli ham (DH) 9.235* (0.482) 9.252* (0.484) 1.080* (0.078) 0 0
Chicken breast (CB) 10.021* (0.506) 9.990* (0.508) 1.931* (0.073) 0 0
Chicken wing (CW) 11.015* (0.535) 10.989* (0.535) 2.712* (0.132) 0 0
Rice and beans (RB) 8.086* (0.408) 8.091* (0.410) 3.213* (0.151) 0 0
38

Pasta (PA) 7.310* (0.417) 7.317* (0.419) 1.614* (0.105) 0 0


Covariances of alternative-specific constants
SK, GB 11.118* (0.528) 11.095* (0.528) 0 0 0
PC, GB 9.327* (0.467) 9.304* (0.467) 0 0 0
PC, SK 11.201* (0.529) 11.186* (0.530) 0 0 0
DH, GB 8.981* (0.453) 8.976* (0.454) 0 0 0
DH, SK 9.813* (0.521) 9.819* (0.522) 0 0 0
DH, PC 8.988* (0.456) 8.989* (0.457) 0 0 0
CB, GB 8.968* (0.437) 8.951* (0.438) 0 0 0
CB, SK 9.602* (0.504) 9.597* (0.506) 0 0 0
CB, PC 9.002* (0.441) 8.987* (0.442) 0 0 0
CB, DH 7.984* (0.438) 8.007* (0.440) 0 0 0
CW, GB 8.641* (0.456) 8.628* (0.456) 0 0 0
CW, SK 10.213* (0.524) 10.207* (0.525) 0 0 0
CW, PC 8.670* (0.458) 8.668* (0.459) 0 0 0

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016


CW, DH 8.944* (0.457) 8.967* (0.459) 0 0 0
CW, CB 8.318* (0.441) 8.318* (0.442) 0 0 0
RB, GB 5.898* (0.394) 5.907* (0.395) 0 0 0
RB, SK 6.239* (0.466) 6.250* (0.469) 0 0 0
RB, PC 6.055* (0.390) 6.070* (0.392) 0 0 0
RB, DH 5.896* (0.389) 5.926* (0.391) 0 0 0
RB, CB 6.041* (0.387) 6.065* (0.389) 0 0 0
RB, CW 6.739* (0.405) 6.736* (0.407) 0 0 0
PA, GB 6.795* (0.412) 6.805* (0.413) 0 0 0
PA, SK 7.517* (0.486) 7.540* (0.489) 0 0 0
PA, PC 6.651* (0.413) 6.675* (0.415) 0 0 0
PA, DH 6.094* (0.400) 6.131* (0.401) 0 0 0
PA, CB 6.634* (0.402) 6.656* (0.405) 0 0 0
PA, CW 6.347* (0.413) 6.365* (0.414) 0 0 0
PA, RB 6.923* (0.385) 6.937* (0.388) 0 0 0

Meat Demand Elasticities Vary with Price, Income, and Product Category
39

N choices (people) 47952 (5328) 47952 (5328) 47952 (5328) 47952 (5328) 47952 (5328)
Log Likelihood 278669.7 278885.5 282049.7 290518.4 290661.1
a
One asterisk represents statistical significance at the 0.01 level or lower.
b
Numbers in parentheses are standard errors.

Downloaded from http://aepp.oxfordjournals.org/ at University of Nebraska-Lincoln Libraries on May 31, 2016

You might also like