ZT 5.7
3
.1712

B-l 712
MClrCh i993

 

A Complete Systems Analysis
of Nutritional Awareness
,_ and Food Demand

Texos A&M University System - The Texos Agricultural Experiment Stotion - Edword A. Hiler, Director - College Stotion, Texos

[Blank Page in Origiaaiflailetin] V

A Complete Systems Analysis of
Nutritional Awareness and Food Demand

John D. Schmitz

~ and

Oral Capps, ]r.*

.42?‘

x l“ ‘>9
. Q g9
gs vs
,0 s ‘gas

,5)

*Respectively, Assistant Professor, Department of Agricultural Economics, University of Wyoming, Laramie

i WY, and Professor, Department of Agricultural Economics, Texas A&M University, College Station TX.

This research was conducted While Schmitz Was a Research Assistant at Texas A&M University.

KEY WORDS: nutritional awareness, food demand analysis, complete systems.

- —.~:

   

 

O

D Appendix B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Contents

Executive Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv

Acknowledgment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

Theoretical Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

Theoretical Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

Functional Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

Dynamic AIDS Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

Augmenting the AIDS model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

The PDL/AIDS model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

Other Nutritional Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

Seasonality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

Price and Expenditure Elasticities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

Information Elasticities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

Data Sources and Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

Economic Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

Meats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

Dairy Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1

Vegetables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

Fruits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

Fats and Oils . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

Aggregate Commodities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

Nutritional Awareness Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

Empirical Results: Aggregate Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

Empirical Results: Submodels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

Meats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

Dairy Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

Vegetables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

Fruits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

Fats and Oils . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . _ _ . _ _ , _ _ , , _ i, _ 19

Elasticity Estimates for Nutritional Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2O

Concluding Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

Areas for Further Research . . . . . . . . . . . . a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

Final Comments . . . . . . . . . . . . . . . . . . . . .' . . . . . . . . . . . . . . . . . . . . . . . . .  .  . . . . . . . . . . . . . . . . . . . 22

Literature Cited . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

Appendix A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

31

iii

Executive Summary

Nutrition and health issues are becoming increas-
ingly important to American consumers. Yet, agricul-
tural economists know relatively little about the role
that such information plays in determining the de-
mand for food. The purpose of this study is to better
understand the role of nutritional awareness in food
demand. A framework is presented in which neoclas-
sical utility theory is augmented to include nutritional
information. This framework is then applied empiri-
cally to estimate the demand for various food com-
modities.

Nutritional information is measured by four alter-
native methods. The first method includes intercept
shifters, while the three remaining approaches utilize
indices to measure nutritional information. One of
these indices (the C-S index) cumulatively counts the
release of ten major articles relating to nutrition and
health. This index considers nutrition in a very broad
context. The second index (the B-S index) is based on

the release of articles which deal with the possible
linkage between cholesterol and coronary heart dis-
ease; as such it considers cholesterol information
specifically. This index is modeled in conjunction
with a polynomial distributed lag. The third method
of measuring nutritional information includes both the
C-S and the B-S indices.

The demands for beef, pork, and whole milk are
negatively affected by cholesterol information, while
the demands for fish, poultry, lowfat milk, cheese, and
frozen dairy products are positively affected. All food
commodities show a highly inelastic response to the
more global measure of nutritional information. The
inclusion of nutritional information leads to a reduc-
tion in serial correlation problems. Failure to include
nutritional information may also result in overstating
the importance of habit persistence in food demand
analyses.

Acknowledgment

The authors wish to acknowledge review comments made by John Penson, Atanu Saha, and Steve Fuller.

iv

O

Introduction

Nutrition and health issues appear to be major
concerns for consumers. In a recent Food Marketing
Institute survey, 95 percent of respondents indicated
concern about the nutritional content of foods. This
concern appears to involve a few particular nutrients,
most notably fat and cholesterol (Borra). In response,
promotional campaigns for food products are increas-
ing emphasis on issues pertaining to health and diet.
However, relatively little has been done to include this
information within food demand studies (Capps and
Schmitz, Brown and Schrader). The omission of this
information may lead to specification errors in de-
mand analysis.

Erdman has labeled the period since 1977 as the
Preventive Nutrition Era. The Preventive Nutrition Era
began with the publication of Dietary Goals for the
United States by the U.S. Senate Select Committee on
Nutrition and Human Needs in 1977. It is commonly
believed that this publication represents the first
major publication which warns of the dangers of
over-consumption of calories or nutrient elements. In
the years prior to 1977, nutritionists were primarily
concerned with under-consumption of nutrients.

The Preventive Nutrition Era emphasizes issues
such as obesity, whereas the previous emphasis was
on the establishment and attainment of Recom-
mended Dietary Allowances (RDAs). In 1988, the U.S.
Surgeon General indicated that 5 of the top 10 causes
of death in the United States had been linked to the
diet (U.S. Dept. of Health and Human Services). The
efforts in the Preventive Nutrition Era are directed to
establishing linkages between diet and health and to
educating the public to these linkages. Cronin and
Shaw reviewed 10 major articles published between
1977 and 1988 concerning preventive nutrition. Al-
though the issues varied somewhat across articles,
they followed a common theme of weight mainte-
nance, limiting consumption offats, sugars, sodium,
and cholesterol; increasing consumption of fiber and
complex carbohydrates; and avoidance or moderation
of the consumption of alcohol.

Although much evidence has been generated to
support the linkage between nutrition and health,
relatively little is known about how consumers use
this information. Ippolito and Mathois state (p. 20),
“the improved understanding (of the role of diet and
the development of chronic disease) will have little
value unless consumers incorporate the new informa-
tion into behavior”. Traditional food demand models
used by agricultural economists are based on prices
and income. Nutritional awareness is not, however,
dependent on these variables. If consumers do assimi-
late this information into their decision-making proc-
ess, failure to include changes in nutritional

information represents a specification error in the
traditional demand models used by agricultural econ-
omists. This specification error may result in incon-
sistent parameter estimates and misleading inferences
about food demand.

The linkage between nutritional awareness and
food demand has been addressed in recent works by
Brown and Schrader; Chang and Kinnucan; and Capps
and Schmitz as well as others. Brown and Schrader
consider the effect of information about cholesterol
on the demand for shell eggs. In their paper, an index
of cholesterol information is generated which is later
used in the works by Chang and Kinnucan and Capps
and Schmitz. The Brown and Schrader index consti-
tutes the number of articles in the available medical
literature which deal with the impact of cholesterol
levels on health. This index represents the cumulative
sum of all articles which establishes a link between
cholesterol and heart disease. This sum is sub-
sequently adjusted by subtracting those articles which
fail to establish this link. Using this index, Brown and
Schrader concluded that shell egg consumption de-
creased as a result of nutritional awareness, all other
things held constant.

Capps and Schmitz used the same index using a
demand systems framework. Annual data were used
to analyze demands for beef, pork, chicken, and fish
in the United States. They found that the cholesterol
index had a significant positive impact on per capita
fish and chicken consumption, but this index had a
significantly negative impact on pork consumption.
Although this index was negatively related to per
capita beef consumption, it was not statistically signifi-
cant.

Chang and Kinnucan expanded the Brown and
Schrader index to include those journal articles not
only in the United States but also in Canada. The
commodities studied included butter, margarine,
shortening, and salad oils. Quarterly data from the
second quarter of 1975 through the third quarter of
1986 were used in this analysis. A variation of the
Almost Ideal Demand System (AIDS) model, which
uses consumption levels in lieu of budget shares as
endogenous variables, was used. In addition to prices
and income, their model included seasonality, adver-
tising, and the cholesterol index. They concluded that
the negative information in the cholesterol index out
weighed the positive information from the advertising
message. They were also able to show that cholesterol
information had a negative affect on butter consump-
tion but a positive affect on salad oil consumption. The
effect of this information for margarine and shortening
was also positive but was not statistically significant.

Objectives

Despite the aforementioned studies, little is yet
known about the role of nutritional awareness in the
retail demand for food. Given that the level of health
and nutrition information available to the consumer
has grown substantially, more knowledge about the
impact of such information on food demand is needed.
If consumers are in fact responding to this informa-
tion, the sharp growth in the level of information
available to consumers should result in a noticeable
impact on food demand. Consequently, ignoring such
variables could give rise to considerable specification
error in demand models.

This study is an attempt to better understand the
role of nutritional awareness in food demand. Specifi-
cally, the objective of this study is to identify and assess
the effects of nutritional information on specific retail
food groups. The central question of this effort is
simply the following: do consumers respond to nutri-
tion and health information? In each of the models
presented later, the null hypothesis that consumers do
not respond to such information is tested. A positive
relationship between nutrition and health information
and the demand for a food would indicate that con-
sumers are consuming more of the food item as a result
of nutritional awareness, ceteris paribzrs, while a nega-

tive relationship between such information and the
demand for a food would indicate a decrease in the
consumption of the food item resulting from nutri-
tional awareness.

Approach

Nutritional awareness may affect food demand in
several ways. The models developed within this study
analyze two predominant methods. These are:

1. budget allocations among the major food group-
ings; (i.e. consume fewer meats and dairy pro-
ducts while consuming more fruits, vegetables,
and cereal products), and

2. budget allocations within any of these major
groups (i.e. for meats, consumers may consume
less beef and pork while consuming more
chicken and fish).

In this study, six complete demand systems are
estimated. Food categories for these systems are
shown in Figure 1. The first demand system examines
the effect of nutritional awareness on major food
groups. This model consists of the eight major food
categories (fresh fruits and vegetables are combined)
shown in Figure 1. This model is estimated using
annual data from 1960 through 1988.

Food Expenditure

Figure 1. Selected food categories and submodel formulations to be analyzed.

Meats Dairy Fresh Fresh Fats &
Fruit Vegetable Oils
Beef Whole Milk Apples Potatoes Butter
Pork Lowfat Milk Oranges Lettuce Margarine
Poultry Cheese Bananas Tomatoes Shortening
Eish frozen Other Other Fresh LSalad Oil
Processed Eggs Cereals Sweets
Fruit & Veg

The next five models consider reallocation within
the subgroups of meat, dairy, fruits, vegetables, and
fats and oils. The three remaining categories are not
considered further because of the lack of data corre-
sponding to individual food items. Quarterly data over
the period from 1968 through 1988 are available for
meats and thus are utilized. The other four subgroup
models use only annual data due to the lack of available
quarterly data.

Overview

The theoretical framework and supporting model
development are discussed in the next section. The
formulations of specific models as well as descriptions
of data used in these model formulations are the
subject of the third section. Empirical results are pre-
sented and discussed in section four. Conclusions,
summary comments, limitations, and future directions
for research are addressed in the final section.

Theoretical Development

Nutritional awareness is akin to advertising in that
both provide information about products for consum-
ers. They primarily differ in that advertising normally
emphasizes a message of technical or sensual charac-
teristics, while nutritional awareness emphasizes a
message of how consumption may affect health. Be-
cause of the similarities between advertising and nu-
tritional awareness, it seems appropriate to include
nutritional awareness information in demand systems
using the same basic techniques as those used success-
fully in the study of the effects of advertising on
demand. Specifically, the use of distributed lag models
may prove to be very helpful in explaining the impact
of nutritional information on food demand.

This study begins by developing an appropriate
empirical framework to determine the implications of
nutritional awareness on food demand. Theoretical
developments and functional form issues are dis-
cussed. The Almost Ideal Demand System (AIDS) is
then presented. The dynamic AIDS model is used as
the basic demand model throughout this study. For
the purposes of this study, it is necessary to expand
the AIDS model to include nutritional information as
well as seasonality. These augmentations are dis-
cussed individually following a discussion of the AIDS
model. This section then proceeds with a discussion
of price, expenditure, and income elasticities and
their associated variances. Finally, the calculation of
short-run, interim, cumulative, and long-run elastici-
ties associated with the polynomial distributed lag
AIDS model is discussed.

Theoretical Framework

The theoretical framework which is used in this
study follows the framework applied by Capps and
Schmitz. In this framework, the utility function in time
period t is expressed as

v, = urea 6cm]. <1>

where 6 (rt) reflects the preferences of the consumer
with regard to commodity vector Qt. This framework
follows from the work of Basmann in the area of
consumer demand under variable preferences. The
vector r represents a vector of exogenous state vari-
ables. These state variables may correspond to stocks
of knowledge, psychological stocks of habits, or physi-
cal stocks of goods. This framework assumes that the
formulation of consumer preferences rests in part on
information about characteristics of Q contained in r.

If the stock of knowledge is constant over time,
then selection of a commodity bundle is not altered.
Thus the term 6 (rt) may be dropped from (1) leaving
the more traditional utility specification. However, if
the stock of information changes over time and the
information in r influences the choice of commodity
bundles, then failure to include this information may
yield biased estimates of structural demand parame-
ters. The information included in r may explain cases
of structural change in the demand for meat products
that have been indicated by several studies (Nyankori
and Miller; Chavas; Dahlgran; Thurman; Eales and
Unnevehr; Moschini and Meilke; Goodwin; Choi and
Sosin).

When modeling health and nutritional awareness,
the vector r may consist of information about any
number of items. The awareness of nutrition and
health may center predominantly on cholesterol, so-
dium, dietary fiber, saturated fats, and caloric intake.
Importantly, the message presented by the medical
profession or nutritionists may not be the same mes-
sage received by consumers. It is the attitude of the
consumer toward nutrition and health which deter-
mines the choice of the commodity bundle, not neces-
sarily the factual information.

Consumer attitudes may be affected by many fac-
tors, for example advertising, which often focuses on
certain aspects of nutrition. However, attitudes are
unique to each individual. Thus the collection and
inclusion of consumer attitudes would require consid-
erable effort. Since practically no data currently exist
in this area, the use of attitudinal variables, which are
specific to individuals, would require crosssectional
studies. Potential for work in this area exists, espe-
cially with the recent release of the 1987-88 Nation-
wide Food Consumption Survey (NFCS).

In time-series applications, proxy variables are
needed for r. Any variable or set of variables highly
correlated with r is a candidate to be used as a proxy.
Proxies may be advertising information or a number
of different measures of scientific information about
health and nutrition. The quality of each alternative
depends on which sources are used most by consum-
ers. Data on sources of consumer information about

health and nutrition factors also will be available in the
1987-88 NFCS.

Functional Form

Marshallian demand functions in compliance with
(1) are given in general form as

q,- = 4F’, M; 9(r,)]- (2)

However, theory cannot specify a functional form
for these demand functions or specify a form for the
6 function. Several alternative complete demand sys-
tems are viable candidates for use in estimating (2) as
well as incomplete demand systems. The Almost Ideal
Demand System (AIDS) has been selected due to con-
sistency with the theoretical framework above, its
ability to deal with aggregate as well as disaggregate
data, its relative ease of use (when the linear approxi-
mation is used), and its flexibility. A dynamic version
of the AIDS model is also often used to circumvent
potential problems of serial correlation. Furthermore,
the dynamic AIDS model can allow for habitual pur-
chase behavior which is often found in consumer
demand studies.

Less is known of the form for the 6 function since
very little previous work has been conducted. In their
study of the effects of an Alar scare in apples, van
Ravenswaay and Hoehn suggest four possible ways
that Alar can affect the demand for apples. Their first
hypothesis assumes that consumers immediately re-
spond to information about Alar. In this case, the
appropriate model contains a binary intercept shifter
variable which allows for a shift on the date the Alar
scare was revealed by the media. This specification,
however, fails to establish a connection between Alar
information and apples. Thus three alternative hy-
potheses were considered. Their second hypothesis
is that “a change in the safety of a product does affect
consumers’ purchases, but the information is eventu-
ally forgotten” (p. 10). This hypothesis gives rise to the
use of a distributed lag framework. Due to the similar-
ity between information and advertising, and the suc-
cess of distributed lags in advertising research, such a
hypothesis seems plausible.

The third hypothesis put forth by van Ravenswaay
and Hoehn is that “a change in risk information is not
forgotten and affects purchases until an an-
nouncement is made that the risk has been eliminated”
(p. 10). This approach in the area of nutrition would
be consistent with the use of a running total of nutri-
tion or health related articles. These running totals
have been used by Brown and Schrader, Capps and
Schmitz, and Chang and Kinnucan in regard to the
development of the cholesterol index. The fourth
hypothesis put forth involves a combination of the
first hypothesis with either the second or third. In this
case, “a consumer’s perception of the magnitude of a
safety problem may increase with subsequent an-
nouncements” (p. l0).

Since little is known of the structure of 6, several
alternative structures are pursued in this analysis.
These include the use of binary intercept shifters,
running totals, and a distributed lag model. Further-
more, the simultaneous use of a distributed lag model
and a running total of articles is investigated. In es-
sence, the hypotheses put forward by van Ravenswaay
and Hoehn are adopted in this study. r

Dynamic AIDS Model

The static AIDS model was first introduced by
Deaton and Muellbauer (1980b). Dynamics were later
added to the static model specification by Blanciforti
and Green who included a lagged quantity variable
(QLH). Later, Eales and Unnevehr incorporated dy-
namics by considering the change in levels of each
variable instead of the variables themselves. Following
the work of Anderson and Blundell, Johnson et al.
presented a dynamic AIDS model with lagged budget
shares instead of the lagged quantities used by Blanci-
forti and Green. Using the Johnson et al. approach, the
dynamic AIDS model can be expressed as

N
w, = a, + EyglnPfi
j-i
X: (5)
+ fliln F

I

* ciwLt-l ‘I ‘n’

where 10,-, is the budget share for good i (zvi, =
PitQit/Xt); Pi, is the price of good j, X, is the total
expenditure on the N goods included within the sys-
tem; sit is a stochastic error term; and

1v
lnP, = a0 + zajlnlv,

1-1

1N N
+ 

i-l j-l

(4)

The parameters which require estimation are the
06s, y’s, B’s, and Cs. Due to the nonlinearity of para-
meters in equation (4), the AIDS model is often ap-
proximated by substituting Stone’s Approximation
(ln(Pt*)) for ln(PQ where

N
lnP: = 2143}, 11113,. <5)
j=l

Equation (5) includes lagged budget shares as is
done by Eales and Unnevehr. This specification avoids
potential problems with simultaneity biasl. The AIDS
model with this substitution is linear in parameters
and thus is commonly referred to as LA/AIDS.

lSimultaneity bias occurs when a change in the dependent variable
(W1) affects the magnitude of one of the independent regressors, in
this case the budget shares in equation (5).

The integrability conditions are required to guar-
antee the existence of an underlying utility function.
To maintain these conditions, it is necessary to impose
the classical restrictions on the demand system (Dea-
ton and Muellbauer, 1980a p. 49-50; Varian, p. 135-
139). Three restrictions, often referred to as the
TRIAD, involve three conditions. Namely, these con-
ditions are homogeneity, Slutsky symmetry, and ad-
ding up. The first two of these three conditions require
restrictions to be placed on the AIDS or LA/AIDS
models (Deaton and Muellbauer, 1980a p.75-78).
Homogeneity requires

N
21:7,]- = 0, Vi = [1---1v], <6)
,.

while Slutsky symmetry requires

vi,- = Yfi, V131’ ; (in). <7>

Homogeneity and Slutsky symmetry restrictions
remain unaltered when switching from a static to a
dynamic structure. However, the adding-up restric-
tions are different when the dynamic AIDS model is
used. In the static model, the presence of adding-up
requires that one equation be dropped from the sys-
tem in order to avoid a singular variance-covariance
matrix. The coefficients for the omitted equation are
determined by

1v 1v 1v
2ai=1, 2yij=0,and EBi=O. (3)
' i=1 i=1

i=1

The same procedure as used in the static proce-
dure could be used if the coefficients for the lagged
budget share also were forced to sum to zero. This
restriction results in some lagged terms being nega-
tive. However, the appropriate range for these coeffi-
cients lies between zero and one because they relate
to habit persistence.

An alternative approach is used here. In this case,
all N equations are estimated. In addition to the homo-
geneity and Slutsky symmetry conditions above, ad-
ding up requires

N 1v 1v
ZYFQ 291F0- ZM t 510i) = 1 » (9)
i=1 i=1

.=l

where O; represents all variables in the model except
prices and expenditure. Equation (9) treats all terms

other than prices and expenditure as part of the inter-
cept.

Augmenting the AIDS model

Through the course of this study, the AIDS model
is augmented by three different methods. These aug-
mentations involve the addition of a Polynomial Dis-
tributed Lag (PDL) formulation, other measures of
nutritional information, and seasonality. To simplify
the presentation of these models, each augmentation
is presented separately.

In each case, the intercept term is expanded. The
homogeneity and symmetry conditions remain as pre-
viously specified while the adding up restriction is
imposed by equation (9). Actually, the lagged budget
share in the dynamic AIDS model already represents
an augmentation of the static model. Each of the
aforementioned augmentations is discussed in turn.

The PDL/AIDS model

The second hypothesis put forth by van Raven-
swaay and Hoehn considers the possibility that con-
sumers hear and respond to a message but forget that
message over time. The incorporation of this idea of
“response and decay” into empirical analyses is often
done through the use of distributed lag models
(Almon; Chen, Courtney, and Schmitz). The use of
distributed lags has been successful in many areas, and
has been especially effective in studies of advertising
(Kinnucan and Forker; Baye, Jansen, and Lee). Due to
similarities in the nature of messages provided by
advertisers and nutritionists, one would expect that
distributed lag models also may work well in the
analysis of nutrition awareness and food demand.

Baye, Jansen, and Lee included a geometric lag
structure for advertising within an AIDS model. The
geometric lag structure allows for an initial response
with geometrically declining response thereafter. Al-
though this structure may be appropriate in the study
of nutrition and food demand, one would not want to
limit the response to such a geometric shape. Alterna-
tively, one could use the polynomial distributed lag
(PDL) structure developed by Almon. The PDL speci-
fication is capable of producing a geometric lag shape
as well as other more flexible shapes. One of the
common shapes displayed by the PDL model is a
humped-shaped pattern. This shape allows a gradual
response to the information followed by a gradual
decline in this response. The PDL structure is pre-
ferred here since it can accommodate a geometric
shape as well as other shapes such as the humped-
shape. '

The use of a PDL requires the analyst to specify a
priori the length of lag and degree of polynomial. In
the PDL model, the degrees of the polynomial should
be less than the length of lag minus one so as to allow
the PDL formulation to reduce the total number of
structural parameters to be estimated. The PDL model
is discussed in detail by Pindyck and Rubinfeld and by
Kmenta.

The reduced-form equation for the PDL can be
combined with the dynamic AIDS model in (3), giving
rise to a Polynomial Distributed Lag augmented Al-
most Ideal Demand System or the PDL/AIDS model.
The reduced-form for the PDL/AIDS becomes

” x
w“ = a‘. + 2761111’)? + Bibi —'
i=1 g

(10)
D!

‘I ciwiJ-l J’ Eat-Ra + 8i!’

d=0

where Di is the degree of polynomial desired for item
i, cdi is the dd‘ reduced-form coefficient for the id‘ item,
and Rd“ is the dth lag of the information variable for
the id‘ equation. The lag structure for Rdi, is calculated
as

L

R0» = grt-k’ Md <11)

L:
Rd, = 21:45,‘, v d = [1---D‘.],
=1

where Li is the length of lag associated with the ith
item.

In the static version of the AIDS model, it is neces-
sary to omit an equation and derive coefficient esti-
mates for this omitted equation. This derivation is
done through the use of the adding-up restrictions.
When the lag structure is allowed to vary across equa-
tions, different variables are placed in each of the
reduced-form equations. Considering this issue, the
coefficients for the Nth equation can be calculated as

N-l

E cdiRdi
c = _='_=1___ _ <12)
an Raw

These can only be imposed at one point, typically
at the means of the data sample. In the dynamic
structure, all the coefficients are estimated so this
derivation is not necessary.

The structural coefficients for the PDL/AIDS can
be obtained by substituting the reduced-form coeffi-
cients into

Di
wh- = c0, + gcfi-V. <15)

These structural coefficients (tun) represent the
impact of the state variable on the it budget share in
the 1th period.

Two additional restrictions known as head and tail
restrictions are often used with a PDL. A head restric-
tion constrains the information (r) included in the PDL
to have no impact in the time period before its release.
Thus a change in r cannot be anticipated. A tail restric-
tion constrains this information to have no impact
after L,- periods have passed. Either, both, or neither
of these restrictions may be used as needed for the
particular modeling application. The head restriction
requires

Dz

Q4‘, E gave,“ = 0, <14)

while the tail restriction requires
D i
_ d =
“(ti-m a g; (Lin) C“ 0' (15)

Other Nutritional Variables

As discussed earlier, alternatives to the PDL speci-
fication also are used to determine the impact of
nutritional information. These alternatives include the
addition of either binary intercept shifters or continu-
ous variables to the model.

The first method used by van Ravenswaay and
Hoehn represents the most straight forward and per-
haps the most naive approach to modeling a “struc-
tural change.” This method simply involves the
selection of critical times when structural changes
may have occurred. An application of this method
requires selecting appropriate times at which demand
shifts may have occurred. The selection of proposed
dates for structural changes is addressed in the data
section. Once periods are selected, they can be incor-
porated into the model using binary variables. These
variables would equal one if the data are from the
period in question and zero otherwise. A dummy
variable associated with one of these periods must be
arbitrarily dropped to avoid a singular matrix resulting
from the “dummy variable trap.”

The third hypothesis put forth by van Ravenswaay
and Hoehn calls for a continuous variable which does
not allow for information decay.2 A good example of
using this hypothesis is the previously described cho-
lesterol index developed by Brown and Schrader. An
alternative and more general index also can be devel-
oped from the articles reviewed by Cronin and Shaw.

zRecall that the second hypothesis was the basis for applying the
polynomial distributed lag model.

3

These ten articles represent presumably major works
which lead to new understandings about nutrition and
health linkages. Using the release dates of these ten
articles, a new index which cumulatively counts these
releases is employed in this study. The value of this
index remains at zero until 1977 and increases with
the release of each article until reaching the value of
ten upon the release of the last article in 1988. This
index is simply added to the AIDS model in the same
fashion as the lagged budget share.

Seasonality

When quarterly or monthly data are used in de-
mand studies, seasonality often plays a significant role.
To allow for seasonality in the AIDS model and its
variations, only a few adjustments need to be made.
Seasonal dummies are easily added to equation (5).
This augmentation results in a new oti which appears
as

H-l

“i * g Pnzsm’ - (16)

where H is the number of seasonal periods in the year;
Sh, is a binary variable equal to one if data are from
period h and zero otherwise; and um is an intercept
shifter. Again, one of these dummy variables must be
arbitrarily omitted to avoid the singularity problem.

Price and Expenditure Elasticities

The magnitude of coefficients determined in the
estimation of the AIDS model are affected by the scale
of the data, making interpretations and comparisons
with the existing literature difficult. To help alleviate
this difficulty, economists often use elasticity esti-
mates. These elasticities are calculated from the esti-
mates of the structural parameters, but the
interpretation of these measures are not dependent on
the scale of the data. Thus elasticities are useful in
making comparisons with previous studies.

Green and Alston review price elasticity calcula-
tions in the AIDS and LA/AIDS models. In their discus-
sion, Green and Alston assume that the budget shares
in Stone’s approximation, equation (5), are not lagged.
Under this assumption, they show that

dln(P') =

N
. lnP . 6. (17)

where 2k,- is the Marshallian cross-price elasticity and
fik- is the Kronecker delta (1 if k equals j and 0 other-
wise). When the budget shares in Stone’s approxima-
tion are lagged, equation (17) simply equals w,» Thus
the formula for the Marshallian elasticities is given as

__ _ + Yii pi w] (18)
ll u w_
I
where 6;)- again refers to the Kronecker delta. The
intercept of the AIDS model does not appear in this
formula. Since each of the augmentations discussed
previously are really just adjustments to the intercept,
the uncompensated elasticity formula is unaffected by
augmentations to the AIDS model. Compensated elas-
ticities may be calculated from the uncompensated
elasticities above by using Slut'sky’s equation

6;’ = eg- + WJ-Tlg» (19)

where n, is the expenditure elasticity calculated as

n, = (HE) <20)

wt

One can also adjust the expenditure elasticity to
an income elasticity. Using the chain rule, the income
elasticity can be shown to be

60,1:

6X I
61 Q‘.

. <21)
ma! x

ax Q, a1 x

9

60H‘ 6x1]:

where X is expenditure on the group and I is income
(Manser; Capps, Tedford, and Havlicek). The last term
is an elasticity between expenditure and income. This
term can be estimated using an auxiliary regression. A
double log OLS regression yields such an elasticity
estimate directly. This auxiliary regression appears as

1n(X,) = a + 1111(1), (22)

where I, is a measure of income and r is the desired
elasticity estimate. The income elasticity is then calcu-
lated as mi.

Using the preceding formulae, a complete set of
compensated and uncompensated price elasticities as
well as expenditure and income elasticities can be
calculated. These represent only point estimates. An
equally important consideration is the variance asso-
ciated with these estimates. Dorfman, Kling, and Sex-
ton address the importance of generating some
measure of variance associated with these elasticity

calculations. The work by Dorfman et al. focuses on
the use of boot-strapping, Taylor’s series expansion,
and other approaches.

Chalfant discusses a technique for estimating the
associated variances’ for elasticity estimates in the
AIDS model. Estimates of these variances may be cal-
culated through use of the variance-covariance matrix
of coefficients. The variance of the elasticity is a linear
function of parameters if wi is assumed to be fixed in
repeated samples. With variance and covariance esti-
mates of the parameters, one can approximate the
variance of the elasticity. Mathematically, the expres-
sion for the variance of the uncompensated price
elasticity appears as

t/ARm-j) = VA -a.. + -_-_Y‘Y' _ 5"“? . <25)

i! wi

Simplifying this expression, one obtains the variance
of the uncompensated price elasticity

VAR(e,j) = izvanqfi)

w? 2W- (24)
+ J-zvanqsg - -2’-€0V(Y,,-,B,)-
W; wi

Following a similar approach, the variance for the
compensated elasticity can be calculated as

VARa-zg) = -l—2VAR(y|j , <25)
i

and the variance of the expenditure elasticity can be
calculated as

vanml.) = lit/Anny). <26)

A variance associated with the income elasticity
cannot be calculated since this calculation involves
the product of two coefficients. The approaches dis-
cussed by Dorfman et al. are applicable in this circum-
stance. Alternatively, one could assume that the
coefficient t from equation (22) is a constant. In this
case, the variance of the income elasticity would equal
the variance of the expenditure elasticity times 1: .

Armed with point elasticities and variances of
these elasticity calculations, one can determine
whether goods are substitutes, complements, or inde-

pendent. Such determinations are made by the sign of
the compensated cross-price elasticities (Hicks-Allen
definition). If negative, the goods are complements, if
positive the goods are substitutes, and if not signifi-
cantly different from zero, they are independent. To
make this determination, the t-test could be used. This
test is calculated as ~

I!

t = -—€i——, (27)

if VAR(ei;-)

where t follows the t-distribution. Since this variance
is only an approximation, a cutoff value of 2 can be
used.

Similar tests may be used to determine whether
demand is elastic or inelastic. The nature of demand
elasticity is determined by whether the uncompen-
sated own-price elasticity is greater or less than one in
absolute value. Additionally, one can determine
whether a good is an inferior, normal, or superior
good by determining whether the income elasticity is
less than zero, between zero and one, or greater than
one, respectively.

Information Elasticities

Nutritional information has been added to the
AIDS model in three different forms. Two of these
forms involve the use of continuous variables. In the
cases where continuous variables are used, elasticity
calculations can be derived. No calculations can be
given for the binary intercept shifter variables.

The simpler case involves the elasticity calculation
for the C-S index. In this case, the index C is common
to all equations. To determine the elasticity, one must
differentiate w, with respect to C. This is

 

(23)

. 8P
X piiQJ + Qi._'
6C
x2

Assuming that prices and expenditure are inde-

pendent of C, equation (28) can be simplified to

E = f: _Q’i <29>

6C X 6C
Note that dwi/dC equals the coefficient estimate asso-
ciated with C. We will call this coefficient zpi. Solving

(29) for 0Q,/6C and multiplying by C/Q; yields, at the
sample means,

6X
' Piqigc“

J

3?. = LE or»
wt

6C

||O|
[GI

- L
ii

9

IO
fin

which gives the elasticity 0f Q,- with respect to C.
Specifically this measure constitutes the percentage
change in quantity demanded that would result if a
one percent change in awareness were to occur in that
period, ceteris paribus.

In some cases such as advertising, it is possible
that C may be specific to certain equations. Thus, it
may be necessary for C to be noted as  Only a slight
modification is needed for a more general derivation.
In the case of C -, it becomes necessary to consider dwi/
0C,» Noting again that expenditure and prices are
exogenous, this expression can be stated as

95 . 53%, 61>

ac, x ac,

Solving for 6Q;/6C,-, we obtain
6Q‘ - awix =  (32)



Again, this result will be converted to an elasticity by
multiplying by Cj/Qi. Evaluating this at the mean, we
obtain
ao, c, z aw, x c].
ac, Q, 6C,- P, Q, ,-

(3 5)

The PDL model is a dynamic structural model.
Several alternative elasticity measures result from this
dynamic structure. These include short-run, interim,
cumulative, and long-run elasticities. Note that the ips
in (33) correspond to the cos in the PDL. In the case
above, one coefficient zpi is involved per equation. In
the case of a PDL, several ips are involved.

The short-run elasticity can be found by replacing
the zpij in (33) with CUQi. Interim elasticities would be
calculated by-replacing 111i)‘ with wk where l is the
number of periods for which the interim elasticity is
desired. Cumulative elasticities are calculated by re-
placing ipi- with (Eton) over the range from 0 to the
number o periods for which the cumulative elasticity
is to cover. A long-run elasticity is a special case of the
cumulative elasticity. The long-run elasticity is the

cumulative elasticity where all values of w are included
in the summation.

Data Sources and Descriptive Statistics

The estimation of the previously mentioned
econometric models require appropriate data. These
data needs can be broken down into two categories:
economic data and nutritional awareness data. The
first section includes a discussion of price and con-
sumption data that are typically included in a demand
system. In the second section, nontraditional data
used to measure the impact of nutritional awareness
are discussed. Within each of these two sections, the
sources of individual data series as well as any special
techniques used in the collection of these series are
documented. Descriptive statistics for each data series
are included with each section.

Economic Data

The estimation of a complete demand system re-
quires information pertaining to prices, expenditures,
and other variables depending on the phenomenon
being investigated. The analyst gathers information
about prices and consumption, and defines total ex-
penditure to be the summation of expenditures on
each item. Thus, in the complete demand system
framework, expenditures are determined from prices
and quantities consumed. This section documents
sources of price and quantity data needed in these
models. For clarity, prices and quantities for each
model will be presented separately.

Throughout this discussion, prices are presented
in cents per pound while quantities are presented in
pounds per capita. For each model, issues which are
specifically encountered in the development of the
respective data series are discussed. Typically, for
aggregate commodities, prices correspond to indices.
To convert from indices to cents per pound, it is
necessary to obtain one “representative” price. This
representative price is then used to convert the index
to cents per pound.

Obtaining a representative price can be a difficult
process, especially for aggregate goods. In each case,
sources for these representative prices are given as
well as any other conversions which may be neces-
sary. Quantities are normally published in pounds per
capita, thus no conversion is necessary.

Meats

Meat is divided into four categories: beef, pork,
poultry, and fish. Beef, pork, and poultry play a major
role in the total food budget. Quarterly data are readily
available for these commodities. Consumption data
for these three commodities are taken from Livestock
and Poultry: Situation and Outlook, which is pub-
lished by the U.S. Department of Agriculture (USDA);
price indices are taken from the Department of Com-
merce, Bureau of Labor Statistics (BLS). These indices
are then converted to cents per pound using selected
representative prices published by the USDA in the

Food Consumption, Prices, and Expenditures (F CPE)
series. Representative prices for 1988 are $3.00 per
pound for beef, $2.10 per pound for pork, and $1.22
for poultry.

Fish data, however, are much less readily avail-
able. The BLS publishes monthly price indices for
many food items, including fish. However, no repre-
sentative price is available to convert this index to
price per pound. A conversion factor of $4.00 per
pound for the year 1988 is used.

Quarterly fish consumption data also are not read-
ily available. The U.S. Department of Commerce
(U SDC) does not publish much data pertaining to food
items and the Department of Agriculture does not
normally include data on fish. Annual fish consump-
tion data (total and per capita) are published in FCPE.
However, quarterly data are needed for fish consump-
tion if quarterly data are to be used in the analysis. As
a result, quarterly fish consumption data are generated
from a variety of sources.

In generating quarterly fish consumption data, it
is assumed that fish disappearance is proportional to
but not necessarily equal to consumption. Fish disap-
pearance is calculated as

01s, = LAND, + IMP,

e4)
+ STOCKH - EXP, - STOCK,

where DIS is the level of disappearance in t, LAND is
the reported landings, IMP is the quantity of imports,
STOCK is the cold storage stocks at the end of the
period, and EXP is the quantity of exports for the
period.

Each of these variables is available or can be cal-
culated on a monthly basis for the period from 1968
through 1988. They are however published in a variety
of places. Landings are found in Fishery Statistics of
the United States (USDC) over the period 1968-1979,
and in Fisheries of the United States (U SDC) over the
period 1980-1988. Cold storage holdings may be
found in the Survey of Current Business, also publish-
ed by the USDC. ~

The USDC also publishes the value of monthly
trade data, but not quantities. This situation makes the
calculation of import and export data more difficult.
Annual quantities are, however, published by the
USDC in Fishery Statistics of the United States, and
Fisheries of the United States. Again, it becomes
necessary to allocate the annual quantities of imports
and exports among the months. Value weights are
generated by dividing the value of imports (exports)
for the month by the total value of imports (exports)
for the year. These weights are then adjusted by the
retail price index for fish to remove seasonal price
variation. The annual import and export quantities are
then multiplied by these weights to determine
monthly import and export levels. Mathematically, the
monthly disappearance is calculated as

1O

01s,, = LAND,” + STOCK”! - STOCK“

Vbw»  VflP-M  (35)

+ VWPJ’  _ V5190" Q
P “W07 P “Pa? ’
Pflvhw Pfishs»

_where m is the month and yr is the associated yearly

total. Vhnp’; and Vcxm- are the value of imports and

exports, respectively, for either the month or the 

yearly total, Pfishm is the price of fish (the CPI for fish)
for month m, and Qimpy, and Qgx y, is the quantity
of imports and exports, respectively, for the year.
Once the monthly disappearance data are ob-
tained, they become weights for annual consumption
to generate monthly or quarterly consumption. Using
these weights, monthly consumption is calculated as

DIS," 6
Qyr (3 )

01s,,

0,, =

where Q is the quantity of consumption for the month
or the year, and DISy, is the summation of DIS over all
the periods in the year.

This weighting procedure is used to capture the
seasonal variation in consumption only. Consumption
levels are still determined by the annual data published
in FCPE.

This process most notably ignores aquaculture.
Since weights are being generated, this procedure
only causes an error to the extent that aquaculture
production does not follow the same seasonal pattern
of the disappearance data calculated above. The addi-
tion of this variable could add considerable difficulty
due to the lack of data in this area. The benefits of this
additional work at this time is considered to be mini-
mal since only weights are desired. Thus no attempt
is made to include aquacultural production.

ter using this weighting procedure, quarterly
price and consumption data for each of the four meat
groups are obtained from the first quarter of 1968
through the last quarter of 1988. Data for these series
are exhibited in Figure 2 (see Appendix A). Descrip-
tive statistics for these series are exhibited in Table 1.

The average quarterly per capita expenditure on
meatproducts over this period is $88.56. Mean prices
are highest for beef and fish with fish being much
more variable. The poultry price is the lowest and least
variable of the meat groups. Beef enjoys the largest
mean consumption, level while fish has the smallest
level of consumption. The coefficient‘ of variation
(standard deviation divided by the mean times 100) is
also provided in these tables. This statistic is included
as another measure of variation for the data. The

Table 1. Descriptive statistics for meats.

Number Standard Coef. of
Variable Units of Obs. Mean Deviation Minimum Maximum Variation

Pbeef ¢/lb 84 216.64 67.87 102.74 306.03 31.28
Ppork ¢/lb 84 151.67 45.65 76.02 223.93 30.10
Ppoult ¢/lb 84 93.06 22.23 55.74 134.67 23.89
Pfigh t/lb 84 215.39 98.74 78.24 403.57 45.84
Qbeef lb/cap 84 20.91 1.62 17.60 25.10 7.75
Qpork lb/cap 84 15.13 1.33 11.30 17.90 8.79
Qpoult lb/cap 84 14.54 2.86 9.70 21.60 19.67
Qfish lb/cap 84 3.20 0.92 2.03 5.54 28.75
Expend $/caD 84 88.65 27.82 40.66 140.49 31.38
consumption levels of poultry and fish are more vari-
able, as determined by this coefficient, than the con- Plow, = 0.8607 + 0.2099 PSKIM _
sumption of beef and pork. Poultry products have (Q0423) '
shown substantial increases in consumption relative (37)
to other meats since the mid-1970s.

R2 = 0.276

Dairy Products

Dairy products correspond to the following:
whole milk, lowfat milk, cheese, and frozen dairy
products; frozen dairy products consist of ice cream,
ice milk, and sherbet. Annual per capita consumption
data are published in FCPE for the years 1960 through
1988. Whole milk prices are also available from FCPE,
while price indices for cheese products and ice cream
are available from BLS.

The lowfat milk price series is more difficult to
obtain since data are not available before 1978.- Gould,
Cox, and Perali encountered this same difficulty. In
their Work, an auxiliary regression of the skim milk
price on the lowfat milk price Was constructed using
monthly data from 1980 through 1987. They chose to
omit the intercept in this auxiliary regression, since
out-of-sample estimates appeared unrealistically high.
An attempt was made to duplicate these results using
monthly data from January 1981 through May 1986.
Although similar results were obtained, a model in-
cluding an intercept appears to out perform the Gould
et al. model. This regression yields the equation

Std. Error of Est. = 0.009

Results from this regression are used to estimate
lowfat milk prices using the skim milk price series.
Skim milk prices are published by the USDC for peri-
ods beginning with 1964. Before 1964, only one milk
price is published. Thus this one price series is used
for both Whole and lowfat milk. Representative prices
for 1986 are published by the USDA, ERS in Dairy
Situation and Outlook. These prices are $1.11 per
half-gallon for whole milk,$1.08 per half-gallon for
lowfat milk, $2.36 per half-gallon for ice cream, and
$2.60 per pound for cheese. Since quantities are in
pounds, it is necessary to convert half-gallons into
pounds. This conversion is done based on the weight
of 8.6 pounds per gallon. Descriptive statistics are
exhibited in Table 2.

The prices for Whole and lowfat milk are highly
collinear (r=0.998). Thus, in the graph of price data
(Figure 3), they appear as one line. The quantities
consumed, on the other hand, differ greatly, with a
strong decrease in whole milk consumption, offset by
a large increase in lowfat consumption. Cheese pro-

Table 2. Descriptive statistics for dairy products.

Number Standard Coef, of
jvariable Units of Obs. Mean Deviation Minimum Maximum Variation
Pwhok, ¢/lb 29 1 7.41 6.35 10.29 26.91 36.74
Pkmm t/lb 29 17.09 6.37 9. 57 26. 57 37.27
chcm ¢/lb 29 162.93 78.09 68.92 274. 22 47.93
Pfmw, ¢/lb 29 33.67 14.07 19.80 57.90 41.79
Qwholc lb/cap 29 187. 59 51.43 106.10 263.90 27.42
Qknvm lb/cap 29 55.50 29.80 13.20 1 01.60 53.69
chase lb/cap 29 14.83 5. l0 8. 30 24.01 34. 39
5mm, lb/cap 29 26.31 0.80 24.30 27.70 3.04
Exoend $/cao 29 77. 57 35.49 39.04 135.47 45.75

11

ducts show the greatest change in price with a sharp
increase during the 1970s. s

Vegetables

Fresh vegetables include: potatoes, lettuce, toma-
toes, and other fresh vegetables. Consumption data for
each of these individual fresh items as well as total
fresh consumption are published in FCPE. Price data
for potatoes, lettuce, and tomatoes are also taken from
FCPE. The price index for all fresh vegetables, also
from FCPE, is used as the price variable for the other
fresh vegetable category. Separate consumption data
are available for carrots, celery, and onions. Prices for
these goods are available from the USDA in Vegetables
and Specialties: Situation and Outlook Report. Al-
though these items are not directly represented be-
cause of their relatively small budget shares, data for
these three items are used to develop the price series
for other fresh vegetables.

Representative prices for vegetable items are ob-
tained from FCPE. Prices from 1985-1988 are averaged
and divided by the average price index for this same
period. Representative prices for the 1985-1988 pe-
riod are $0.25 per pound for potatoes, $0.58 for
lettuce, $0.81 for tomatoes, and $0.26 for other fresh
vegetables. Descriptive statistics for these vegetables
are given in Table 3.

The consumption of potatoes show a steady de-
cline from 1960 through 1975 but remain steady after
1975 (see Figure 4). Modest growth in the consump
tion of lettuce and stronger growth in the consump

tion of fresh vegetables has occurred since 1975.
Prices for these series move in similar fashions
throughout the sample period. Potatoes have the larg-
est budget share for the first 5 years of this period.
Other fresh vegetables hold the largest budget share
after 1965.

Fruits

Fruits consist of four fresh commodities: oranges,
apples, bananas, and other fresh fruit. Price and con-
sumption data are readily available for bananas, ap-
ples, and oranges in FCPE. The other fresh fruit
category includes all citrus and noncitrus fruit except
oranges, apples, and bananas. Again, as was done for
vegetables, the price for fresh fruit is used as the other
fresh fruit price.

Representative prices from FCPE for 1988 are
$0.73 per pound for apples, $0.53 for oranges, and
$0.42 for bananas. The remaining fresh fruits prices
given in FCPE are averaged to obtain a representative
price of $0.89 per pound. Descriptive statistics for
these variables are shown in Table 4.

As was the case for vegetables, price patterns are
similar for all of fresh fruits (see Figure 5). The con-
sumption of each of these items remains relatively flat
through the mid-1970s followed by slow growth for
all the items except oranges. The similarity of patterns
across items results in little change in the budget
shares over the data period.

Table 3. Descriptive statistics for fresh vegetables.

Number Standard Coef. of

m, ylariable Units of Obs. Mean Deviation Minimum Maximum Variation
l)potat‘xs ¢/lb 29 14.08 7.27 5.66 27.24 51.63
Plwucc ¢/lb 29 28.60 16.46 11.11 68.43 57.55
Pmmmoc, ¢/lb 29 47.44 20.71 22.71 87.97 43.66
Pmhc, vcgcmbks ¢/lb 29 27.96 12.72 12.31 50.61 45.49
Qpotatoc, lb/cap 29 56.03 1 1.47 43.80 79.80 20.47
ma,“ lb/cap 29 22.30 1.73 19.90 25.70 7.76
“nnmoc, lb/cap 29 11.66 1.47 9.60 15.20 12.61
oo,,,c,,.cgc,,b,c_, lb/cap 29 35.61 5.72 27.70 49.90 16.06
ExDend $/caD 29 30.14 16.51 13.48 69.98 54.78

Eble 4. "Descriptive statistics for fresh fruits.
Number Standard Coef. of

A Yariable Units of Obs. Mean Deviation Minimum Maximum Variation
Pomngcs ¢/lb 29 21.42 13.64 9.38 53.00 63.69
Pappks ¢/lb 29 34.94 19.15 14.58 73.00 54.81
bananas t/lb 29 22.27 10.14 12.93 42.00 45.53
Pom“ m“, ¢/lb 29 39.00 22.2 1 17.24 89.00 56.95
Qdnmgcs lb/cap 29 14.38 1.65 1 1.50 18.60 11.47
Qwpk, lb/cap 29 16.67 1.37 14.30 20.40 8.22
bananas lb/cap 29 19.73 2.58 15.80 25.70 13.08
omcflmi, lb/cap 29 31.33 3.11 25.80 38.10 9.93
Expend S/cap 29 26.26 16.44 11.69 64.37 62.60

12

Fats and Oils

The fats and oils category also consists 0f four
items: butter, margarine, shortening, and vegetable
oils. Prices for these items were taken from USDA, Oil
Crops: Situation and Outlook whenever possible.
These prices are in cents per pound so no repre-
sentative prices are needed. In cases of missing data,
indices for butter and margarine are available from
FCPE, while shortening and salad oil prices are pro-
jected from the Fat and Oil CPI in FCPE. Consumption
for each of these items also is given in FCPE. Descrip-
tive statistics are given in Table 5.

Over the data period, the consumption of salad
oils and shortening has increased substantially, While
the consumption of butter has decreased (see Figure
6). Butter prices show a much greater rate of growth
since 1975 than the other items, offsetting its declin-
ing consumption in the determination of budget
shares since that time. In the pre-1975 period, the
budget share for butter dropped substantially, while
shortening and salad oils experienced growth. Since
1975, budget shares have remained relatively stable,
except for margarine which still exhibits a decreasing
budget share.

Aggregate Commodities

Aggregate commodities consist of eight food
groups. Four of these are from previously defined
groups: meats; dairy; fresh vegetables and fruits; and
fats and oils. The remaining four categories include:
processed vegetables and fruits; eggs; cereal products;
and sugar and sweets. To maintain more equitable
budget shares, fresh fruits and fresh vegetables were
combined into one category for this model.

Data for three of the first four groups are obtained
from the sources presented earlier. The consumption
level of each category within a group may be summed
to obtain total consumption. A major food group price
may be obtained by dividing the total expenditure on
all items in the group (the sum of price times quantity
of each item) by the total quantity consumed. This
process results in prices and quantities in the same
units as in the disaggregate categories. This approach
is also consistent with the weak separability assump-

tion that the expenditure on a subgroup is determined
in a previous budgeting stage.

Since quarterly data are used for meats, it is neces-
sary to add the observations from each of the four
quarters to obtain annual data. Additional data for the
years 1960 through 1967, which are not included in
the quarterly data, are obtained from FCPE. Data for
the other categories are obtained from FCPE. No miss-
ing observations were encountered in gathering these
data. Representative prices for the new categories are
needed as well from FCPE. These prices per pound
for 1988 are $1.82 for processed fruits, $0.61 for
processed vegetables, $0.94 for cereal products, $0. 57
for sweets, and $0.57 per dozen for eggs. Egg prices
are converted to a pound basis by dividing by 1.57
pounds per dozen. A weighted-average price for proc-
essed fruits and vegetables was developed from these
respective prices. Descriptive statistics are shown in
Table 6.

Meats are the dominate category in this model
with about 45 percent of the foodbudget (see Figure
7). The remaining budget is split nearly equally among
the other items, with no notable trends. Processed
fruit and vegetables show a definite drop in consump
tion in the late 1960s, and dairy products show steady
losses in consumption throughout the period. Prices
for these groups show slow but steady growth
through the early 1970s with accelerated growth in
later periods.

Nutritional Awareness Data

In addition to the traditional economic data pre-
viously discussed, nontraditional factors also are in-
cluded in the model formulation to measure the effect
of nutritional awareness. Since nutritional awareness
is not directly measurable, proxy variables must be
usedqThree different alternatives are used in this
project. These alternatives as well as data uses have
been discussed in the model development section.
This section only discusses the data used for these
proxies.

The first alternative uses binary variables to divide
time periods. These time periods are: pre-1977, 1977-
1979, 1980-1985, and 1984-1988. The base period
corresponds to the pre-1977 period. Three dummy

Table 5. Descriptive statistics for fat and oils.

Number Standard Coef. of

Variable Units of Obs. Mean Deviation Minimum Maximum Variation
Pbune, ¢/1b 29 130.17 57.41 74.40 217.00 44.10
Pmmaflnc ¢/lb 29 50.75 22.41 26.00 86. 50 441.60
Pshoncmng ¢/lb 29 52.67 24.50 26. 30 92.00 46.52
Qbuue, lb/cap 29 5.28 1.03 4. 20 7. 50 19. 51
Qmaqwmc lb/cap 29 10.71 0.72 9.30 1 2.00 6.72
Qshoncmng lb/cap 29 17.35 2.76 12.60 22.90 15.91
Qsalad o“ lb/cap 29 1 7.47 4.89 9.20 25.00 27.99
Expend S/cap 29 34.76 19.25 14.49 66.56 55.38 

Table 6. Descriptive statistics for agggate commodities.
Number Standard Coef. of
Variable Units of Obs. Mean Deviation Minimum Maximum Variation

Pmms ¢/lb 29 139.48 55.62 72 .86 219.96 39.88
PM“, ¢/lb 29 28.09 14.37 12.73 52.57 51.16
PM“, raw ¢/1b 29 26.90 14.51 11.50 56.78 53.94
Pproducc f&v ¢/lb 29 53.80 26.46 25.40 103.34 49.18
Pm and 0,, ¢/lb 29 64.97 28.44 55.70 108.76 43.77
Pcggs r/lb 29 20.27 4.67 13.85 28.95 23.04

m“, ¢/lb 29 47.75 24.27 22.79 94.00 50.83
Pswms ¢/lb 29 18.93 11.03 7.59 37.00 58.27
Qmcm lb/cap 29 206.69 16.42 173.50 233.28 7.94
Qdmy lb/cap 29 284.23 16.64 257.67 309.70 5.85
Qfrcsh f8“,       
Qpmduccmv lb/cap 29 177.31 23.74 121.17 202.65 13.39
Qfa, and o“ lb/cap 29 50.80 7.07 38.70 62.20 13.92
Q6885 lb/cap 29 36.66 3.55 3 1.00 42.60 9.68
Qccml lb/cap 29 149.79 8.00 139.80 171.80 5.34
Qswws lb/cap 29 129.94 11.0 1 1 14.00 1 52.70 8.47
Ezipfildx $/caD  29 _, _ 670.00 334.81 297.83 1275.70 49.97

variables are created to “measure” structural shifts
which may have occurred in these periods. A new era
in the study 0f nutrition began in 1977, making this
regime a probable period (pre versus post 1977) for
structural change. In the period following 1977, ten
major articles on nutrition and health were published.
The time period following 1977 is divided into three
periods, 1977-1979, 1980-1983, and 1984-1988. The
post-1977 regime is subdivided based on the work of
Cronin and Shaw (Table 7).

This first alternative considers structural change
with little regard for the source of this change. An
alternative approach is to generate a proxy variable for
nutritional information for use in demand models.
Brown and Schrader have developed such an index

based on cholesterol information. This index consists
of the number of articles published in medical journals
which establish a link between cholesterol and heart
disease, less any published articles which fail to find
such a link. Data for this index are published by Brown
and Schrader. Updated data for this index have been
obtained from Brown and Schrader and included in
this study.

The Brown and Schrader index only considers
articles involving cholesterol. However, the nutri-
tional awareness message involves many nutrients in
addition to cholesterol. One could develop such an
index with a broader source of nutrient elements. A
simpler approach is to select a few very important
articles regarding the link between nutrition and

Table 7. Articles and publication dates for articles reviewed by Cronin and Shaw.

Dept. of Health and Human Services

Article Year
Number Title and Organization Published
(1) Dietary Goals for the United States, 2nd ed. U.S. Senate Select Committee on Nutrition and Human 1977
Needs
(2) Health People: Surgeon Generals Report on Health Promotion and Disease Prevention, U.S. 1979
Dept. of Health,Education and Welfare
(3) Concepts of Nutrition and Health”, Council on Scientific Affairs, American Medical Association 1979
(4) Recommended Dietary Allowances, 9th Ed. Committee on Dietary Allowances, Food Nutrition 1980
Board, National Research Council, National Academy of Sciences
(5) Toward Healthful Diets, Food and Nutrition Board, National Research Council, National Academy 1980
of Sciences
(6) Diet, Nutrition, and Cancer, Committee on Diet, Nutrition, and Cancer, National Research 1982
Council, National Academy of Sciences
(7) “Nutrition and Cancer: Cause and Prevention; A Special Report”, American Cancer Society 1984
(8) Nutrition and Your Health: Dietary Guidelines forAmericans, 2nd Ed. U.S. Dept. of Agriculture 1985
and Dept. of Health and Human Services
(9) “Dietary Guidelines for Healthy American Adults”, American Heart Association 1988
(10) “NCI Dietary Guidelines: Rational", National Cancer Institute, National Institute of Health, U.S. 1988

l4

health. Cronin and Shaw review 10 such articles. A
list of these articles as well as their date of publication
is given in Table 7. An index is developed which
increases from 0 to 1O as these articles are released.
Since it is based on the review by Cronin and Shaw, it
is referred to as the Cronin-Shaw (C-S) index. The C-S
index is similar in nature to the Brown and Schrader
(B-S) index but addresses nutritional awareness in a
more global context. This index assumes that each
article carries equal weight as does the B-S index.
However, the C-S index measures nutritional aware-
ness differently than the BS index. Thus the two may
be used simultaneously.

Relevant descriptive statistics are given for these
variables in Table 8. The data forthe B-S Index are not
lagged, unlike the index as published by Brown and
Schrader. Furthermore, the first difference of the B-S
index is used in this analysis. Brown and Schrader use
the accumulation of articles, which can be considered
a stock variable. In this project, a distributed lag frame-
work is used. The framework necessitates the use of
the flow of articles which corresponds to the first
difference of the stock.

model indicates a problem with serial correlation. This
serial correlation problem is corrected by employing
a nonlinear estimation procedure in SHAZAM. Typi-
cally, when a serial correlation correction is con-
ducted in a demand system frame work, the same
value for p is imposed on all the equations. This
imposition maintains the adding-up restriction
(Berndt and Savin). In the dynamic AIDS model as
specified in this study, all N equations are estimated.
Thus unique estimates of p are obtained for each
equation in the system.

The nutritional information leads to mixed results
across scenarios. When used independently, both the
C-S index and B-S index have significant impacts on
the consumption of meats, but this impact is in oppo-
site directions. Yet, when the two indices are com-
bined, neither have significant impacts. The C-S index
also has a significant impact on the consumption of
fresh fruits and vegetables, fats and oils, eggs, and
cereal products. The C-S index indicates an increase
in the consumption of eggs. This finding is contrary to
the findings of Putler and of Brown and Schrader who
found a decrease in egg consumption due to nutri-
tional information.

Table 8. Descriptiove statistics for nutritional awareness information.

Number Standard Coef. of
Variable Units of Obs. Mean Deviation Minimum Maximum Variation
L70s Binary 29 ()_1 ()3 NA O. 1. NA
E805 Binary 29 0.138 NA O. 1. NA
L80s Binary 29 0.172 NA O. 1. NA
C-S Index Articles 29 2.345 3.352 0. 10. 42.94
BS Index Net Number 29 34.966 26.124 O. 107. 74.71
of Article

Empirical Results

Nutritional awareness is measured by considering
five different econometric scenarios. These scenarios
utilize: (1) no information, (2) dummy shifters, (3) a
counter of nutritional information articles (C-S Index),
(4) a counter of cholesterol information articles (B-S
Index) using a Polynomial Distributed Lag (PDL), (5)
and a combination of these two indices. Head and tail
restrictions are imposed whenever the PDL is used.
Tests of these head and tail restrictions do not reject
their use in any instance.

To conserve space, only the results from the best-
case scenario are presented here. However, some
details of other scenarios are discussed. The results
pertaining to these omitted scenarios can be found in
Schmitz.

Empirical Results: Aggregate Model

The aggregate model represents the first stage of
a two-stage budgeting process. As such, results from
this model consider allocations among major food
subgroupings. Unlike the submodels, the aggregate

l5

The B-S Index indicates a positive impact on meat
items and negative impacts on dairy products and
eggs. Because the meats group includes both red and
white meat products, a positive relationship is con-
ceivable. The negative impact on dairy and eggs ap-
pears as expected in light of the high cholesterol
content of these groups. I

The two indices also perform well when used
together. By adding the C-S index to the model which
contained the B-S index, significant relationships for
processed fruits and vegetables, cereal products, and
sweets were evident. Meats, however, no longer show
a significant response to either index. Again, due to
the mixture of red and white meats in the meat cate-
gory, it is not at all surprising that no significant
relationships are found. The nutritional indices in the
fresh fruit and vegetable equation as well as in the fats
and oils equation also show no significance. This result
may indicate that the response to nutritional informa-
tion involves substitutions within the submodel, but
may not affect the consumption of a particular group

as a whole. The signs for the cereal and the sweet
groups are opposite 0f what is expected.

The scenario involving the B-S index used alone
appears to perform better than the other scenarios.
Results from this model then are presented in Table
A1. The compensated price, uncompensated own-
price, and income elasticity estimates for this scenario
are presented in Table 9, along with some results from
other studies.

Blanciforti, Green, and King consider four aggre-
gate commodities: meats; fruits and vegetables; cere-
als; and other foods. However, the fruit and vegetable
category used by Blanciforti, et al. includes both fresh
and processed items, and thus, is not directly compa-
rable. Price elasticities for meats obtained i11 this study
are slightly lower than those obtained by Blanciforti,
while the price elasticities for cereal products are
considerably lower. Income elasticity estimates are
much higher in this study than those obtained by
Blanciforti. The elasticities for eggs are comparable to
those obtained by Huang. Price elasticities for these
two studies are similar; the income elasticity for eggs
from this study is 0.7, while Huang obtained an elas-
ticity estimate of -0.03.

Empirical Results: Submodels

Meats

The meats model is the first of five second-stage
or submodels to be discussed. It differs from the other
models in this study in that it utilizes quarterly data

over the period from 1968 through 1988. Since quar-
terly data are used, binary variables are included to
allow for seasonality. F-tests on these shifters indicate
that significant seasonal patterns are exhibited by
these data.

The results for this model are affected by the
inclusion of nutritional information, especially when
either of the indices are included. E-tests are con-
ducted to determine the significance of these vari-
ables. The nutritional information in each scenario
used here has significant impact on the meats model.
In the scenario that combines the two indices,» both of
these indices are significant. Furthermore, their im-
pact is significant in each equation. In the case of the
dummy shifters, the shifter for the late 1970s is not
significant.

The scenarios that use the indices indicate a move-
ment away from beef-to the other three meats, most
notably to poultry products. When these two indices
are combined, they give mixed results. In the com-
bined scenario, the B-S index indicates movement
away from beef and pork, while the C-S index indi-
cates movement away from beef and fish. However,
the negative coefficients associated with pork and fish
in these cases are not significant. The coefficients
associated with the lagged budget shares decrease in
magnitude from the 0.5 to 0.7 range in the base
scenario (without nutritional information) to the 0.2
to 0.3 range in the combined indices scenario.

Tableg9iémSielected elasticity estimates for the Aggregate Model.

511' Fresh Proc. Fats 8:
lgomp. El. Meats Dairy F&V F&V Eggs Oil Cereals Sweets
M61115 0.151* 0048* 0.005 0.079* 0.010* 0.007 0.020 0017*
Dairy 0182* -0.274* 0104* 0.124* 0.002 0.054 0062 0078*
Frfish F~ & V. 0.024 -0.149* 0197 0.143 0.024 0.106 0155 0251*
Proccsscd 0.248* 0103* 0.083 0871* 0.027 0.023 0.275 0.114
F. & v. ~
E888 0324* 0.015 -0.152 0.294 0143" -0.036 0386 0.084
Fats & Oils 0.060 0.127 0.174 0.064 0009 0166 0148 0102
Cereals 0.083 -0.068 0119 0.366 0.047 -0.070 0143 0002
Swws 0.213 0.262* 0586* 0.460 0.031 -0.145 -0.007 0975"‘
Uncvmp- 0590* 0373* 0271 -1.039* 0155 -0.225 0253 4015"‘
Own-Prc. ~
Income 0.742 0.633 0.673 0.8903 0.700 0.889 0.782 0.856
Huang
Own-Prc N/A N/A N/A N/A 015 N/A N/A N/A
Income N/A N/A N/A N/A .()_()3 N/A N/A N/A
Blanciforti
et.al.
Own-PK 0.68 N/A 066 N/A N/A N/A 042 N/A
In¢<>m¢ 0.48 N/A 0.47 N/A N/A N/A 0.16 N/A

‘Indicates that elasticity estimate is significantly different from zero (not computed for income elasticities).

N‘ T-value could not be computed.

16

The results obtained from the combined indices the dairy model when considered independently.

Fecase appear to perform better than results from other When combined, each index remains statistically sig-
scenarios. Results from the combined indices scenario nificant, and they jointly show a significant impact on
are presented in Table A2, while the elasticities are each equation.

i‘ exhibited in Table 10. Also included in this elasticity As in the theats model, the Combined ththees
tabie ate estimates item Huang as Weii as CaPPs and scenario offers the best results, overall. Coefficient
sehmitz- The estimates hem Capps anti sehmitz uittet estimates for this model are given in Table A3, while
item Huang in that Capps and sehmitz ineiutie nutti‘ price and income elasticities are included in Table 11.
tienai intetmatien- The most notable impact of nutritional information on

The addition of the nutritional information to this the elasticity estimates involves the own-price elastic-
study results in little change in the own-price elasticity ity for lowfat milk. This estimate increases from -0.78
estimate for beef and lower own-price estimates for to-1.00 as aresult of including nutritional information.
the other meats when compared with results obtained The own-price elasticity of whole milk also is in-
without nutritional information. Own-price elasticity creased, while cheese and frozen products show a
estimates for poultry and pork appear to be in the decrease in own-price elasticity‘ as a result of this
general range as those obtained from other studies, information. Results from the dairy model are com-
while the own-price elasticities for beef and fish are pared to two studies: Huang; and George and King.
slightly higher than those obtained in other studies. Both of the comparison studies have whole and lowfat
Income elasticities obtained for these meats are higher milk combined as one estimate. The elasticity esti-
than those found by Huang. In general, all of the mates obtained here (near 0.75) are considerably
systems estimated here have larger income elasticity larger than estimates obtained by either of these stud-
estimates than estimates found in other studies. ies (near 0.30). This result is not unexpected since the

_ - se aration of these two items ields more disa e ate
Dairy Products colinmodities. Own-price etiasticity estimggers gfor

The dairy 1110061 P¢ff0fII1$ Well ill eflCh 0f the five cheese products obtained here are similar to those in
SCUIQfiOS. Serial COIIQIHIiOII (1068 110K appear I0 be a the comparison studies, while the estimate for frozen
problem, and R-square values remain high across sce- pi-Qduets is similar to the estimate by George and King
IIKIiOS- but considerably different from the estimate by

Q With one C-S index, a significant shift away from Huang-

consumption of frozen items to consumption of
cheese products is evident. Although not significant,
a decrease in the consumption of whole milk and an
increase in the consumption of lowfat milk also is

Vegetables

The vegetable model considers the consumption
of potatoes, lettuce, tomatoes, and other vegetables,

associated with the C-S index. The B-S index reveals a eaeh in ttesh t°tm~ The base metiei has a iowet 800d‘

statistically significant shift from whole milk to lowfat nessehnt than in the aggregate, meats, and tiaiTY med‘

milk and cheese products. This shift is consistent with eis eiiseusseti PYeViOuSiY-

the expected results. In this model, the dummy variables, as a proxy for
F-tests are performed for the dairy model as they nuttitien» appeal‘ t0 (i0 mueh bettet than in Perv i0t15

Wete fet- the meats medet The dummthvatiahle see submodels. These coefficients associated with the

nario shows no significant impacts on the system. The dummy Vatiabies ate signitieant in neatiY evetY ease-
C-S and B-S indices perform well, both independently The iate 1970s ‘and iate i980s shew a signitieant

and jointly. Both indices have significant impacts on deviation from the Piei 977 Petieti- Ftests tutthet
Table 10. Selected elasticity estimates for the Meats Model.
Compensated elasticity Beef Fish Poultry Pork
Beef ' -0.3677‘ 0.0348 0.0652‘ 0.2676‘
Fish 0.2377 0.3185 0.0108 0.0700
Poultry 0.2148‘ 0.0052 0.3229‘ 0. 1028‘
'P0rk 0.5267‘ 0.0202 0.0614 0.6083‘
“ UnComp. Own-Price 0.8949‘ -0.4039 0.4900‘ 0.8287‘
Income 0.6934 0.7659 0.7241 0.5704
Huang
Own-Price a 0.62 N/A 0.53 0.73
Income e 0.45 N/A 0.36 0.44
Capps & Schmitz
 Own-Price 0.73 0.32 0.27 0.98
Income N/A N/A N/A N/A
‘Indicataes that elasticity estimate is significantly different from zero (not computed for income elasticities).

17

r}

_'I‘Kable 1 1. Selected elasticity estimates for the Dairy Model.

Compensated elasticity Whole Lowfat Cheese Frozen
Whole -0.3750‘ 0.1740 0.0796 0.1214
Lowfat 0.6468 08661‘ 0.4035‘ -0. 1842
Cheese 0.1160 0.1581‘ 0.3125‘ 0.0385
Frozen 0.4676 -0.1909 0.1017 ~0. 3784‘
UnComp. Own-Price -0.7553‘ -1.0050‘ -0.6665‘ -0. 5053'
Income 0.5587 0.7584 0.7574 0.7181
Huang

()wn-Price -0.26 -0.26 -0.33 -0. 12
Income -0.22 -0.22 0.59 0.01
George 8r King
()wn-Price -0.35 -0.35 -0.46 -0. 53
income 0.20 0.20 0.25 0.33

‘Indicates that elasticity estimate is significantly different from zero (not computed for income elasticities).

indicate that the binary variables have a significant
explanatory role as a group on the model as well as on
each equation in the model. These variables indicate
a movement away from other vegetables and toward
potatoes and tomatoes.

The C-S index performs in a similar fashion to the
dummy shifters. In this case, a significant shift from
lettuce to potatoes is indicated as well as a shift, albeit
insignificant, from other fresh vegetables to tomatoes.
The presence of the C-S index left the lagged budget
share relatively unaffected from the base scenario. The
serial correlation problem indicated in the base case
is eliminated in this scenario.The B-S index does not
perform as well in this model.

The B-S index is based on cholesterol information.
Vegetables, as well as fruits which are discussed next,
contain no cholesterol. Thus it is not surprising that
this index performs poorly in these two models. How-
ever, it is included since some may consider this index
as a proxy for a more global measure of nutritional
awareness.

The impact of the B-S index is opposite of the
dummy variable and C-S index cases. This index indi-
cates a significant shift from tomatoes to the fresh

vegetables. The B-S index does not perform any better
when combined with the C-S index.

The C-S index yields better results than other
scenarios in this model. Coefficient estimates for this
scenario are given in Table A4. As has been the case
in the earlier models, the lagged coefficients are
smaller in magnitude in cases where nutritional infor-
mation is included relative to the base case. This result
indicates that omission of this information may lead to
an over-estimation of habit persistence. The elasticity
estimates for this scenario are presented in Table 12.

The own-price elasticity of potatoes when nutri-
tional information is included is half of the estimated
level when no such information is included. A de-
crease in own-price elasticity is evident for lettuce,
while for tomatoes and other fresh vegetables, own-
price elasticities are larger with the presence of nutri-
tional information. Huang and George and King
provided elasticity estimates for similar items. Esti-
mates for potatoes and other fresh vegetables in the
base-case are similar in magnitude to those obtained
in these other two studies. The elasticity of tomatoes
is comparable between this study and the study of
George and King. Income elasticity estimates and the

Gable 12. Selected elasticity estimates for the Vegetable Model.

W Compensated elasticity Potatoes Lettuce Tomatoes Other Vegetables
Potatoes -0.0880 0.0130 0.1006‘ -0.0255
Lettuce 0.0163 -0.01 52 —0.0062 0.0052
Tomatoes 0.1380‘ -0.0068 -0. 1639 0.0328
Other Vegetables -0.0197 0.0032 0.0 184 -0.0020
UnComp. Own-Price 0.1934 02937‘ 03520‘ 04302‘
Income 0.3228 1.0675 0.7909 1.01 31
Huang ‘

Own-Price -0.37 -0.14 -0.56 -0.21

Income 0.16 0.23 0.46 0.28
George 8: King

Own-Price -0.31 -0.14 -0.56 -0.21

Income 0.12 0.15 0.17 0.15
‘Indicates that elasticity estimate is significantly different from zero (not computed for income elasticities).

K

own-elasticity estimate for lettuce also are much
higher in this study than in either of the reference
studies.

Fruits

The fruit model consists of fresh consumption of
oranges, apples, bananas, and other fresh fruits. As in
the vegetable model, goodness-of-fits are low in this
model, relative to the previous models. Serial correla-
tion does not appear to be a problem. The lagged
budget shares are also smaller in this model, ranging
from 0.22 to 0.37. 9

The dummy-variable scenario shows no signifi-
cance, either on an individual coefficient basis or on
the basis of F-tests. The B-S index, as found in the
vegetable model, also shows no significant impact on
the consumption of fresh fruits, either alone or when
combined with the C-S index. The C-S index is signifi-
cantly positive for oranges and maintains this sigr1ifi-
cant sign when combined with the B-S index. As seen
in the earlier model, the coefficients associated with
the lagged budget share diminish in magnitude when
nutritional information is included.

Although the C-S index does not perform as well
as in the previous systems, it performs better than the
other nutritional measures in the fruit system. Coeffi-
cient estimates for this model are given in Table A5

'\while the resulting elasticities are presented in Table

13. Oranges are the most affected by the presence of
this information. The own-price elasticity increases
from -0.90 to -1.06 as a result of this information, while
the income elasticity decreases from 0.75 to 0.53.
Again, these estimates can be compared to estimates
obtained from Huang and George and King. Price
elasticity estimates for oranges are close to the esti-
mate by Huang, while bananas more nearly match the
estimate obtained by George and King. Own-price
elasticities for other fresh fruits are much higher than
those obtained in these two comparison studies. Ap-
ples provide an interesting case where the estimate
obtained here (-0.40) is much higher than the -0.2

obtained by Huang but lower than the-0.72 obtained
by George and King.

Fats and Oils

The fats and oils model is the final submodel
analyzed. This model contains four items: butter; mar-
garine; shortening; and salad oils.

The dummy variable case appears to do quite well
in the fats model. These shifters show an initial surge
in the budget share for butter but decreased share in
the final two time periods. The other three items show
gained shares with most of these coefficients statisti-
cally significant. The late 1970s and the early 1980s
show a significant deviation in shares from the pre-
1977 period. The late 1980s, however, are not signifi-
cantly different than the pre-1977 era. These tests also
show that the impact of these shifters is significant on
each equation except salad oils.

The BS index performed better in this model than
in either the fruits or the vegetables model. This index
shows a significant shift from butter to shortening as
well as shifts, albeit insignificant, away from margarine
and toward salad oils. When combined with the C-S
index, this pattern is unchanged. However, signifi-
cant shifts are indicated away from margarine and
toward shortening. The C-S index indicates the shift
from butter to salad oils that was detected in the B-S
index scenario.

As seen in each of the submodels, the lagged
budget share coefficients once again have smaller
magnitudes when the nutritional information is added
to the system. Thus again, we should recognize the
potential for over-stating the effect of habit persist-
ence if nutritional information were not included in
the model.

In this model, the dummy variables perform better
than in the other scenarios considered. Coefficient
estimates for this scenario are presented in Table A6
while selected elasticities are presented in Table 14.
The inclusion of nutritional information results in little
change in the own-price and income elasticity esti-

Table 13. Selected elasticity estimates for the Fruit Model.
Compensated elasticity Oranges Apples Bananas Other Fruit
Oranges -0.9927‘ 0.0668 0.0589 0.8671‘
Apples 0.0341 -0.1565 -0.0143 0.1367
Bananas 0.0382 -0.0182 -0.4364‘ 0.4164‘
a Other Fruit 0.2131‘ 0.0658 0.1576‘ 04365‘
UnComp. Own-Price -1.0647‘ 0.4000‘ 06223‘ 0.9351‘
Income 0.5335 0.9218 0.8948 0.9082
Huang
Own-Price -1.00 -0.20 -0.40 -0. 24
Income 0.49 -0.35 -0.04 035
I~\George & King
Own-Price -0.66 -0.72 -0.61 -0.60
Income 0.26 0.14 0.14 0.40
‘Indicates that elasticity estimate is significantly different from zero (not computed for income eleaticities).

Egfable 14. Selected elasticity estimates for the Fats and Oils Model.

Compensated elasticity Butter Margarine Shortening Salad Oil
Butter 0.2108 0.0213 0.0282 0.1612
Margarine 0.0282 0.4762‘ 0.163 1 0.2848
Shortening 0.0227 0.0994 0.061 1 -0.0609
Salad Oil 0.1036 0.1382 0.0485 -0.1933
UnComp. Own-Price 0.3801‘ 0.601 1‘ 0.3756‘ 0.5847‘
Income 0.6456 0.6307 0.9669 0.9589
Huang

Own-Price 0.17 -0.27 N/A N/A

Income 0.02 0.1 1 N/A N/A
George 8: King

Own-Price 0.65 0.85 -1.02 0.69

Income 0.32 0.00 0.03 0.28

‘Indicates that elasticity estimate is significantly different from zero (not computed for income elasticities).

mates, except for butter, Where the own-price elastic-
ity decreased from 0.57 to 0.38 as a result of the
nutritional information.

Again, these results are compared to the works of
Huang and George and King. Own-price elasticities
obtained in this study are comparable to those ob
tained by George and King, except for the case of
shortening. George and King found the own-price
elasticity of shortening to be -1.02 while this study
estimates it at 0.38. Again, this study finds the income
elasticity to be much higher than estimates by either
Huang or George and King.

Elasticity Estimates for Nutritional Measures

Elasticity estimates can be obtained for nutritional
information as well as for prices and income. Formulas
for these calculations are given in the model develop-
ment section. Elasticity estimates for the best-case
scenarios of each model are given in Table 15. These
estimates are only calculated for the indices used in
the best-case scenario. The best-case scenario for fats
and oils is the dummy-shifter case, thus no elasticity
estimates are available. When the PDL is used, both
short-run and long-run elasticities are given.

Among the aggregate commodity groupings, the
response to nutritional information, in the long-run is
highest for eggs with an estimate of 0.16. The re-
sponse to nutritional information for cereal and baking
products and sugars and sweets is nearly as high as
eggs with estimates of 0.14 and 0.15, respectively.
The response for the two fruit and vegetable groups
are lowest with estimates of 0.01 for both the fresh
and the processed items.

In the submodel analysis, meat items have the
highest responses to nutritional information. This re-
sponse is greatest for fish where a 1 percent increase
in cholesterol information results in a 6 percent in-
crease in fish consumption in the long-run, ceteris
paribus. The response for poultry and beef also are
elastic where a similar 1 percent increase results in a
2 percent increase in poultry consumption and a 1.2

2O

percent decrease in beef consumption, ceteris pari-
bus.

These elasticity estimates indicate a substitution
of lowfat milk for whole milk as a result of cholesterol
information. A 1 percent increase in cholesterol infor-
mation results in a 0.11 percent decrease in whole
milk consumption and a 0.18 percent increase in
lowfat milk consumption, ceteris paribus, in the long-
run.

Subgroups for fruits and for vegetables use only‘

the C-S index. Lettuce shows a 0.13 percent decrease
in consumption resulting from a 1 percent increase in
nutritional information. Meanwhile oranges and pota-
toes show increases of 0.11 and 0.10 percent, respec-
tively, in response to a 1 percent increase in nutritional
information.

Concluding Comments

The groups analyzed in this study include both
aggregate and disaggregate items as well as groupings
of foods derived strictly from animal sources, plant
sources, and both sources. Demand studies often ad-
dress one of these particular groups, but seldom dis-
cuss multiple groups within one study. But to more
fully understand the role of nutritional information,
this information needs to be applied to a variety of
different food groups.

Since these groups are diverse, it would seem
unlikely that one particular nutritional measure would
always work best. Each of the four measures of

nutritional information performed well in certain‘

models, while none performed particularly well in
every case. One would anticipate that different groups
would respond to nutritional messages in different
ways. Thus, one should anticipate that different meas-
ures would perform better in some cases and worse
in others.

These final pages offer a few conclusions based 01$

the results of this study. After discussing the conclu-
sions, limitations and areas for further research are
considered, and final comments made.

In studies which address the issue of structural
change within food groups, nutritional information is
often cited as a possible cause among a host of choices.
This study uses alternative measures of nutritional
information to explain these structural changes. At
least one measure of nutritional information is signifi-
cant in every model, holding all other factors constant.
The significance of these variables certainly supports
the hypothesis that nutritional information may have
led to structural changes in food demand.

Failure to include nutritional information may lead
to serial correlation problems in demand models. Se-
rial correlation is often considered as an indicator that
relevant variables are omitted from the model. Serial
correlation is indicated in the base scenario of several
models. In each of the submodels studied, at least one
of the measures of nutritional information was able to
circumvent serialcorrelation problem.

Failure to include nutritional information also may

'\ result in overstating the importance of habit persist-

ence. The coefficients associated with lagged budget
shares are interpreted as measures of habit persist-
ence. Throughout this study, it is apparent that the
inclusion of nutritional information reduced the mag-

21

‘Kgble 15. Nutritional elasticities for the Best-Case Scenario from each model.
‘ as Index B-S Index
Food Group Short-run Long-run C-S Index
MAIN
Mcats 0.01 5 0.075 N/A
Dairy -0.014 0.06s N/A
Fresh Fruits and Vegetables 0.002 0.01 1 N/A
Produce Fruits and Vegetables 0.003 0.013 N/A
Fats and Oil 0.01 1 0.055 N/A
Eggs 0.032 0.162 N/A
Cereal and Baking Products 0.027 0.135 N/A
Sugar and Sweets 0.029 0.147 N/A
MEATS
Beef 0.097 -1.169 0.010
Fish 0.498 5.974 0.000
Poultry 0.160 1.920 0.002
Pork 0.037 0.448 0.002
DAIRY
Whole Milk 0.021 0.105 0.001
Lowfat Milk 0.035 0.176 0.003
Cheese 0.013 0.066 0.023
Frozen Dairy Products 0.006 0.031 0.044
VEGETABLES
Potatoes N/A N/A 0- 102
Lettuce N/A N/A ‘0- 126
Tomatoes N/A N/A 0-041
Other N/A N/A -0.027
FRUITS
Oranges N/A N/A 0.1 1 4
Apples N/A N/A -0.04 1
Bananas N/A N/A 0.000
Other N/A N/A 0.008
FATS AND OILS
Butter N/A N/A N/A
Margarine N/A N/A N/A
Shortening N/A N/A N/A
Salad Oil N/A N/A N/A
Conclusions nitude of these coefficients. Thus failure to include

this information in a dynamic setting overstates the
degree of habit persistence. Heien and Durham con-
sidered habit formation, using cross-sectional data,
and concluded that time series studies overstate the
role of habits. Based on the results contained within
this study, failure to consider nutritional factors may
account for such an error.

Nutritional information appears to have a small
impact on the elasticity estimates of the traditional
demand variables, specifically prices and income,
when used in a dynamic setting. The lagged budget
shares are affected more than the traditional variables
by the presence of nutritional information.

Nutritional awareness variables are significant in
both the aggregate model and the submodels. Because
of intergroup substitutions, the full impact of nutri-
tional information cannot be captured by the use of
these submodels alone, since submodels cannot cap-
ture the substitution among the major categories. For
example, to measure the full effect of nutritional infor-
mation on the consumption of beef, one must not only
consider substitutions among meats, but also the
change in consumption of meats vis-a-vis other items
in the overall food budget. The extent of error that

results from the failure to capture intergroup substitu-
tions depends on the amount of intergroup substitu-
tion.

One must be concerned with how nutritional
information is measured. In this study, intercept shift-
ers, an index of major nutrition articles, and an index
of cholesterol information are used to measure nutri-
tional information. The variables used here provide
alternative proxy variables for variables which are
difficult to capture. Numerous alternatives are avail-
able. However, these alternatives may perform better
or worse depending on how they are developed and
applied as well as the groups being estimated. Thus,
there is no best measure of nutritional awareness for
all cases. In studies of this type, alternative measures
should be tried and critically evaluated.

Areas for Further Research

In this study, time-series data are used to deter-
mine the impact of nutritional information on food
demand. Although helpful in determining the dynamic
adjustment at a global or macro-level, it does not yield
insight into household or micro-level activities. Nu-
merous questions in this area are still left unanswered.
Some of the most notable questions may be: (1) do the
impacts of nutritional information differ geographi-
cally? (2) do two-income families respond differently
than single-income households? (3) do white-collar
families behave differently in response to nutritional
information than blue-collar families? and (4) what
impact does age of the household head and the num-
ber of children have on food demand response to
nutritional awareness? Questions such as these re-
quire the use of cross-sectional data. The recently
released 1987-88 Nationwide Food Consumption Sur-
vey provides data that may be used to address these
issues.

Consumers receive information about nutrition
and health factors from many different sources. Some
of these sources include medical and nutrition profes-
sionals, radio, television, and printed news, advertis-
ing, food labels, and in-store promotional campaigns.
The effectiveness of these alternative sources needs to
be evaluated. Such an evaluation could have important
impacts, especially in the areas of advertising and food
labeling. This evaluation could also be used to better
direct programs which are designed to educate the
public about issues in nutrition and health.

Limitations

The development of adequate measures of aware-
ness or information is always difficult. Additional work
is needed to establish better measures of nutritional
awareness. The index developed by Brown and
Schrader is a good idea but is really only appropriate
for cholesterol. Measures which are more global in
nature are needed. The C-S index is developed as an
alternative in this study. This index is a simple meas-
urement. It does, however, perform favorably when

22

used alone as well as when used in unison with the B-Sfl V

index.

Other indices could be developed in relation to
numerous specific nutrients, for example fiber, satu-

rated/unsaturated fats, sodium, and sugar. The devel- w

opment of such measures could improve models, such
as the fruits, vegetables, and fats and oils, which
respond to nutrients other than cholesterol. Further-
more, such measures would allow a disentangling of
effects associated with these different nutrients,
thereby allowing a better understanding of how indi-
vidual products respond to new information.

A final limitation involves the distinction between
commodities and products. The time-series data used
in this study results in a commodity approach to the
study of demand. As a result, little can be said about
the behavior of individual products, the role of adver-
tising and promotion, or potential impacts resulting
from changes in product form. Such issues may play
important roles in the overall impact of nutritional
awareness on the demand for food. These issues can
be explored but only at a product level. Scanner data
may be employed in such studies to obtain the degree
of disaggregation necessary to look at product level
demands.

Final Comments
The Preventive Nutrition Era as labeled by Erdman

has had notable impacts on how the consumer con- ,

siders what food items to purchase. Thus, the role of
nutritional awareness in the determination of food
demand cannot be ignored. In the time period of the
1970s and 1980s, consumer awareness has changed
drastically. Such awareness is expected to continue in
the future, and therefore, it is important to continue
to study this dynamic adjustment process.

Literature Cited

Almon, S., “The Distributed Lag Between Capital Ap-
propriations and Expenditures,” Econometrica,
33(1965):7&96.

American Cancer Society, Nutrition and Cancer:
Causes and Prevention, Amer. Cancer Society
Special Report, CA 1984341216.

American Heart Association, Nutrition Committee,
“Dietary Guidelines for Healthy American Adults:
A statement for Physicians and Health Profession-
als,” Circulation, 77(1988):721A-4A.

American Medical Association, Council on Scientific

Affairs, Concepts of Nutrition and Health, journal ‘t’

of the American Medical
242(1979):2535-8.

Anderson, G. and R. Blundell, “Testing Restrictions in
a Flexible Dynamic Demand System: An Applica-
tion to Consumers’ Expenditure in Canada,” Re-
view 0f Economic Studies, 1985, p. 597-410.

Basmann, R., “A Theory of Demand with Variable
Consumer Preferences,” Econometrica, 24
(1956):47-58.

Association,

'\ Baye, M. R., D.W. Jansen, and J.W. Lee, Advertising
Effects in Complete Demand Systems, Working
paper, Dept of Econ., Texas A & M Univ., Jan.

.\ 1990.

Berndt, E.R., and N.E. Savin, “Estimation and Hypoth-
esis Testing in Singular Equation Systems With
Autoregressive Disturbances,” Econometrica,
45( 19751957-57.

Blanciforti, L.A. and R.D. Green, “An Almost Ideal
Demand System Incorporating Habits: An Analysis
of Expenditures on Food and Aggregate Commod-
ity Groups,” Review of Economics and Statistics,
1983, p. 511-515.

Blanciforti, L.A., R.D. Green, and G.A. King, US Con-
sumer Behavior Over the Postwar Period: An
Almost Ideal Demand System Analysis, Giannini
Foundation Monograph No. 40, Aug. 1986.

Borra, S.T., “A Healthy Diet with Animal Product Op
tions: What the Food Marketer and Consumer are
Doing,” Food Marketing Institute, Washington
D.C., November 1988.

Brown, D.J. and L.F. Schrader, “Cholesterol Informa-
tion and Shell Egg Consumption,” American jour-
nal of Agricultural Economics, 72(1990):
548-555.

Butrum, R.R., C.K. Clifford, E. Lanza, “NCI Dietary
Guidelines: Rationale,” American journal of
Clinical Nutrition, 1988;48(Suppl) 888-95.

Capps, O. Jr., and J.D. Schmitz, “A Recognition of
Health and Nutrition Factors in Food Demand
Analysis,” forthcoming in Western journal of Ag-
ricultural Economz'cs, July 1991.

Capps, O. Jr. , J.R. Tedford, andJ. Havlicek, Jr., “House-
hold Demand for Convenience and Nonconven-
ience Foods,” American journal of Agricultural
Economics, 67(1985):862-69.

Chalfant, J.A., “A Globally Flexible, Almost Ideal De-
mand System,” journal of Business and Econom-
ics Statistics, 5(1987):233-247.

Chang, H.S. and H. Kinnucan, Advertising, Informa-
tion, and Product Quality: The Case of Butter,
Working paper, Auburn University.

Chavas, J.-P., “Structural Change in the Demand for
Meat,” American journal of Agricultural
Economics, 65(1985):148-153.

Chen, D., R. Courtney, and A. Schmitz, “A Polynomial
Lag Formulation of Milk Production Response,”
American journal of Agricultural Economics,
52(1972):77-83. A

Choi, S. and K. Sosin, “Testing for Structural Change:
The Demand for Meat,” American journal of
Agricultural Economics, 72(1990):227-236.

“Cronin, F.J. and A.M. Shaw, “Summary of Dietary

Recommendations for Healthy Americans,” Nutri-
tion Today, Nov/Dec 1988:2633.

Dahlgran, R.A., “Complete Flexibility Systems and the
Stationarity of U.S. Meat Demands,” Western jour-

H

nal of Agricultural Economics, 12(l987): 152-
163.

Deaton, A.S., and J. Muellbauer, Economics and Con-
sumer Behavior, Cambridge Univ. Press, 1980a.

, “An Almost Ideal Demand System,” Ameri-
can Economic Review, 70(1 980b):3 12-3 26.

Dorfman, J.H., C.L. Kling, and R.J. Sexton, “Confi-
dence Intervals for Elasticities and Flexibilities:
Reevaluating the Ratios of Normals Case,” Ameri-
can journal of Agricultural Economics,
72(1990):1002-17.

Eales, J.S., and L.J. Unnevehr, “Demand for Beef and
Chicken Products: Separability and Structural
Change,” American journal of Agricultural
Economics, 70(1988):52l-32.

Erdman, J.W. Jr., “Nutrition: Past, Present, and Fu-
ture,” Food Technology, September 1989:220-
227.

Food Marketing Institute, Supermarkets Nutrition
and Health Programs and Information, Wash-
ington D.C. Food Market Institute, 1985.

George, P.S., and G.A. King, Consumer Demand for
Food Commodities in the United States with
Projections for 1980, Giannini Foundation
Monograph No. 26, March 1971.

Goodwin, B.K., Multivariate Gradual Switching Sys-
tems and the Stability of U.S. Meat Demands." A
Bayesian Analysis, Kansas Agricultural Experi-
ment Station No. 90-2, July 1989.

Gould, B.W., T.L. Cox, and F. Perali, “The Demand for
Fluid Milk Products in the U.S.: A Demand Systems
Approach,” Western journal of Agricultural
Economics, 15(1990):1-12.

Green, R. and J.M. Alston, “Elasticities in AIDS Mod-
els,” American journal of Agricultural Econom-
ics, 72(1990):442-445.

Heien, D., and C. Durham, “A Test of the Habit Forma-
tion Hypothesis Using Household Data,” Review
of Economics and Statistics, 75(1991): 189-199.

Huang, K.S, US. Demand for Food: A Complete Sys-
‘tem of Price and Income Elasticities, U.S. Dept of
Agriculture, Economic Research Service, Techni-
cal Bulletin No. 1714, 1985.

Ippolito, P.M. and A.D. Mathois, “Information, Adver-
tising, and Health Choices: A Study of the Cereal
Market,” Federal Trade Commission, April 1990.

Johnson, S.R., R.D. Green, Z.A. Hassan, and A.N. Safy-
urtlu, “Market Demand Functions,” in Food De-
mand Analysis: Implications for Future
Consumption, eds. Oral Capps, Jr. and Benjamin
Senauer, Dept. of Agricultural Economics, Virginia
Poly. Institute and State Univ., 1986.

Kinnucan, H. and O.D. Forker, “Seasonality in the
Consumer Response to Milk Advertising with Im-
plications for Milk Promotion Policy,” American
journal of Agricultural Economics,
68(1986):562-571.

Kn1cnta,]., Elements of Econometrics, MIT Press, 2nd
Ed., 1986.

Manser, M ., “Elasticities 0f Demand for Food: An An-
alysis Using Non-Additive Utility Functions Allow-
ing for Habit Formation,” Southern Economics
journal, 42(1976):879-91.

Moschini, G., and K. Meilke, “Modeling the Pattern of
Structural Change in U.S. Meat Demand,” Ameri-
can journal 0f Agricultural Economics,
71(1989):253~261.

National Academy of Sciences, National Research
Council, Food And Nutrition Board, Committee
on Dietary Allowances, Recommended Dietary
Allowances, 9th Ed., Washington D.C., National
Academy Press, 1980a.

National Academy of Sciences, National Research
Council, Food And Nutrition Board, Toward
Healthful Diets, Washington D.C., National Acad-
emy Press, 1980b.

National Academy of Sciences, National Research
Council, Committee on Diet, Nutrition, and Can-
cer, Diet, Nutrition, and Cancer, Washington
D.C., National Academy Press, 1982.

Nyankori, J.C.O., and G.H. Miller, “Some Evidence and
Implications of Structural Change in Retail De-
mand for Meats,” Southern journal of Agricul-
tural Economics, 14(1982):65-70.

Padberg, D.l., “Non-Use Benefits of Mandatory Con-
sumer Information Programs,” journal of Con-
sumer Policy, Jan. 1977.

Pindyck, R., and D. Rubinfeld, Econometric Models
and Economic Forecasts, McGraw-Hill Co., New
York, 2nd Ed., 1981.

Putlcr, D.S., The Effect of Health Information on Shell
Egg Consumption, Dept. of Agriculture and Re-
source Economics, Working Paper No. 448, Cali-
fornia Agricultural Experiment Station, Aug. 1987.

Schmitz, ].D., The Impact of Nutritional Awareness
on Food Demand: A Complete Systems Ap-
proach, PhD. Dissertation, Dept. of Agricultural
Economics, Texas A&M Univ., 1991.

Schmitz, ].D., and R.M. Nayga, ]r., “Food Nutritional
Quality: A Pilot Study on Consumer Awareness,”
journal of Food Distribution Research, June
1991:1935.

Thurman, W.N., “The Poultry Market: Demand Stabil-
ity and Industrial Structure,” American journal of
Agricultural Economics, 69(1987):30-37.

U .S. Dept. of Agriculture, Dairy Situation and Out-
look, USDA,ERS, Washington D.C., various issues.

24

U.S. Dept. of Agriculture, Food Consumption, Prices,
and Expenditure, 1967-88, USDAJERS, Statistical
Bulletin No. 804, Washington D.C., various issues.

U.S. Dept. of Agriculture, Livestock and Poultry: Situ-
ation and Outlook Report, Commodity Econom-
ics Division, Economic Research Service,
Washington D.C., various issues.

U.S. Dept. of Agriculture, Oil Crops: Situation and
Outlook, Commodity Economics Division,
Economic Research Service, Washington D.C.,
various issues.

U.S. Dept. of Agriculture, Vegetables and Specialties:
Situation and Outlook Report, Commodity
Economics Division, Economic Research Service,
Washington D.C., various issues.

U.S. Dept. of Agriculture and U.S. Dept. of Health and
Human Services, Nutrition and YourHealth: Die-
tary Guidelines for Americans, 2nd Ed., Home
and Garden Bulletin No. 252, Washington D.C.,
Govt. Printing Office, 1985.

U.S. Dept. of Commerce, CPI: Detailed Report, Wash-
ington D.C., Govt. Printing Office, various issues.

U.S. Dept. of Commerce, Fishery Statistics of the
United States, National Marine Fisheries Service,
Washington D.C., various issues.

U.S. Dept. of Commerce, Fisheries of the United
States, National Marine Fisheries Service, Wash-
ington D.C., various issues.

U.S. Dept. of Commerce, Survey of Current Business,
Bureau of Economic Analysis, Washington D.C.,
various issues.

U.S. Dept. of Commerce, U.S. Exports: Schedule A
Commodities by Country, Bureau of Census,
Washington D.C., various issues.

U.S. Dept. of Commerce, U.S. General Imports and
Imports for Consumption: ScheduleA Commodi-
ties by Country, Bureau of Census, Washington
D.C., various issues.

U.S. Dept. of Health Education and Welfare, Public
Health Service, Healthy People: The Surgeon Gen-
eral’s Report on Health Promotion and Disease
Prevention, DHEW Publ. No. 79-55071, Washing-
ton D.C.: Govt. Printing Office. 1979.

U.S. Senate Select Committee on Nutrition and Human
Needs, Dietary Goals of the United States, 2nd
Ed. Washington D.C.: Govt. Printing Office, 1977.

van Ravenswaay, E.O. and J.P. Hoehn, The Impact of
Health Risk on Food Demand: A Case Study of
Alar and Apples, Staff Paper N0. 90-31, Dept. of
Agri. Econ., Michigan State University, June 1990.

Varian, H.R., Microeconomic Analysis, W.W. Norton
& Co., 2nd Edition, 1984.

-. -v

U

P‘

1k

_¢~_~._¢._n@_<¢_n¢_~¢_.»_¢-_@~_.»_-_.~_n~_v»_n~_-__~_°~_m¢_¢»_
3.:.x.x.x.xu_x*xrx.xH_:_xu.xu.:urxu_x_xu_x_xu_xu_xu,
0

fiifizmoznom

€-_~._¢»_n-_w¢_@»_~¢_.-_¢¢_@~_-_>~_¢>_»>_¢~_@»_-_.~_=~_@¢_¢3
3,:_x.:_x:auSn~:u-xu.:_x,:u_x.x.:_xu.x_x.xu:a~=n~.
0

00F

00a

00m

00¢

azaofimwzwu

68o: 30a @0323 .5.“ gm 3325 wda 52.5% 603.5 d 0.59m

:2“. If... >mh4DOm . . . . .. 0E9. I “mmm I

_ww_~w_@w_mm_¢m_nm_~w_Fm_ow_@~_m>_-_@>_m>_¢>_n>_~>_P>_o>_mu_mo_
¢n~P¢n~P¢n~_¢@~F¢@~P¢n~.¢@~F¢m~F¢@~_¢n~F¢n~F¢n~F¢n~_¢@u_¢@~_¢n~F¢n~F¢n~F¢@~F¢n-¢m~F

--__-_-_-__-_~A—-_-—--—----------------_--—-—-—--—----

w¢<zw

mounwwm QED
< 535004

0

~  » ~

25

mm? 83 m»? Q22 82 8,

#~141<4-4€#-¢-4~4~#<._A.w..-I~<

om

00F
9.:
ooN
omN

oom

40.7040302305-

6520: .00c=_¢u
>22: =05 3 on occ ma 52E» 05E
0.22.2 nan 20E: S». 02.0w 02.5 9:.

mm? 82 mSP E2 $3 QSWP

    

d 4 #4 ‘J - q _ I - - n -

 

........................................ mrém
 2:
............................. 
, . , , . , , > , > . . . . . 0 . . _ , , > .%&hw;;;;E::;:E;::;E:.com
,>;.,:1»,@@;;;¢:;;z;:;1;1:;;i¢;.¢m~

com

Qzzolahzwu

z

A»

£80: g 000030..” 00w wuuaim 003:5 v60 6088525 600E .m 0.59m

._.<“_>>O._

mm? 39

230E

wwmmIO

$3

w._o_._>> I

33

-

q - - q -

_.....__.__ ....._........V..,

wm<zw

o

To
Nd
m...
¢.o
m...
m6
v.0
m6

26

68o: vzaaomo.» ©3033 .3.“ 3.52m “v9.5a v5“ 603555 600E é 953w

£2 82 m8. 22 m3- 8w_

4d--¢-~<---<<-¢-n-u-‘q

0N

3

8

om mmzho I-Il wmowizo»

oow wu:»hw4 III wwo»<#o¢ ill.

<:._<u\wnz:o._

ma? 82 £2 22 $3 ammo»

- - - - - - q - - - - - - - - - a - - d - - q q - -

........................................................................................................  Pd
Nd

ma? 39 m»? 22 m2! 8F

dwddqllqnaddidcddnddqdiqdddd

m6

 ON

o¢

8

8

P
nznomfikzwu oo

27

mm? .82 R3 E2 m2! oww.

-<<~a<~<-<<<-<¢<--<<¢¢41

<.._._<o\mnz:o._

mm? 32 m2! 33 £2 8.

J\—<-4-<d-¢--<<-<-¢n-<<q

QZDOEwhZmQ

 

c

to
Nd
m6
.1.
m6
0.0

.

‘ Q
68oz 35am ©3023 .5.“ mun-aw H0325 v5 52.526 600E .m 95mm
mmzho I.| m<z<z<m ...... ..
mm._.E< I wwuz<mo |.l
mm? 33 £3 E9 mm? 83
wm<zm

28

div: =0 fig .8 ©3023 .5.“ wuuufim 89:5 U5“ 52.5% 600E .0 PBME

$2 82 m»? 22 $2 25..

¢-<~<-<-<<<<<<¢~J<-4¢J-1¢-<

.=0 Q55 Isl uziwEozw
wz_m<om<s I muhhnm |||
<_._._<u\waz:o._om
$3 89 m3? E2 mmm- owmop
F6
Nd
m6
mmmF cum? m>@P ¢~m_ mwmr ow P
S m6

2:

2.;

ooN

omN
nzaomwwhzwu

29

¥

mm? 39 33 o5- $2 3F

qd-<--q-qq-qndq1-qqqqnd

‘..Q.‘.."
. . . . _ . . ...I...lqQ:.I.I:(.O..,.Q..I.............
.I.Q

<t¢<uqwazaom

mm? 89 m3- 22 mm? 8.

<a4-<--<<--u¢--nJ.<q-<

omN

nzaomihzwu

Q .,
68o: $085.5 oquwuuwwa wuuuoqum wow 8.85m 3H5; win 603353? 600E K 0.5M?‘

35.55  3E Ill >§dom¢ |¢|

2E 13E >E<Q l wims |||..

mm? 82 m5? E3 $9 83

q q q q q q q q q q q q q q q q q q q q q q q q q q q o

m...
m6

50

Appendix B
Coefficient Estimates of Best-Scenario Models

Table A1. Coefficient Estimates for the Aggregate Model using the B-S Index.

Fresh Processed
Meats Dairy Fruit & Veg. Fruit & Veg. Eggs Fats Cereals Sweets
Const 0.22512‘ 0.35097‘ 0.23217‘ 0.05598 003102‘ 0.03110 0.08385‘ 0.02812
(0.08180) (0.00917) (0.04537) (0.08131) (0.01679) (0.03339) (0.04591) (0.02908)
Meats 0.17969‘ 0.03175‘ 0.03512‘ 0.02880‘ 0.00165 0.01968‘ 0.03926‘ 0.02343‘
(0.0 1 247) (0 .00307) (0.00948) (0.01 349) (0.00208) (0.00463) (0.00867) (0. 0052 1)
Dairy -0.03175‘ 007200‘ 0.02209‘ -0.00226 0.00136 000043 0.02005‘ 0.00509‘
(000307) (000535) (000394) (000679) (000172) (000345) (000643) (000286)
Fresh F.&V. -0.03512‘ 0.02209‘ 0.05952‘ 0.00001 0.00309 0.00461 0.02165‘ 0.01783‘
(000948) (0.00394) (0.01326) (0.01 447) (000200) (000537) (0.00990) (0.006 l 5)
Proc. F.&V. 0.02880‘ 0.00226 0.00001 0.00205 0.00199 0.00401 0.02393 0.01122
(0.01349) (0.00679) (0.01 447) (0.03334) (000328) (000774) (001994) (0.02151)
Eggs 0.00165 0.00136 0.00309 0.00199 0.01111‘ 0.00114 0.00649‘ 0.00064
(000208) (0.00172) (0.00200) (0.00328) (000091) (000169) (000319) (000167)
Fats 0.01968‘ 0.00043 0.00461 -0.00401 0.001 14 003964‘ 0.01290 -0.00694
(000463) (0.00345) (0.00537) (000774) (0.001 69) (0.00633) (0.00973) (000507)
Cereals 0.03926‘ 0.02005‘ 0.02 165‘ 0.02393 0.00649‘ 0.0 1 290 0.08046‘ 0.00403
- (0 .00867) (000643) (000990) (0.01 994) (0.003 1 9) (0.00973) (002444) (0.00835)
Sweets -0.02343‘ 0.00509‘ 0.0 1 783‘ 0.01 1 22 0.00064 0.00694 0.00403 0.00037
(0.00521) (0.00286) (00061 5) (0.02 1 51) (0.00l67) (0.00507) (000835) NA
Expend 0.01044 0.0 1988‘ 0.00946‘ 0.02453‘ 0.001 04 0.00860‘ 0.00307 0.00464
(000788) (00009 1) (000436) (0.00783) (0.00150) (000268) (000441) (000950)
LAG(W) 0.13358‘ 0.20705‘ 0.20052‘ 0.10377‘ 0.07618 0.08616 0.09236 0.05787
(003722) (0.0415 1) (005627) (004916) (007482) (006538) (005847) (008592)
B-S(0) 0.00143‘ -0.00036‘ 0.00004 0.00009 0.000 1 3‘ 0.00009 0.00065 0.00023
(000060) (00001 1) (000031) (000074) (000007) (000018) (000040) (000030)
B-S(1) 000107‘ 0.00027‘ 0.00003 0.00007 0.00010‘ 0.00006 0.00049 0.00017
(000045) (000008) (0.00023) (000055) (000005) (000014) (000030) (000022)
B-S(2) 0.00036‘ 0.00009‘ 0.00001 0.00002 0.00003‘ 0.00002 0.00016 0.00006
(0 .000 1 5) (000003) (000008) (0.00018) (000002) (000005) (0.00010) (000007)
RHO 0.55839‘ 0.31806‘ 051228‘ 0.57772‘ 0.86378‘ 0.55045‘ 0.55153‘ 075694‘
(004706) (008379) (006792) (004781) (0.07093) (0.10524) (008627) (005537)
R-Square 0.9570 0.9743 0.9038 0.8114 0.9955 0.9136 0.9163 0.8912
RUNS 0.1839 0.1319 0.9402 -2.7586‘ 0.9402 1.0155 -1.6617 0.1986
Durbmh 2.0393‘ 0.9821 0.7360 2.0763‘ 2.7652~ 0.8294 1.248s 0.3246

‘Coefficient or test significant at the a = 0.10 level.

Table A2. Coefficient Estimates for the Meats Model using the Combined Indiees.

Beef Fish Poultry Pork
Const 0.25415* -0.05247 0.04598 0.46523‘ .
(0.06630) (0.06114) (0.04022) (0.05633) Q
8661 006212* 002042* 004586~ 0.00416
(0.01633) (0.01162) (0.00885) (000957)
Fish 002042~ 004539~ -0.01079 3 001418-
(0.01 162) (0.01392) (0.00776) (0.00847)
Poultry 004586* 0.01079 008093* 002428
(000885) (000776) (0.00988) (0.00743)
Pork 0.00416 001418~ 002428~ 003430'
(000957) (000847) (000743) (001009)
Expend 001655* 0.01050 0.01209* 003914*
(000883) (000857) (000594) (000663)
Q2 0.01937* 002072* 001407* 001551'
(0.00335) (000281) (000227) (000209)
Q3 002835* 003469* 001173* 001741*
(000324) (000331) (000222) (000243)
Q4 002792* 0.01328* 002844~ 001413’
(00031 1) (000386) (0.00209) (000272)
LAG(w) 0.29200* 0.25169* 0.28479* 0.28229*
(005767) (006086) (005674) (006432)
88(0) 000011- 000008* 000006* -0.00002
(000004) (000003) (000002) (000003)
88(1) 000009’ 000007* 000005* -0.00002 
(000004) (000003) (000002) (000002) *,
88(2) 000002* 000001* 000001* 0.00000
(000001) (000000) (000000) (000000)
c8 Index 000607* 0.00038 000366* 000295‘
(000096) (000084) (000058) (000071)
R-Square 0.9338 0.8979 0.9305 0.8421
RUNs 0.8019 0.2410 0.6148 -1.7182
Durbmh 1.7496 -1.2970 20307* 1.6628
‘Coefficient or test signiificant at the 01 = 0.10 level.
4
g.

32

-LTable A3. Coefficient Estimates for the Dairy Model using the Combined Indices.

Whole Lowfat Cheese Frozen
, Const 0.88083' 0.11372 0.47194‘ 0.00124
(0.32099) (0. 16008) (0.24886) (0.07755)
Whole 0.07834 0.02373 0.10397‘ 0.00190
(0.08322) (0.04952) (0.02814) (0.02780)
Lowfat 0.02373 0.00152 0.01136 0.03661‘
(0.04952) (0.03499) (0.0 1676) (0.0 1689)
Cheese -0. 10397‘ 0.01 136 0.1 1702‘ 0.02440‘
(0.02814) (0.01676) (0.02676) (0.01020)
Frozen 0.00190 0.03661‘ 0.02440‘ 005911‘
(0.02780) (0.01689) (0.01020) (0.01259)
Expend -0.07 1 14" 0.0 1744 0.04401 000969
(004035) (0.02391) (003762) (0.01 156)
LAG(w) 0.74366‘ 0.82786‘ 0.66307‘ 0.52521‘
(007134) (0.09549) (009027) (009709)
B-S(0) 000195‘ 0.00095‘ 000093‘ 0.00015
(000059) (000040) (0.00056) (000022)
B-S(1) -0.00146‘ 000071‘ 000070‘ 0.00011
(000044) (000030) (000042) (000017)
B-S(2) 0.00049‘ -0.00024‘ 0.00023‘ 0.00004
(0.0001 5) (000010) (000014) (000006)
C-S Index 0.00011 0.00010 000217‘ 000154‘
(0.00138) (000099) (000132) (000064)
R-Square 0.9989 0.9966 0.9975 0.9405
.-\ RUN5 -1.6617 0.5745 -1.6617 0.6330
Durbmh 1.5410 0.7823 0.0853 -1.2354
Ilgoefficient or test significant at the a = 0.10 level.
Table A4. Coefficient Estimates for the Vegetable Model using the C-S Index.
Potatoes Lettuce Tomatoes Other Fresh
Const 1.33030‘ 0.33548‘ 0.02781 0.43303‘
(0.19485) (0. 10584) (0.1 3436) (0.20011)
Potatoes 0.1701 1‘ 005128‘ 002355‘ 009527‘
(002476) (0.01 374) (0.01300) (002680)
Lettuce -0.05 128‘ 0.16218‘ 0.041 16‘ 0.06974‘
(0.01374) (001749) (001556) (0.02327)
Tomatoes 0.02355‘ 0.04116‘ 0.12303‘ 0.05832‘
(0.01 300) (0.01556) (00274 1) (002834)
Fresh 0.09527‘ 0.06974‘ 0.05832‘ 0.22333‘
(0.02680) (002327) (0.02834) (0.05 102)
Expend 0.1 5623‘ 0.06937‘ 0.00255 0.08941‘
yx (002662) (0.01 577) (001753) (003086)
‘ mew) 0.37085‘ 0.41519‘ 0.47061? 0.40709
(007063) (006858) (009025) (007208)
C-S Index 0.00826‘ 0.00823‘ 0.00230 0.00272
(000243) (000165) (000170) (00031 2)
R-Square 0.9341 0.9148 0.7569 0.8461
RUNS 0.6663 0.0269 -2.8335‘ 3.2122‘
F“ Dmbimh 1.1136 0.7200 0.9298 1.9 195

|'Coefficient or test significant at the a = 0.10 level.

33

Table A5. Coefficient Estimates for the Fruit Model using the C-S Index.

w.

Oranges Apples Bananas Other Fresh a’;
Const 0.35594‘ 0.07573 0.12752 0.26259‘
(0.08154) (0.10499) (0.08585) (0.1 2647) Q"
Oranges -0.01278 -0.01888 0.01412 0.04578
(0.02134) (0.01989) (0.01163) (0.03757)
Apples 0.01888 o.14051* 0.04434* ' 0.07729
(0.0 1989) (0.03264) (0.0 1678) (0.05047)
Bananas 0.01412 0.04434‘ 0.06900‘ 0.01054
(0.0 1 163) (0.01678) (0.0l522) (00242 1)
Fresh 0.04578 -0.07729 0.01054 0.04205
(003757) (005047) (0.0242 1) (0.09429)
Expcnd 0.04476‘ 0.01497 0.00616 0.02363
(001245) (00181 1) (001379) (001842)
LAG(w) 0.21666‘ 0.18567‘ 0.18591‘ 0.16151
(0. 10640) (009772) (009516) (00991 2)
C-S Index 0.0041 3‘ 0.00279 0.00000 0.00120
(000160) (000240) (000172) (000256)
R-Square 0.6635 0.7592 0.6564 0.6556
RUNS 0.0269 1.7106 0.5413 0.9532
Durbinh 0.1225 -1.0855 -1.0012 0.6929
‘Cocfficicnt or test significant at the a = 0.1 0 level.
Table A6. Coefficient Estimates for the Fats and Oils Model using Dummy Shifters. *'
Butter Margarine Shortening Salad Oil
Const 0.32706‘ 0.38824‘ 0.00283 0.14781
(0. 10909) (009127) (009364) (0.11842)
Butter 0.1 25 1 5‘ 0.03 178* 005367‘ 0.03970
(002816) (0.0 1034) (002301) (002817)
Margarine 0.03178*" 0.05939* 0.01814 0.00947
a (001034) (002858) (001681) (003052)
Shortening 0.05367' ‘ 0.01814 0.18155‘ 0.10973‘
(00230 1)_ (0.0l68l) (003834) (004099)
Salad Oil 0.03970 0.00947 0.10973‘ 0.15890‘
(002817) (003052) (004099) (006163)
Expcnd -0.05041* 0.04102‘ 0.04199 0.04944
(0.01868) (001331) (0.01771) (0.02209)
LAG(w) 0.55645‘ 0.38284‘ 0.26516‘ 0.44587‘
(009562) (0.1 1305) (0.11371) (0. 10995)
L7()s 002537‘ 0.00892‘ 002291‘ 00073 1
(0.01191) (000417) (000888) (0.01181)
E805 0.03870‘ 001400‘ 0.01380 0.02908‘ a
(001461) (0.00521) (000895) (001377) P‘
L805 -0.00387 0.02677‘ 0.01734 0.02191
(002708) (0.01208) (0.01871) (0.02593)
R-Square 0.9765 0.9498 0.7128 0.9682
RUNS 0.4171 2.1628 -1.3188 0.5413
Durbm), 1.5482 4.8480 0.3307 0.8280
‘Coefficient or test significant at the a = 0.10 level. 

34