Download Consumer knowledge, food label use and grain consumption in the

Applied Economics, 2008, 40, 437–448
Consumer knowledge, food label
use and grain consumption
in the US
Biing-Hwan Lina and Steven T. Yenb,*
a
US Department of Agriculture, Economic Research Service,
1800 M Street NW, Room N2110, Washington, DC 20036–5831, USA
b
Department of Agricultural Economics, The University of Tennessee,
302 Morgan Hall, 2621 Morgan Circle, Knoxville, TN 37996–4518, USA
Responding to mounting evidence of the association between whole-grain
consumption and a reduced risk of heart problems and other diseases, as
well as an increased probability of body weight maintenance, the US
Government has strongly encouraged its citizens to increase consumption
of whole grains. However, compared against the 2005 Federal dietary
recommendations, in 1994–1996 only 6% of Americans met the current
recommended whole-grain consumption. To narrow this huge gap between
actual and recommended consumption of whole grains, considerable
changes in consumer behaviour will be needed. A demand system with two
censored consumption equations and endogenous food label use and
nutrition knowledge variables is estimated to investigate the factors that
affect the consumption of whole and refined grains. Food label use and
nutrition knowledge are found to play important roles in the consumption
of refined- and whole-grain products, as are sociodemographic variables.
The results can be used to help develop effective nutrition education
messages and targeting strategies to promote consumption of whole grains
in Americans’ diets.
I. Introduction
As a staple food in Americans’ diets, grain
products are available to consumers in two basic
forms – refined and whole grains, both of which can
be enriched. Compared to refined grains, whole
grains provide greater amounts of vitamins, minerals,
fibre and other protective substances. Despite these
nutritional advantages, Americans tend to favour the
consumption of refined grains over whole grains.
Responding to mounting evidence of the association between whole-grain consumption and a reduced
risk of heart problems and other diseases, as well as
an increased probability of body weight maintenance,
the US Government has been promoting consumption of grains, especially whole grains, in Americans’
diets. The Healthy People 2010 aims at increasing the
proportion of persons consuming at least six daily
servings of grain products, with at least three servings
of whole grains (US Department of Health and
Human Services, 2005). Data from the most recent
US Department of Agriculture’s (USDA) food
consumption survey indicate that only half of
Americans consumed six or more servings of grain
*Corresponding author. E-mail: [email protected]
Applied Economics ISSN 0003–6846 print/ISSN 1466–4283 online ß 2008 Taylor & Francis
http://www.tandf.co.uk/journals
DOI: 10.1080/00036840600690298
437
B.-H. Lin and S. T. Yen
438
products a day, and only one in ten consumed three
or more servings of whole-grain products a day
during 1994–1996 (Kantor et al., 2001).
The 2005 Dietary Guidelines for Americans include
several changes in the recommendations for grain
consumption (US Department of Agriculture and US
Department of Health and Human Services, 2005).
First, the recommendations for total grains have been
revised downward. For example, the recommended
total grain consumption is now five ounce-equivalent
(servings) instead of six for a 1600-calorie diet.
Second, at least half of total grains consumed
should come from whole grains. Third, the new
guidelines cover a much wider range of food energy
intakes from 1000 to 3100 calories, compared to the
1600–2800 calories specified in the previous guidelines. Under these new guidelines, the average
American age two and above consumed slightly
more than the recommended total grains (103%),
according to data from the 1994–1996 Continuing
Survey of Food Intakes by Individuals (USDA,
2000). Americans over-consumed refined grains
averaging 75% over the recommendation. It is a
major challenge for Americans to meet the new
guidelines on whole grains, as the 1994–1996
consumption amounted to 31% of the recommended
level and only 6% of consumers met the
recommendation.
The grain industry and the public health community share an interest in increasing whole-grain
consumption, with marketing and public health
campaigns aiming at promoting such consumption.
Designing effective promotional or marketing strategies requires a good knowledge of grain consumption
patterns. What are the factors associated with low or
high consumption of grains and whole grains? Which
subpopulations are particularly behind in meeting the
recommendation? Such information is currently very
limited (Moutou et al., 1998; Kantor et al., 2001;
Harnack et al., 2003).
The objective of this study is to conduct a
regression analysis to identify social, economic,
demographic, knowledge and behavioural factors
that are associated with consumption of whole-grain
and refined-grain products. Heterogeneity of preference has traditionally played a part in consumer
demand and the roles of sociodemographic factors
are often investigated in empirical studies. Other
factors considered in the empirical literature include
consumer knowledge and behaviour. The literature
on the effects of dietary knowledge and food-label use
on food and nutrient intake and diet quality has
proliferated since the release of the 1994–1996 Diet
and Health Knowledge Survey (DHKS) data and the
passage of the 1990 Nutritional Label and Education
Act (NLEA). Dietary knowledge has been linked to
food consumption, including fats and oils (Chern
et al., 1995; Kim and Chern, 1999), fat-modified
foods (Coleman and Wilson, 2002), egg (Brown and
Schrader, 1990; Yen et al., 1996; Kan and Yen, 2003),
meat (Kinnucan et al., 1997; Kaabia et al., 2001) and
25 food groups consumed at and away from home
(Lin et al., 2003). Dietary knowledge has also been
linked to the diet quality of children (Variyam et al.,
1999b), elderly (Howard et al., 1998) and female
household heads (Ramezani and Roeder, 1995). With
respect to nutrient intake, there are reported links
between knowledge and intake of fibre (Variyam
et al., 1996), energy and nutrient density (Bhargava,
2004) and fat and cholesterol (Variyam et al., 1997,
1999a). The use of the nutrition fact panel mandated
under NLEA has been found to affect the intake of
fat (Kreuter and Brennan, 1997; Neuhouser et al.,
1999) and fats, cholesterol, sodium and fibre (Kim
et al., 2000).
Unlike sociodemographic factors, consumer
knowledge and behaviour are likely to also
be determined by some of the factors that determine
consumption; that is, they are likely to be endogenous. In this study, we investigate the roles
of consumer knowledge and food-label use
(as a knowledge-promoting device) as well as sociodemographic factors, in the consumption of grain
products, using data from a national food consumption survey in the US.
As in other empirical analyses based on survey
data, the sample we use contains a notable proportion of observations not consuming whole grains.
This is the issue of censored dependent variable.
In addition, as stated, consumer knowledge and
food-label use are potentially endogenous. It is well
known that statistical procedures not accommodating
censoring or endogeneity produce biased estimates.
To accommodate these data features we construct a
system of censored equations with dual endogenous
regressors. Such an econometric specification has not
been reported in the literature.
II. Data
The USDA has conducted periodic food consumption surveys in the US since the 1930s. The most
recent food consumption surveys, the 1994–1996
Continuing Survey of Food Intakes by Individuals
(CSFII) and its companion DHKS provide the data
for this study (USDA, 2000). The CSFII is the only
national survey that includes a comprehensive coverage of dietary and health knowledge and attitudes.
Consumer knowledge, food label use and grain consumption
Each year of the CSFII survey comprises a nationally
representative sample of noninstitutionalized persons
residing in the US.
In the CSFII, two nonconsecutive days of dietary
data for individuals of all ages were collected 3 to 10
days apart through in-person interviews using
24-hour recalls. The 1994–1996 CSFII data provide
information on the food intakes of 15303 individuals,
who provided a list of food items and their amounts
consumed. After the respondents reported their first
day of dietary intake, an adult 20 years old was
randomly selected from each household to participate
in the DHKS. The DHKS questions cover a wide
range of issues, including self-perceptions of the
adequacy of nutrient intakes, awareness of
diet–health relationships, knowledge of dietary
recommendations, perceived importance of following
dietary guidance, use and perceptions of food labels
and behaviours related to fat intake and food safety.
Of the 7842 households eligible to participate in the
DHKS, respondents from 5765 households completed the survey.
The USDA created several technical databases,
including the Pyramid Servings Database (PSD), to
support use of the CSFII data. The PSD converts the
amount of food consumed into the number of
servings for comparison with dietary recommendations in the 1995 and 2000 Dietary Guidelines for
Americans. The PSD shows, for each food consumed,
the number of servings from 30 food groups,
including refined and whole grains. However, in the
2005 Dietary Guidelines, recommendations on food
consumption are expressed in cups (for fruits,
vegetables and dairy products) and ounce-equivalents
(grains and meat) instead of servings. This does not
affect the measurement of grains because one ounceequivalent is identical to one serving for grain
products. Therefore, the PSD is still directly applicable to the current recommendation on grain
consumption.
Socioeconomic and demographic data for the
sample households and their members are also
collected in the CSFII. The explanatory variables
for grain consumption (refined and whole grains
separately) include household income, household
size, household structure, gender, age, race/ethnicity,
location and season (see Table 1 for variable
definitions and sample statistics). We hypothesize
that the use of a nutrition label and the perceived
importance of consuming plenty of grain products
also affect grain consumption, and these two
variables are endogenized in a system of four
equations. In addition to income, gender, age and
race/ethnicity, the use of a nutrition label and
perceived importance of grain consumption are
439
hypothesized to be affected by education, exercise,
smoking, whether the respondent is a meal planner
and whether anyone in the household is on a special
diet. An attitudinal variable (pessimistic) is also used
which indicates whether the individual agrees with the
statement that ‘some people are born to be fat and
some thin; there is not much you can do to change
this’.
In the DHKS, respondents were asked about their
perceived importance (very, somewhat, not too, or
not at all important) in choosing a diet with plenty of
breads, cereals, rice and pasta. The answers were
grouped into important (very or somewhat) and not
important (not too or not at all). The respondents
were also asked when they buy foods, do they often,
sometimes, rarely, or never use the information on:
(1) the short phrases on the label like ‘low fat’ or
‘light’ or ‘good source of fibre’, (2) the list of
ingredients, (3) the nutrition panel listing the
amount of nutrients and (4) claims on health benefits
of nutrients or foods. These four possible answers are
grouped into use (often or sometimes) and nonuse
(rarely or never). As explained later, the four
definitions of label use produce similar results.
Label use, perceived importance of grain consumption and many other exogenous variables come from
the DHKS, hence our analysis is limited to the CSFII
adult sample. Excluding observations with missing
values, the final sample contains 5501 adults. Of the
sample, 72.8% consumed whole-grain products,
whilst almost all individuals (99.8%) consumed
refined-grain products.
III. Econometric Model
We develop an estimation procedure for an equation
system with censored dependent variables and binary
endogenous regressors. In what follows observation
subscripts are suppressed for brevity. The endogenous regressors, food-label use ( y1) and nutrition
knowledge ( y2), are both binary and therefore
specified as probit:
yi ¼ 1ðz0 i þ ui > 0Þ,
i ¼ 1, 2
ð1Þ
The censored equations for whole grains ( y3) and
refined grains ( y4) are
yi ¼ maxð0, x0 i þ i1 y1 þ i2 y2 þ ui Þ,
i ¼ 3, 4 ð2Þ
In the previous equations, 1() is a binary indicator
function, z and x are exogenous vectors of
explanatory variables, i and i are conformable
vectors of parameters, gi1 and gi2 are scalar
B.-H. Lin and S. T. Yen
440
Table 1. Variable definitions and sample statistics (n ^ 5501)
Variable
Endogenous variables
Label use
Importance
Whole grains
Refined grains
Definition
Use the short phrases on the label like ‘low fat’ or ‘light’ or
‘good source of fibre’: 1 ¼ often or sometimes; 0 ¼ rarely or
never.
Perceived importance in choosing a diet with plenty of breads,
cereals, rice and pasta: 1 ¼ very or somewhat; 0 ¼ not too
important or not at all.
Daily consumption of whole grains (servings), 2-day average
Consuming sample (n ¼ 4003)
Daily consumption of refined grains (servings), 2-day average
Household characteristics: continuous exogenous variables
Income
Household income as percent of poverty
HH size
Number of persons in the household
Mean
0.61
0.74
0.28 (0.34)
0.38 (0.34)
1.34 (0.61)
160.90 (137.10)
2.56 (1.46)
Binary exogenous variables (yes ¼ 1; no ¼ 0)
Household characteristics
HH type 1
Household is dual-headed, with children
HH type 2
Household is dual-headed, without children
HH type 3
Household is single-headed, with children
HH type 4
Household is single-headed without children (reference)
Special diet
A family member is on a special diet
0.28
0.36
0.08
0.28
0.27
Individual characteristics
Male
Age 20–30
Age 31–40
Age 41–50
Age 51–60
Age >60
Black
Hispanic
Asian
Other race
White
Quarter 1
Quarter 2
Quarter 3
Quarter 4
Midwest
South
West
Northeast
Rural
Suburb
City
<high school
High school
Some college
College
Meal planner
Exercise
Smoker
Pessimistic
0.50
0.14
0.18
0.18
0.18
0.32
0.11
0.08
0.02
0.01
0.78
0.23
0.26
0.28
0.24
0.25
0.35
0.20
0.19
0.27
0.44
0.30
0.22
0.34
0.21
0.23
0.70
0.48
0.26
0.43
Note: SD in parentheses.
Gender is male
Age 20–30
Age 31–40
Age 41–50
Age 51–60
Age 61 or older (reference)
A nonHispanic Black
Of the Hispanic origin
Asian Pacific Islander
None of the above nor White
NonHispanic White (reference)
Dietary recalls taken in January–March
Dietary recalls taken in April–June
Dietary recalls taken in July–September
Dietary recalls taken in October–December (reference)
Resides in a Midwestern state
Resides in a Southern state
Resides in a Western state
Resides in a Northeastern state (reference)
Resides in a rural area
Resides in a suburb
Resides in the central city (reference)
Did not complete high school (reference)
Completed high school education
Attended college for less than 4 years
Had 4 or more years of college education
Main meal planner of household
Exercised vigorously: at least twice a week
Currently smoking cigarettes
Agrees with statement that some are born to be fat and
some thin
Consumer knowledge, food label use and grain consumption
parameters, and the error terms e [u1, u2, u3, u4]0
are distributed as e N(0, ) with probability
density function (pdf) f(u1, u2, u3, u4). The covariance
matrix is defined with error correlations ij and
SD i such that 1 ¼ 2 ¼ 1 (because variables y1
and y2 are both binary).1 The model considered
here is an extension of the Tobit system (Amemiya,
1974), as well as the Kuhn–Tucker model (Wales
and Woodland, 1983) when prices are constant,
in that endogenous variables are present in
Equation 2. It also generalizes the multiple- and
double-selection models (Catsiapis and Robinson,
1982; Tunali, 1986) and multiple-treatment models
(Keane and Moffitt, 1998) in that there are
multiple outcomes in Equation 2 and that, in
addition, these outcome variables are censored.
The likelihood function needs to be constructed
for different ‘sample regimes’, characterized by
zero/positive outcomes of whole grains and refined
grains, and is described subsequently.
Both whole grains and refined grains are censored
Consider first a sample regime in which whole grains
and refined grains are both zeros (not consumed).2
To facilitate description of the estimation procedure,
define
dichotomous
indicators
k1 ¼ 2y1 1,
a
diagonal
matrix
D ¼ diag
k2 ¼ 2y2 1,
(k1, k2, 1, 1) and vector r ¼ [r1, r2, r3, r4]0 , where
ri ¼ z0 i for i ¼ 1, 2 and ri ¼ x0 i þ gi1y1 þ gi2y2
for i ¼ 3, 4. The sample regime is characterized
by
inequality
e* Dr,
where
e ¼ De ¼
½u1 , u2 , u3 , u4 0 N ð0, Þ such that ¼ DD0 , for
which the likelihood contribution is
Z 1 r1 Z 2 r2 Z r3 Z r4
fðu1 , u2 , u3 , u4 Þdu4 du3 du2 du1
L1 ¼
Z1 1 1 1
¼
fðe Þde
ð3Þ
e Dr
Only whole-grain equation is censored
Consider a regime in which refined grains are
consumed but whole grains are not. Using
Equations 1 and 2 and following a similar procedure,
441
the likelihood contribution is
Z 1 r1 Z 2 r2 Z r3
hðu1 , u2 , u3 ju4 Þdu3 du2 du1
L2 ¼ gðu4 Þ
1
1
1
ð4Þ
where g(u4) is the marginal density of u4,
hðu1 , u2 , u3 ju4 Þ is the conditional density and
u4 ¼ y4 (x0 4 þ gi1y1 þ gi2y2).3 The likelihood contribution for a regime with refined grains censored
follows from Equation 4 symmetrically with equation
subscripts 3 and 4 reversed.
Neither whole-grain nor refined-grain equation
is censored
The likelihood contribution is
Z 1 r1 Z 2 r2
L3 ¼ gðu3 , u4 Þ
hðu1 , u2 ju3 , u4 Þdu2 du1
1
ð5Þ
1
where ui ¼ yi (x0 i þ gi1y1 þ gi2y2), i ¼ 3, 4, and
moments of the marginal distribution g(u3, u4) and
conditional distribution hðu1 , u2 ju3 , u4 Þ are similar
to those in Equation 4, with partition of e*
(and covariance matrix ) at the second row
(and column).
The sample likelihood function for the system is the
product of the likelihood contributions L1, L2 or L3
over the sample, depending on the regime for each
observation. The integrals in Equations 3, 4 and 5 can
be evaluated by simulation or quadrature. The model
reduces to an exogenous system when error correlations are zero between the two binary equations
(food-label use and nutrition knowledge) and those
of the consumption equations. The corresponding
parametric restrictions are
i, j ¼ 0 8 i ¼ 3, 4; j ¼ 1, 2
ð6Þ
and a test for these parametric restrictions amounts
to one for the joint endogeneity of y1 and y2. The
restricted model implied by Equation 6 can be
estimated by a bivariate probit for Equation 1 and
Amemiya’s (1974) Tobit system for Equation 2.
Further, when all error correlations are equal to
zero, that is,
ij ¼ 0
8i > j
ð7Þ
1
We present the bivariate censored system with dual endogenous binary regressors for ease of exposition. A larger system
with a different number of endogenous variables can be estimated with slight changes in the likelihood function. Estimation of
the censored system is slightly easier with continuous (vs. discrete) endogenous variable(s) which do not introduce additional
integrals as the binary variables do (Yen et al., 2005).
2
As noted above zero observations of refined grains occur in a very small proportion of the sample, but a two-zero regime is
nevertheless reflective of the sample.
3
Given the distribution e* N (0, ), partition at the third row with submatrices 11, 12, 21 and 22 such that 11 is
3 3 and 22 is 1 1. Then using properties of the multivariate normal distribution (Kotz et al., 2000), we have
1
u4 N (0, 22) and u1 , u2 , u3 ju4 N ð12 1
22 u4 , 11 12 22 21 Þ:
B.-H. Lin and S. T. Yen
442
the model reduces to one with separate probit and
Tobit models from Equations 1 and 2. Tests of these
restricted models against the unrestricted model can
be carried out by likelihood-ratio (LR) tests.
To examine the marginal effects of explanatory
variables, we derive relevant probabilities, conditional
means and unconditional means of the dependent
variables. Partition the error vector e at the second row
such that e ¼ ½e01 , e02 0 ¼ ½u1 , u2 ju3 , u4 0 , with corresponding partitioning in the covariance matrix
11 12
¼
ð8Þ
21 22
Then, using properties of the multivariate normal
distribution (Kotz et al., 2000), e2 can be expressed
conditional on e1 as
e2 ¼ 21 1
11 e1 þ "
ð9Þ
such that " is independent of e1 and x,
e1 ¼ [y1 z0 1, y2 z0 2]0 ,
and
" N ð0, 22 21 1
11 12 Þ. Substituting Equation 9 into Equation
2 gives the conditional Tobit system
yi ¼ max 0, x0 i þ i1 y2 þ i2 y2 þ "i , i ¼ 3, 4
ð10Þ
where
" ¼ ½"3 , . . . , "m 0 ¼ 21 1
11 e1 þ " N ð0, Þ
ð11Þ
1
1
such that ¼ 21 1
11 11 12 þ 22 21 11 12 .
The covariance of "* in Equation 11 follows from
the independence between " and e1. The motivation of
writing the conditional Tobit system as Equation 10
is that the parameters ’s and g’s can be estimated
consistently with ‘unobserved heterogeneity’ due to
omission of 21 1
11 e1 . Denote the univariate standard normal pdf as ðÞ, cdf as () and the SD of "i
as i which is the squared root of the i-th diagonal
element of in Equation 11. Then, the probabilities,
conditional means and unconditional means of yi
(for i ¼ 3, 4) are
0
ðx i þ i1 y1 þ i2 y2 Þ
Prðyi > 0Þ ¼ ð12Þ
i
Eðyi jyi > 0Þ ¼ x0 i þ i1 y1 þ i2 y2
½ðx0 i þ i1 y1 þ i2 y2 Þ=i þ i
½ðx0 i þ i1 y1 þ i2 y2 Þ=i 0
ðx i þ i1 y1 þ i2 y2 Þ
Eðyi Þ ¼ i
ðx0 i þ i1 y1 þ i2 y2 Þ
0
ðx i þ i1 y1 þ i2 y2 Þ
þ i i
ð13Þ
ð14Þ
Marginal effects of explanatory variables x, y1 and y2
can be derived by differentiating Equations 12, 13
and 14. The effects of each discrete variable can be
calculated as the changes in these probabilities and
means resulting from a finite change (e.g. from 0 to 1)
in the variable, while holding all other variables
constant (e.g. at sample means).
IV. Results
The four-equation system, consisting of binary
equations for food-label use and perceived importance of grains and censored equations for whole and
refined grains, is estimated by maximizing the
likelihood function described above. Asymptotic SE
of the parameter estimates are calculated by inverting
the outer products of the gradients (Berndt et al.,
1974).
One specification issue relates to choice of the
variables in the probit equations and consumption
equations. Unlike instrumental variable estimation
for which exclusion restrictions are often needed, in
ML estimation the nonlinear identification criteria
(i.e. linear independence of the first derivatives of the
likelihood function with respect to parameters) are
met due to the functional form and distributional
assumptions for the current system. However, to
avoid over-burdening the nonlinear functional forms
for parameter identification, we make a priori
assumptions in that regard. First, whilst it is hard
to contemplate how education per se might affect
grain consumption, the educational variables are
more likely to affect label use and perceived
importance of grains, so they are only included in
the probit equations through which they affect
consumption indirectly. Other variables used solely
in the probit equations include being a meal planner
and behavioural variables such as exercise and
smoker, and the attitudinal variable ‘pessimistic’.
On the other hand, because grain consumption is
more likely to vary across regions, household types,
seasons, household size and urbanization categories
than label use and perceived importance of grains,
these variables are used only in the consumption
equations.
As discussed above, there are four alternative
variables representing the use of food labels – the
list of ingredients, short phrases, nutrition panel and
health claims. These alternative specifications of label
use produce similar results. For brevity, we only
present the results from using short phrases. This is
because a short-phrase example given during the
interview is related to the fibre content of foods.
Consumer knowledge, food label use and grain consumption
Whole grains are known for their rich fibre content.
Also estimated for comparison are the two restricted
models, by imposing the restrictions in Equation 6
and 7. Based on the results of LR tests, the hypothesis
of joint exogeneity of food-label use and nutrition
knowledge was rejected (LR ¼ 189.55, df ¼ 4,
p-value < 0.001), as was the hypothesis of complete
independence among all equations (LR ¼ 469.22,
df ¼ 6, p-value < 0.001). The results suggest that use
of a food label and consumer knowledge should both
be treated as endogenous, and that the two binary
equations and two consumption equations should be
estimated as a system.
Maximum-likelihood estimates for the system are
reported in Table 2. All but one of the six error
443
correlation coefficients are significant at the 10%
level of significance or lower, corroborating results of
the LR tests which suggest endogeneity of label use
and perceived importance of grains and simultaneity
of the system. At the 10% level of significance
or lower, two-thirds of the variables are significant in
the label use and whole grains equations, and
about half are significant in the perceived importance
and refined-grain equations. In addition, label use
has a positive effect on whole-grain consumption,
as does perceived importance in both the whole-grain
and refined-grain equations, all at the 5% level or
lower.
The use of food labels and perceived importance of
grains are affected by household financial and human
Table 2. Parameter estimates of demands for whole and refined grains with endogenous food-label use and health belief
Label use
Importance of grains
Whole grains
Refined grains
Variable
Coeff.
SE
Coeff.
SE
Coeff.
SE
Coeff.
SE
Constant
Income/103
Male
Age 20–30
Age 31–40
Age 41–50
Age 51–60
Black
Asian
Other race
Hispanic
Special diet
High school
Some college
College
Meal planner
Exercise
Smoker
Pessimistic
Midwest
South
West
Rural
Suburban
HH type 1
HH type 2
HH type 3
Quarter 1
Quarter 2
Quarter 3
HH size
Label use
Importance
SD ( i)
Error correlations
Importance
Whole grains
Refined grains
0.186***
0.487***
0.481***
0.061
0.023
0.074
0.114**
0.007
0.171
0.251
0.117*
0.224***
0.168***
0.190***
0.323***
0.075*
0.126***
0.273***
0.153***
0.070
0.146
0.043
0.061
0.056
0.055
0.054
0.059
0.155
0.181
0.068
0.041
0.051
0.058
0.061
0.045
0.037
0.042
0.037
0.484***
0.211
0.010
0.149***
0.122**
0.123**
0.069
0.208***
0.124
0.269
0.070
0.051
0.067
0.125**
0.368***
0.133***
0.134***
0.163***
0.058*
0.070
0.155
0.043
0.060
0.057
0.055
0.054
0.056
0.149
0.196
0.068
0.042
0.047
0.055
0.060
0.043
0.035
0.040
0.035
0.208***
0.094
0.042***
0.061***
0.054***
0.065***
0.055***
0.118***
0.253***
0.117**
0.028
0.021
0.054
0.059
0.017
0.022
0.022
0.020
0.019
0.022
0.054
0.060
0.023
0.016
1.190***
0.060
0.148***
0.156***
0.124***
0.053**
0.046*
0.091***
0.487***
0.037
0.048
0.018
0.070
0.074
0.024
0.028
0.028
0.027
0.025
0.029
0.054
0.077
0.032
0.020
0.043**
0.005
0.137***
0.034**
0.025*
0.055**
0.027*
0.095***
0.015
0.003
0.023
0.078
0.168***
0.421***
0.451***
0.019
0.018
0.020
0.016
0.015
0.026
0.017
0.028
0.017
0.017
0.016
0.063
0.061
0.059
0.011
0.041*
0.087***
0.191***
0.077***
0.010
0.008
0.002
0.031
0.031
0.027
0.044**
0.050
0.059
0.210**
0.592***
0.025
0.023
0.027
0.023
0.020
0.035
0.022
0.042
0.023
0.023
0.022
0.088
0.088
0.094
0.008
0.178***
0.023
0.163***
0.169**
0.068
0.025
0.083
0.088
0.510***
0.156*
0.065
0.087
Notes: Log-likelihood value ¼ 14611.412.
***, ** and * indicate significance at 1%, 5% and 10% levels, respectively.
444
capital, demographics, life style, diet/health attitude
and the respondent’s role in the household. Higher
household income, as a percent of the poverty level,
contributes to the use of food labels but not the
perceived importance of grain consumption. Higher
educational attainment results in a greater likelihood
to use food label and to perceive grain consumption
as important. Meal planners and respondents who
engage in vigorous exercise at least twice a week are
more likely to use food labels and to perceive grain
consumption as important. Conversely, cigarette
smokers and individuals who are pessimistic about
the effect of food choice on health are less likely to
use food labels and to perceive grains as important.4
Gender, age, race and ethnicity have some effects
on label use and perceived importance. Compared
with their respective counterparts, males are less
likely to use food labels and younger adults more
likely to perceive grain consumption as important.
Compared with Whites, Blacks are less likely
to perceive grain consumption as important.
As expected, when a household member is on a
special diet, the respondent is more likely to use food
labels in grocery shopping.
The effects of variables on label use and perceived
importance are quantified further by calculating the
effects of these variables on the probabilities of label
use and perceived importance. The results suggest
that gender and college education have the most
notable effects on the probabilities, with male 18%
less likely to use food labels than female, and
individuals with a college education 12% more
likely to use food labels and 10% more likely
to perceive grain consumption as important.
The marginal effects of other variables on the
probabilities of label use and perceived importance
of grain consumption are presented in the appendix.
The parameter estimates for the two grains
equations suggest notably different consumption
patterns of whole grains and refined grains. All else
equal, men consume more of both whole and refined
grains than women. Younger individuals (age 20–60
in four age categories) consume less whole grains but
more refined grains than their older counterparts
(age >60). Compared to Whites, Blacks consume less
of both whole grains and refined grains, Asians
consume less whole grains but more refined grains
(likely white rice), and individuals of other races
consume less whole grains. Regional differences are
present, with individuals from the Midwest and West
4
B.-H. Lin and S. T. Yen
consuming more whole grains, whilst individuals
from the Midwest, South and West consuming less
refined grains, than those residing in the Northeast.
Urbanization also plays a role, with individuals
residing in rural areas consuming less of both whole
and refined grains, and individuals from suburban
areas consuming less whole grains, than those in the
cities.
Household structure is classified into four
categories – dual- or single-headed with or without
children, with single-person households being the
reference group. Respondents from households with
children (dual or single-headed) consume fewer
servings of whole grains. This is consistent with past
findings that children prefer to consume white bread
(Moutou et al., 1998; Harnack et al., 2003).
Finally, food-label use increases the consumption
of whole grains, and the perceived importance of
grains increases the consumption of both whole and
refined grains. The variables that affect the use of
food labels and/or perceived importance may have
both indirect and direct effects on grain consumption.
For instance, whilst household income does not affect
the consumption of whole or refined grains directly, it
has an indirect effect on whole-grain consumption by
way of the label-use variable. Education and several
other variables are found to affect label use and
perceived importance so they have indirect effects on
grain consumption. Our results are consistent with
the finding that whole-grain consumption increases
with education (Bhargava and Hays, 2004).
The parameter estimates are used to calculate the
marginal effects on the probability as well as
conditional and unconditional levels of whole-grain
consumption, based on Equations 12, 13 and 14.5
For statistical inference, asymptotic SEs for these
marginal effects are calculated with the method
(Rao, 1973, p. 388). Results are presented in Table 3.
Because of the Tobit parameterization, the effects of
each explanatory variable on the probability, conditional level and unconditional level are qualitatively
consistent in signs and significance. However, the
marginal effects allow further quantification of these
effects of the variables. The most notable effects are
seen in the perceived importance of grains.
Specifically, relative to others, individuals who
perceive consumption of enough grains as important
are 36% more likely to consume whole grains and,
conditional on consumption, consume 0.17 more
serving per day than others. The corresponding
The effects of explanatory variables on the probabilities of food label use and importance of grains can be derived from
Pr( yi ¼ 1) ¼ (z0 i) for i ¼ 1, 2. These marginal effects are presented in the appendix.
5
The refined-grain equation is hardly censored so the effects of explanatory variables on the unconditional means are reflected
by the parameter estimates of the equation.
Consumer knowledge, food label use and grain consumption
445
Table 3. Marginal effects of explanatory variables on the probability, conditional level and unconditional level of whole-grain
consumption
Probability
Variable
3
Income/10
HH size
Male
Age 20–30
Age 31–40
Age 41–50
Age 51–60
Black
Asian
Other race
Hispanic
Special diet
Midwest
South
West
Rural
Suburban
HH type 1
HH type 2
HH type 3
Quarter 1
Quarter 2
Quarter 3
Label use
Importance
Conditional level
Unconditional level
Effect
SE
Effect
SE
Effect
SE
0.077
0.063
0.034
0.050***
0.045***
0.054***
0.045***
0.099***
0.220***
0.099*
0.022
0.017
0.034**
0.004
0.106***
0.028**
0.020*
0.045**
0.022*
0.080***
0.012
0.003
0.019
0.138***
0.355***
0.048
0.051
0.014
0.018
0.018
0.017
0.016
0.020
0.048
0.052
0.018
0.013
0.015
0.014
0.015
0.013
0.012
0.022
0.014
0.024
0.014
0.013
0.013
0.051
0.041
0.044
0.036
0.020***
0.028***
0.025***
0.030***
0.025***
0.052***
0.101***
0.051**
0.013
0.010
0.020**
0.002
0.068***
0.016**
0.012*
0.025**
0.013*
0.042***
0.007
0.001
0.011
0.077***
0.174***
0.028
0.029
0.008
0.010
0.010
0.009
0.009
0.009
0.018
0.024
0.011
0.007
0.009
0.008
0.010
0.007
0.007
0.012
0.008
0.012
0.008
0.008
0.008
0.027
0.021
0.063
0.052
0.028***
0.039***
0.035***
0.042***
0.036***
0.074***
0.142***
0.072**
0.019
0.014
0.029**
0.003
0.096***
0.023**
0.017*
0.036**
0.018*
0.060***
0.010
0.002
0.015
0.110***
0.243***
0.039
0.042
0.011
0.014
0.014
0.013
0.012
0.013
0.025
0.034
0.016
0.011
0.013
0.012
0.014
0.011
0.010
0.017
0.011
0.017
0.011
0.011
0.011
0.039
0.028
Note: ***, ** and * indicate significance at the 1%, 5% and 10% levels, respectively.
marginal effect on the unconditional level suggests
that, overall, individuals who perceive consumption
of enough grains as important consume 0.24 more
serving per day than others. The effects of being
Asian are among the other more notable. Relative to
Whites, Asians are about 22% less likely to consume
whole grains and consume 0.10 and 0.14 fewer
serving per day, conditional and unconditional on
consumption, respectively.
V. Concluding Remarks
Responding to mounting evidence of the association
between whole-grain consumption and a reduced risk
of heart problems and other diseases, as well as a link
between whole-grain consumption and body weight
maintenance, the US Government has strongly
promoted consumption of whole grains. However,
Americans tend to over-consume refined grains and
under-consume whole grains. Compared against the
2005 Federal dietary recommendations, an average
American consumed about the right amount of grain
products in 1994–1996 but the balance between
refined and whole grains deviated notably from
the recommended half-and-half pattern. Only 6%
of Americans met the current recommended
whole-grain consumption in 1994–1996.
The food manufacturing sector has quickly
responded to the Federal call for more whole-grain
consumption. In anticipation of the 2005 Dietary
Guidelines and consumer reactions to them, many
companies launched new branded packaged foods
with higher whole-grain contents in 2004 (Buzby
et al., 2005). For example, General Mills
reformulated all its breakfast cereals to whole
grains, Nestle launched a frozen entrée line made
with 100% whole grains, and Sara Lee launched its
Heart Healthy Plus line of fortified 100% wholegrains and multigrain breads. That same year,
ConAgra introduced a new whole-grain flour called
‘Ultragrain White Whole Wheat’. Increased supply
may generate greater demand. Obviously, other
factors, such as price and taste, can influence
consumers’ acceptance of whole grains.
However, the gap between actual and recommended consumption of whole grains is huge.
B.-H. Lin and S. T. Yen
446
Effective nutrition information strategies may play an
important role in closing the gap. The Federal
Government has revamped its Food Guide Pyramid
with MyPyramid (USDA, 2005), which provides
useful tips to incorporate whole grains into our diet.
Findings reported in this article can be used to design
and target educational campaigns for promoting
whole-grain consumption. For example, children are
known to prefer white breads and our results show
that adults from households with children tend to
consume less whole grains. Apparently, adults and
children from the same household eat alike.
Therefore, strategies targeting children must also
address parental preference, and vice versa.
Children’s food choices have been shown to be
influenced by TV commercials (Hastings et al., 2003).
Therefore, nutritional messages appealing to children
during children’s TV-watching hours may be effective
in promoting whole-grain consumption among children and their parents. A similar advertising-style
approach has been used by the US Centers for
Disease Control and Prevention (2004) to promote
physical activity to youth with positive results.
Besides the USDA food guidance system symbolized by MyPyramid, the other major publicly
regulated source of nutrition information is the
nutrition label on food packages, which is regulated
by the US Food and Drug Administration.
Currently, the ‘Nutrition Facts’ panel, which is the
required element of nutrition labelling, does not
include information on whole-grain content. Food
manufacturers may choose to use an optional frontof-package whole-grain claim and, with the increased
emphasis on whole grains, more food products will
likely carry this label. In its absence, however,
consumers must make inferences from the fibre
content listed on the Nutrition Facts panel and/or
the ingredient list, or product name. Because product
names are frequently confusing (Liebman, 1997), this
confusion may subvert the intentions of motivated
consumers. Therefore, changes in the information on
the Nutrition Facts panel may be an important policy
step to increasing whole-grain consumption.
Acknowledgements
Senior authorship is shared. The authors benefited
from insightful discussions with Lung-Fei Lee.
Thanks are also due to Kristin Marcoe and Andrea
Carlson for assistance with data preparation, and to
Jean Buzby, Lisa Mancino, Joanne Guthrie and an
anonymous referee for helpful comments and suggestions. Regular disclaimer applies. Research for this
article was supported by USDA-ERS Cooperative
Agreement 43-3AEM-4-80052. The views in this
article are those of the authors and do not necessarily
reflect the views or policies of the US Department of
Agriculture.
References
Amemiya, T. (1974) Multivariate regression and simultaneous equation models when the dependent variables
are truncated normal, Econometrica, 42, 999–1012.
Berndt, E. K., Hall, B. H., Hall, R. E. and Hausman, J. A.
(1974) Estimation and inference in nonlinear structural
models, Annals of Economic and Social Measurement,
3, 653–65.
Bhargava, A. (2004) Socio-economic and behaviour factors
are predictors of food use in the National Food Stamp
Program Survey, British Journal of Nutrition, 92,
497–506.
Bhargava, A. and Hays, J. (2004) Behavioural variables
and education are predictors of dietary changes in the
women’s health trials: feasibility study in minority
populations, Preventive Medicine, 38, 442–51.
Brown, D. J. and Schrader, L. F. (1990) Cholesterol
information and shell egg consumption, American
Journal of Agricultural Economics, 72, 548–55.
Buzby, J. C., Farah, H. A. and Vocke, G. (2005) Will 2005
be the year of the whole grain? Washington, DC: US
Department of Agriculture, Economic Research
Service, Amber Waves, 3, 12–7.
Catsiapis, G. and Robinson, C. (1982) Sample selection
bias with multiple selection rules: an application to
student aid grants, Journal of Econometrics, 18,
351–68.
Chern, W. S., Loehman, E. T. and Yen, S. T. (1995)
Information, health risk beliefs, and the demand for
fats and oils, Review of Economics and Statistics, 77,
555–64.
Coleman, L. M. and Wilson, M. A. (2002) Consumers’
knowledge and use of fat-modified products, Journal
of Family and Consumer Sciences, 4, 26–33.
Harnack, L., Walters, S. A. and Jacobs Jr, D. R. (2003)
Dietary intake and food sources of whole grains
among US children and adolescents: data from the
1994–96 Continuing Survey of Food Intakes by
Individuals, Journal of the American Dietetic
Association, 103, 1015–9.
Hastings, G., Stead, M., McDermott, L., Forsyth, A.,
MacKintosh, A. M., Rayner, M., Godfrey, C.,
Caraher, M. and Angus, K. (2003) Review of
Research on the Effects of Food Promotion to
Children, a report prepared for the Food Standards
Agency, UK. Center for Social Marketing, The
University of Strathclyde. Available at http://www.
food.gov.uk/healthiereating/promotion/readreview/
(accessed 7 May 2005).
Howard, J. H., Gates, G. E., Ellersieck, M. R. and
Dowdy, R. P. (1998) Investigating relationships
between nutritional knowledge, attitudes, and beliefs,
and dietary adequacy of the elderly, Journal of
Nutrition for the Elderly, 17, 35–54.
Consumer knowledge, food label use and grain consumption
Kaabia, M. B., Angulo, A. M. and Gil, J. M. (2001) Health
information and the demand for meat in Spain,
European Review of Agricultural Economics, 28,
499–517.
Kan, K. and Yen, S. T. (2003) A sample selection model
with endogenous health knowledge: egg consumption
in the USA, in Health, Nutrition and Food Demand
(Eds) K. Rickertsen and W. S. Chern, CABI
Publishing, Cambridge, MA, pp. 91–103.
Kantor, L. S., Variyam, J. N., Allshouse, J. E.,
Putnam, J. J. and Lin, B. (2001) Choose a variety of
grains daily, especially whole grains: a challenge to
consumers, Journal of Nutrition, 131, 473S–86S.
Keane, M. and Moffitt, R. (1998) A structural model of
multiple welfare program participation and labor
supply, International Economic Review, 39, 553–89.
Kim, S. R. and Chern, W. S. (1999) Alternative measures of
health information and demand for fats and oils in
Japan, Journal of Consumer Affairs, 33, 92–109.
Kim, S., Nayga Jr, R. M. and Capps Jr, O. (2000) The
effect of food label use on nutrient intakes: an
endogenous switching regression analysis, Journal of
Agricultural and Resource Economics, 25, 215–31.
Kinnucan, H. W., Xiao, H., Hsia, C. J. and Jackson, J. D.
(1997) Effects of health information and generic
advertising on US meat demand, American Journal of
Agricultural Economics, 79, 13–23.
Kotz, S., Johnson, N. L. and Balalrishnan, N. (2000)
Continuous Multivariate Distributions, Vol. 1: Models
and Applications, 2nd edn, John Wiley & Sons,
New York.
Kreuter, M. W. and Brennan, L. K. (1997) Do nutrition
label readers eat healthier diets: behavioural correlates
of adults’ use of food labels, American Journal of
Preventive Medicine, 13, 277–83.
Liebman, B. (1997) The whole grain guide, Nutrition Action
Health Letter. March – US Edition. Center for Science
in the Public Interest, Washington, DC. Available at
http://www.cspinet.org/nah/wwheat.html (accessed 18
May 2005).
Lin, B., Variyam, J. N., Allshouse, J. and Cromartie, J.
(2003) Food and Agricultural Commodity Consumption
in the United States: Looking Ahead to 2020. US
Department of Agriculture, Washington, DC.
Economic Research Service, Food and Rural
Economics Division, Agricultural Economics Report
No. 820, February.
Moutou, C., Brewster, G. W. and Fox, J. A. (1998) US
consumers’ socioeconomic characteristics and consumption of grain-based foods, Agribusiness, 14,
63–72.
Neuhouser, M. L., Kristal, A. R. and Patterson, R. E.
(1999) Use of food nutrition labels is associated with
lower fat intake, American Journal of Dietetic
Association, 99, 45–53.
Ramezani, C. A. and Roeder, C. (1995) Health knowledge
and nutrition adequacy of female heads of households
in the United States, Journal of Consumer Affairs, 29,
381–402.
447
Rao, C. R. (1973) Linear Statistical Inference and its
Applications, John Wiley & Sons, New York.
Tunali, I. (1986) A general structure for models of doubleselection and an application to a joint migrationearnings process with remigration, Research in Labor
Economics, 8, 235–84.
US Centers for Disease Control and Prevention (2004)
National campaign to get kids physically active is
working: survey findings prove the VERB campaign
is motivating youth to get active, CDC Press release,
February 17. Available at http://www.cdc.gov/od/
oc/media/pressrel/r040217.htm (accessed 18 May
2005).
US Department of Agriculture (2000) Continuing Survey of
Food Intakes by Individuals 1994–96 and 1998, CDROM, Agricultural Research Service, Washington,
DC.
US Department of Agriculture (2005) MyPyramid.gov:
Steps to a Healthier You. Available at http://www.mypyramid.gov/ (accessed 6 May 2005).
US Department of Agriculture and US Department of
Health and Human Services (2005) Dietary Guidelines
for
Americans,
2005.
Available
at
http://
www.healthierus.gov/dietaryguidelines/ (accessed 6
May 2005).
US Department of Health and Human Services. (2005)
Healthy People 2010. Available at http://healthypeople.gov/(accessed 6 May 2005).
Variyam, J. N., Blaylock, J. and Smallwood, D. M. (1996)
Modelling nutrition knowledge, attitudes, and dietdisease awareness: the case of dietary fibre, Statistics in
Medicine, 15, 23–35.
Variyam, J. N., Blaylock, J. R. and Smallwood, D. M.
(1997) Diet-Health Knowledge and Nutrition: The
Intake of Dietary Fats and Cholesterol, US
Department of Agriculture, Washington, DC.
Economic Research Service, Technical Bulletin
No. 1855.
Variyam, J. N., Blaylock, J. and Smallwood, D. (1999a)
Information, endogeneity, and consumer health behaviour: application to dietary intakes, Applied
Economics, 31, 217–26.
Variyam, J. N., Blaylock, J., Lin, B., Ralston, K. and
Smallwood, D. (1999b) Mother’s nutrition knowledge
and children’s dietary intakes, American Journal of
Agricultural Economics, 81, 373–84.
Wales, T. J. and Woodland, A. D. (1983) Estimation of
consumer demand systems with binding nonnegativity
constraints, Journal of Econometrics, 21, 263–85.
Yen, S. T., Jensen, H. H. and Wang, Q. (1996) Cholesterol
information and egg consumption in the US: a
nonnormal and heteroscedastic double-hurdle model,
European Review of Agricultural Economics, 23,
343–56.
Yen, S. T., Lin, B. and Davis, C. (2006)
Consumer knowledge and meat consumption at
home and away from home. Unpublished manuscript,
October.
B.-H. Lin and S. T. Yen
448
Appendix
Table A1. Marginal effects of explanatory variables on the probability of food-label use and perceived importance
of grain consumption
Label use
Variable
3
Income/10
Male
Age 20–30
Age 31–40
Age 41–50
Age 51–60
Black
Asian
Other race
Hispanic
Special diet
High school
Some college
College
Meal planner
Exercise
Smoker
Pessimistic
Importance of grains
Effect
SE
Effect
SE
0.187***
0.183***
0.023
0.009
0.028
0.043**
0.003
0.064
0.092
0.044*
0.085***
0.064***
0.072***
0.120***
0.029*
0.048***
0.106***
0.059***
0.056
0.016
0.024
0.021
0.021
0.020
0.022
0.056
0.063
0.025
0.015
0.019
0.021
0.022
0.017
0.014
0.017
0.014
0.064
0.003
0.043***
0.036**
0.036**
0.020
0.067***
0.039
0.073
0.022
0.015
0.020
0.037**
0.103***
0.041***
0.040***
0.051***
0.018*
0.047
0.013
0.017
0.016
0.016
0.016
0.019
0.049
0.047
0.021
0.012
0.014
0.016
0.015
0.014
0.010
0.013
0.011
Note: ***, ** and * indicate significance at the 1%, 5% and 10% levels, respectively.