Sie sind auf Seite 1von 16

Determinants of US fast food

consumption 1994-1998
Jasper Fanning
Department of Agricultural Economics,
Kansas State University, Manhattan, Kansas, USA
Thomas Marsh
School of Economic Sciences, Washington State University,
Pullman, Washington, USA, and
Kyle Stiegert
Food System Research Group, University of Wisconsin-Madison,
Fitchburg, Wisconsin, USA
Abstract
Purpose Fast food (FF) consumption increased dramatically through the 1990s in the USA,
accounting for nearly 35.5 percent of total away-from-home expenditures in 1999. Given dramatic
changes in food consumption, and heightened public concern about health and obesity, there is a
considerable need for research to understand better the factors affecting US FF consumption. This
paper aims to ll this gap.
Design/methodology/approach In this paper, logistic regression is applied to analyze the
socioeconomic and demographic factors inuencing the likelihood of consuming FF using United
States Department of Agriculture data from the Continuing Survey of Food Intakes by Individuals from
1994 to 1996 and the Supplemental Childrens Survey of 1998.
Findings In general, the expected likelihood of FF consumption increases until around 20-30 years
of age and then decreases; increases as household income grows until about $50,000-60,000 and then
decreases; and decreases as household size grows. Further, males from the Midwest and South regions
that live outside central cities in Metropolitan Statistical Areas have the highest likelihood of
consuming FF.
Originality/value While much literature has addressed key questions about expenditure on food
away from home, this study complements previous work by focusing on food items consumed from FF
facilities in the 1990s. In addition, the results nd highly signicant and important (statistically and
economically) interactions between the likelihood of FF consumption and age, income, and
household size.
Keywords Consumption, Fast foods, United States of America
Paper type Research paper
Introduction
Food away from home (including fast food (FF)) maintained a large and growing
component of US food expenditure through the 1990s. As a share of disposable personal
income, food away fromhome expenditure rose from3.6 percent in 1970 to 4.0 percent in
2000, while food at home (FAH) decreased from 10.3 percent in 1970 to 5.8 percent in
2000. Moreover, in 1967, FF accounted for 14.3 percent of total away-from-home
expenditures and by 1999 it reached 35.5 percent (US Department of
Agriculture/Economic Research Service). The emergence of the FF marketing
platform is commonly considered a leading force in the increased incidence of obesity
The current issue and full text archive of this journal is available at
www.emeraldinsight.com/0007-070X.htm
Determinants
of US fast food
consumption
5
British Food Journal
Vol. 112 No. 1, 2010
pp. 5-20
qEmerald Group Publishing Limited
0007-070X
DOI 10.1108/00070701011011164
and diabetes in children. Bowman et al. (2004, p. 113-4) analyzed food consumption
patterns in 6,212 children in the USA and stated:
Fast food consumers ate more total fat, more saturated fat, more total carbohydrates, more
added sugars, more sugar sweetened beverages, less uid milk, and fewer fruits and
non-starchy vegetables, differences that were statistically signicant in most age categories.
They also suggested that FF consuming children were nearly twice as likely to become
obese compared to those that did not consume FF. Given dramatic changes in food
consumption patterns, as well as general public concern about health, obesity, food
quality (Lin and Franzao, 1999; Jekanowski, 1999; Kuchler and Lin, 2002; Schlosser,
2002; Piggott and Marsh, 2004; Marsh et al., 2004), particularly among children, it is
important to identify the potentially complex relationships among socioeconomic
factors affecting FF consumption.
While much literature has addressed key questions about expenditure on food away
from home (Byrne et al., 1996 or Senauer, 1979), our study complements previous work
by focusing on food items consumed from FF facilities in the 1990s[1]. McCracken and
Brandt (1987) examined the factors inuencing expenditures on food away from home,
at restaurants, FF establishments, and other commercial facilities using United States
Department of Agricultures (USDA) 1977-1978 Nationwide Food Consumption Survey.
They found household income (except at FF facilities), time value, size, and composition
to be predominately signicant factors. Byrne et al. (1998) examined the National Panel
Diary survey data covering the period from 1982 to 1989, focusing on quick-serve,
mid-scale and up-scale commercial food service facilities. The study reported signicant
income, household size, regional, gender, and urban impacts on food away from home
participation and expenditure. Jekanowski et al. (2001) investigated the effect of price,
income, and demographic characteristics on FF. Using aggregated FF measures at the
Metropolitan Statistical Area (MSA) level, they suggested that growth in the FF
consumption was related to an increasing supply of convenience. In other words,
consumers have increased consumption of FFbecause it provided the incentives to do so
with respect to price, time, and taste. Jekanowski (1999) suggested that the limited menu
aspect of most major chains means that their growth can have a potentially large impact
on a selected segment of the agricultural marketing system. Moreover, any menu
changes by a major rm can have enormous and almost immediate effects on particular
agricultural industries. Lin et al. (2001) examined the inuence of FF growth on frozen
potato consumption. The authors reported that 13 percent of consumers eat French fries
at FF establishments each day accounting for 67 percent of the French fry market. They
also reported that French fry consumption varies by age, region, urbanization, race, and
ethnicity, but independent of income. Steward et al. (2004, 2005) examined determinants
of food expenditure data from the Bureau of Labor Statistics. This study found
signicant associations between FF expenditure and education, household size, income,
age, work hours, and race. In all, these studies provide valuable insight into expenditure
on FF, other food away from home (OFAFH), and FAH that is useful in better
understanding issues about structure of the food industry and distribution sectors,
nutritional intake, and demand for goods at different marketing levels.
The main objective of the current study is to identify and quantify several empirical
relationships between an individual consumers socioeconomic and demographic
characteristics and their likelihood to consume food items from FF facilities in the USA
BFJ
112,1
6
during the 1990s. We specify and t two models using individual survey data from1994
to 1998, which were collected and maintained by the USDA. A binary logit regression
model is used to analyze consumption patterns on a per day basis, and a multinomial
logit regression model is used on a per eating occasion basis. The results are both
comparable to ndings fromprevious studies and can be used in the future as a baseline
to evaluate results fromnewer data. For instance, we drawfromthe same data source as
that of Bowman et al. (2004). The focus of their study was principallyonconsumers inthe
four to 19 age brackets with emphasis on group differential effects of FF consumption,
on energy intake and diet quality. In the current study, we are particularly interested in
what the data reveals about the complex nonlinear interactions of socioeconomic factors
to explain consumption patterns across all ages. We also provide information that
compares results across modal eating experiences (FF, FAH, and OFAFH), by timing
of consumption, and across the binary and multinomial logit models to offer a
robustness check of model results.
The remainder of the paper proceeds in the following manner. First a random utility
model is specied for a consumer making choices among multiple foods to motivate
specication of the empirical models. Second, we describe in more detail the data used
in the analysis and some empirical issues. Third, we specify and estimate empirical
relationships between FF consumption and socioeconomic and demographic variables
on both a daily and eating occasion basis. Fourth, model results are discussed. Fifth,
concluding comments are provided.
Random utility model
Suppose that a decision maker faces multiple choices such as FF or not FF.
Alternatively, the decision maker could choose among FF, OFAFH, and FAH. In either
case an unordered choice model can be motivated by a random utility framework
(Greene, 2003). The random utility function motivates each choice, which can be
represented as:
U
*
ij
V
ij
1
ij
1
where:
V
ij
deterministic utility function.
1
ij
unobserved random variable.
Subscript i (i 1, . . . , N) the individual.
Subscript j ( j 1, . . . , J ) the food of choice among the J foods.
The deterministic part of the utility function is dened as V
ij
X
i
b
j
, where X
i
is a
1 K matrix of individual is characteristics (dened ahead) and b
j
is the K 1
vector of parameters to be estimated for j 1, . . . , J. We assume the decision maker
has a higher likelihood of choosing food choice j over all other food types if and only if
U
*
ij
is the maximum from among the J utilities or:
U
*
ij
max U
*
i1
; . . . ; U
*
iJ
_ _
, Prob U
*
ij
. U
*
ik
_ _
;k j:
The empirical analogue to the random utility model of choice is a statistical probabilistic
choice model. The theoretical link between randomutility maximization and probabilistic
Determinants
of US fast food
consumption
7
multinomial logit regression model has been developed by McFadden (1981) and others
(McFadden and Train, 2000; Maddala, 1982). Fromthe multinomial model, the probability
that individual i with characteristics X
i
will choose food choice j is:
P
ij

e
V
ij

J
l0
e
V
il
_ _ 2
where we impose the normalization that b
0
0. The marginal effect of change invariable
X
k
on the probability P
ij
of the multinomial logit regression model is:
P
ij
X
k

e
V
ij
=

J
l0
e
V
ij
_ _ _ _
V
ij
V
ij
X
k
: 3
In the empirical relationships below, we rst investigate the determinants of anindividual
who is choosing between consuming FFor non-FF( j 1, 2 J ) on a daily basis. Second,
we assume an individual considers a food choice decision between FAH, FF, and OFAFH
with ( j 1, 2, 3 J) on a per eating occasion basis. Both relationships provide valuable
insight into the inuence of demographic and socioeconomic variables on FF.
Data and empirical issues
The data used in this study are fromthe continuing survey of food intakes by individuals
(CSFII) of 1994-1996 and the supplemental childrens survey (SCS) of 1998. This survey
conducted by the USDA measures individual food consumption. The goal of the
1994-1996 survey is to obtain a nationally representative sample of noninstitutionalized
persons residing in households in the USA. The goal of the SCS of 1998 was to obtain
nationally representative samples of noninstitutionalized persons nine years of age or
younger residing in the USA. In both surveys individuals were interviewed on two
nonconsecutive days during which they were asked to recall their food intake in the
previous 24 hours. Along with the dietary information, specic demographic information
was collected such as age, income, race, and gender. Design and operation of the survey,
including discussion of sample design, is provided in data documentation.
Observations recorded food intake amounts per individual for day 1 or 2. In other
words, each individual food item consumed on day 1 or 2 of the survey was recorded as
a separate observation. We restricted the empirical analysis of the current study to day
1, which is statistically appropriate when used with sample weights provided as part of
the data base.[2] For the per day analysis, food consumption was aggregated to a per
day basis for each individual, yielding a nal data set with 21,662 observations. For the
per eating occasion analysis, the nal data set contained 135,277 observations. FF
items were identied as food from FF establishments as dened in the USDA
survey[3]. Table I presents variable denitions and descriptive statistics of the data on
a daily basis.
The empirical analysis for this study progressed in several steps. The rst step
analyzes the likelihood of consuming FF (or not) on a daily basis using a binary logit
regression. The second step extends that of the rst by examining the likelihood of
consuming FAH, FF, or OFAFH (restaurant, vending, etc.) on a given eating occasion
using multinomial logit regression[4]. Examining FF consumption on a daily basis
provides an empirical relationship with which most applied analysts are familiar and
BFJ
112,1
8
V
a
r
i
a
b
l
e
D
e

n
i
t
i
o
n
M
e
a
n
S
D
M
i
n
i
m
u
m
M
a
x
i
m
u
m
F
F
1
i
f
c
o
n
s
u
m
e
d
F
F
,
e
l
s
e
0
0
.
2
3
0
.
4
2
0
1
H
o
u
s
e
h
o
l
d
s
i
z
e
C
o
u
n
t
o
f
h
o
u
s
e
h
o
l
d
m
e
m
b
e
r
s
3
.
7
6
1
.
7
7
1
1
6
I
n
c
o
m
e
A
p
p
r
o
x
i
m
a
t
e
h
o
u
s
e
h
o
l
d
i
n
c
o
m
e
3
8
,
6
2
9
.
4
7
2
7
,
2
2
1
.
6
7
0
1
0
0
,
0
0
0
A
g
e
A
g
e
o
f
h
o
u
s
e
h
o
l
d
m
e
m
b
e
r
2
5
.
3
9
2
5
.
4
7
0
9
0
R
e
g
i
o
n
1
1
i
f
N
o
r
t
h
e
a
s
t
,
e
l
s
e
0
0
.
1
8
0
.
3
8
0
1
R
e
g
i
o
n
2
1
i
f
M
i
d
w
e
s
t
,
e
l
s
e
0
0
.
2
3
0
.
4
2
0
1
R
e
g
i
o
n
3
1
i
f
S
o
u
t
h
,
e
l
s
e
0
0
.
3
5
0
.
4
8
0
1
R
e
g
i
o
n
4
1
i
f
W
e
s
t
,
e
l
s
e
0
0
.
2
4
0
.
4
3
0
1
U
r
b
a
n
1
1
i
f
M
S
A
/
c
e
n
t
r
a
l
c
i
t
y
,
e
l
s
e
0
0
.
3
0
0
.
4
6
0
1
U
r
b
a
n
2
1
i
f
M
S
A
/
o
u
t
s
i
d
e
c
e
n
t
r
a
l
,
e
l
s
e
0
0
.
4
6
0
.
5
0
0
1
U
r
b
a
n
3
1
i
f
N
o
n
-
M
S
A
,
e
l
s
e
0
0
.
2
3
0
.
4
2
0
1
S
e
x
1
1
i
f
M
a
l
e
,
e
l
s
e
0
0
.
5
1
0
.
5
0
0
1
S
e
x
2
1
i
f
F
e
m
a
l
e
,
e
l
s
e
0
0
.
4
9
0
.
5
0
0
1
R
a
c
e
1
1
i
f
w
h
i
t
e
,
e
l
s
e
0
0
.
7
5
0
.
4
3
0
1
R
a
c
e
2
1
i
f
b
l
a
c
k
,
e
l
s
e
0
0
.
1
4
0
.
3
4
0
1
R
a
c
e
3
1
i
f
A
s
i
a
n
/
p
a
c
i

c
i
s
l
a
n
d
e
r
,
e
l
s
e
0
0
.
0
3
0
.
1
6
0
1
R
a
c
e
4
1
i
f
A
m
e
r
i
c
a
n
I
n
d
i
a
n
/
A
l
a
s
k
a
n
n
a
t
i
v
e
,
e
l
s
e
0
0
.
0
1
0
.
0
9
0
1
R
a
c
e
5
1
i
f
O
t
h
e
r
,
e
l
s
e
0
0
.
0
8
0
.
2
7
0
1
O
r
i
g
i
n
1
1
i
f
M
e
x
i
c
a
n
/
M
e
x
i
c
a
n
A
m
e
r
i
c
a
n
/
C
h
i
c
a
n
o
,
e
l
s
e
0
0
.
0
7
0
.
2
5
0
1
O
r
i
g
i
n
2
1
i
f
P
u
e
r
t
o
R
i
c
a
n
,
e
l
s
e
0
0
.
0
1
0
.
1
1
0
1
O
r
i
g
i
n
3
1
i
f
C
u
b
a
n
,
e
l
s
e
0
0
.
0
0
0
.
0
5
0
1
O
r
i
g
i
n
4
1
i
f
O
t
h
e
r
S
p
a
n
i
s
h
/
H
i
s
p
a
n
i
c
,
e
l
s
e
0
0
.
0
5
0
.
2
2
0
1
O
r
i
g
i
n
5
1
i
f
N
o
n
e
o
f
t
h
e
a
b
o
v
e
,
e
l
s
e
0
0
.
8
7
0
.
3
4
0
1
Y
e
a
r
1
1
i
f
1
9
9
4
,
e
l
s
e
0
0
.
2
6
0
.
4
4
0
1
Y
e
a
r
2
1
i
f
1
9
9
5
,
e
l
s
e
0
0
.
2
5
0
.
4
3
0
1
Y
e
a
r
3
1
i
f
1
9
9
6
,
e
l
s
e
0
0
.
2
4
0
.
4
3
0
1
Y
e
a
r
4
1
i
f
1
9
9
7
,
e
l
s
e
0
0
.
2
6
0
.
4
4
0
1
W
t
4
_
d
a
y
1
S
a
m
p
l
e
w
e
i
g
h
t
f
o
r
d
a
y
1
i
n
t
a
k
e
1
2
,
0
8
9
.
2
4
1
1
,
0
5
0
.
5
3
3
4
0
2
2
6
,
6
9
2
N
o
t
e
s
:
R
a
w
d
a
t
a
f
o
r
d
a
y
1
c
o
n
s
u
m
p
t
i
o
n
w
i
t
h
t
o
t
a
l
n
u
m
b
e
r
o
f
o
b
s
e
r
v
a
t
i
o
n
s
n

2
1
,
6
6
2
;
M
S
A
-
A
g
e
o
g
r
a
p
h
i
c
a
r
e
a
c
o
n
s
i
s
t
i
n
g
o
f
a
l
a
r
g
e
p
o
p
u
l
a
t
i
o
n
n
u
c
l
e
u
s
t
o
g
e
t
h
e
r
w
i
t
h
a
d
j
a
c
e
n
t
c
o
m
m
u
n
i
t
i
e
s
t
h
a
t
h
a
v
e
a
h
i
g
h
d
e
g
r
e
e
o
f
e
c
o
n
o
m
i
c
a
n
d
s
o
c
i
a
l
i
n
t
e
g
r
a
t
i
o
n
w
i
t
h
t
h
a
t
n
u
c
l
e
u
s
;
d
e

n
e
d
b
y
t
h
e
F
e
d
e
r
a
l
O
f

c
e
o
f
M
a
n
a
g
e
m
e
n
t
a
n
d
B
u
d
g
e
t
f
o
r
u
s
e
i
n
t
h
e
p
r
e
s
e
n
t
a
t
i
o
n
o
f
s
t
a
t
i
s
t
i
c
s
b
y
a
g
e
n
c
i
e
s
o
f
t
h
e
F
e
d
e
r
a
l
G
o
v
e
r
n
m
e
n
t
;
r
e
g
i
o
n
s
a
s
d
e

n
e
d
b
y
t
h
e
1
9
9
0
C
e
n
s
u
s
o
f
P
o
p
u
l
a
t
i
o
n
:
N
o
r
t
h
e
a
s
t
-
C
o
n
n
e
c
t
i
c
u
t
,
M
a
i
n
e
,
M
a
s
s
a
c
h
u
s
e
t
t
s
,
N
e
w
H
a
m
p
s
h
i
r
e
,
N
e
w
J
e
r
s
e
y
,
N
e
w
Y
o
r
k
,
P
e
n
n
s
y
l
v
a
n
i
a
,
R
h
o
d
e
I
s
l
a
n
d
,
V
e
r
m
o
n
t
;
M
i
d
w
e
s
t
-
I
l
l
i
n
o
i
s
,
I
n
d
i
a
n
a
,
I
o
w
a
,
K
a
n
s
a
s
,
M
i
c
h
i
g
a
n
,
M
i
n
n
e
s
o
t
a
,
M
i
s
s
o
u
r
i
,
N
e
b
r
a
s
k
a
,
N
o
r
t
h
D
a
k
o
t
a
,
O
h
i
o
,
S
o
u
t
h
D
a
k
o
t
a
,
W
i
s
c
o
n
s
i
n
;
S
o
u
t
h
-
A
l
a
b
a
m
a
,
A
r
k
a
n
s
a
s
,
D
e
l
a
w
a
r
e
,
D
i
s
t
r
i
c
t
o
f
C
o
l
u
m
b
i
a
,
F
l
o
r
i
d
a
,
G
e
o
r
g
i
a
,
K
e
n
t
u
c
k
y
,
L
o
u
i
s
i
a
n
a
,
M
a
r
y
l
a
n
d
,
M
i
s
s
i
s
s
i
p
p
i
,
N
o
r
t
h
C
a
r
o
l
i
n
a
,
O
k
l
a
h
o
m
a
,
S
o
u
t
h
C
a
r
o
l
i
n
a
,
T
e
n
n
e
s
s
e
e
,
T
e
x
a
s
,
V
i
r
g
i
n
i
a
,
W
e
s
t
V
i
r
g
i
n
i
a
;
W
e
s
t
-
A
l
a
s
k
a
,
A
r
i
z
o
n
a
,
C
a
l
i
f
o
r
n
i
a
,
C
o
l
o
r
a
d
o
,
H
a
w
a
i
i
,
I
d
a
h
o
,
M
o
n
t
a
n
a
,
N
e
v
a
d
a
,
N
e
w
M
e
x
i
c
o
,
O
r
e
g
o
n
,
U
t
a
h
,
W
a
s
h
i
n
g
t
o
n
,
W
y
o
m
i
n
g
Table I.
Variable denitions and
descriptive statistics on a
per day basis
Determinants
of US fast food
consumption
9
comparison to descriptive statistics on a per day basis reported in previous studies.
Examining FF consumption per eating occasion in relation to FAH and OFAFH
extends the initial analysis to incorporate tradeoffs between the alternative food types
and to capture temporal or seasonal effects. It also provides a robustness check by
comparing empirical results across the binary and multinomial logit regression
models.
The USDA surveys used a complex multistage sampling procedure to provide
representative samples in the 50 states and Washington, DC. As a result, the binary
logit model is estimated with a weighted regression using PROC LOGISTIC in SAS.
The multinomial logit model is estimated as a weighted regression using PROC
LOGISTIC in SAS with the generalized logit (GLOGIT) option[5]. Sample weights are
needed to appropriately estimated standard errors of parameters. Statistical estimates
and goodness of t statistics are available from authors upon request.
Binary logit per day basis
We conceptualize an empirical relationship between an individuals discrete decision to
consume food items at a FF establishment (or not) and characteristics of those
individuals. Let U
i1
1 if an individual consumes FF and U
i2
0 if the individual
does not consume FF, where the subscript i represents the ith observation of the data
set. Dene X
i
to be a 1 K vector of socioeconomic and demographic factors for the
ith observation. As discussed above, an appropriate regression technique for a model
with a discrete (U
ij
1 or 0) dependent variable is binary logit regression (Judge et al.,
1988; Greene, 2003). Formulating a binary logit regression model yields the predicted
probability:
ProbU
i1
1jX
i
1 e
2X
i
^
b

21
4
where
^
b is the estimated parameter vector. In equation (4), the probability an
individual consumes FF conditional on the individuals characteristics, X
i
, is denoted
by ProbU
i1
1jX
i
. Yen (1993), Yen and Jones (1997), and McCracken and Brandt
(1987) have used similar approaches to empirically model an individuals decision of
whether or not to consume a specic good.
Marginal effects of the logit regression model, which provide the change in
ProbU
i1
1jX
i
given a change in the kth individual characteristic, X
ik
, are a
special case of equation (3), and can be expressed as:
ProbU
i1
1jX
i

X
ik

1 e
2X
i
^
b
_ _
21
X
i
b
_

_
_

_
X
i
b
X
ik
_ _
: 5
Estimated elasticities based on equation (5) are calculated using parameter estimates,
^
b, from the logit regression model and the mean values of selected socioeconomic and
demographic variables, X
i
.
Model variables
The selection of model variables is motivated by economic theory, previous research
(discussed above), and their potential to provide insights to food policy issues.
BFJ
112,1
10
Socioeconomic and demographic variables included regions of the country, urbanization
status, household size, income, age, gender, race, origin, and time variables represented
in each year of the survey (Table I). Economic theory links food consumption in the
household to income, age, household size, and other socio-demographic characteristics
(see, for example, discussion of household production theory in Deaton and Muellbauer,
1980). McCracken and Brandt (1987), Byrne et al. (1996, 1998) and others use similar
economic justication for variable selection. While these studies have reported
statistically signicant effects on FF expenditure due to age and household size, the
effects of income have not been consistent across studies.
To complete the model specication, the functional formof the explanatory variables
is specied to be a multivariate polynomial. Based on the literature review and
preliminary investigations of the data, it is hypothesized that a second order
multivariate polynomial maintained sufcient exibility to capture probability
responses from the explanatory variables. First order interaction terms are included
to test for signicance of the relative inuence of each variable on one another. The
structure of X
i
b is dened as:
X
i
b b
0

K
j1
b
i
X
ij

K
j;k1;j.k
a
1jk
X
ij
X
ik

3
j1
b
2j
X
2
ij
6
where the continuous variables are X
i1
age (years), X
i2
income ($US), and X
i3
household
size[6]. The time and the discrete variables (region, urban, gender, race, and origin, and
urban) enter into the linear shift and interaction terms of the regression equation, but not
in the quadratic terms[7].
Results per day basis
The logit regression model contained 167 variables made up of an intercept term as
well as linear, quadratic, and interaction terms that are dened in equation (6). The
data are aggregated to a daily basis providing 21,662 observations for day 1. The null
hypothesis that H
0
: b 0 is rejected for each model with a p-value less than 0.0001
for the likelihood ratio, score, and Wald tests. The predicted and observed probabilities
are 64.5 percent (34.1 percent) concordant (discordant), indicating an adequate
goodness-of-t. The regression models are formulated so that the representative
individual (RI) is a white male not of Spanish/Hispanic origin living in a MSAs, central
city in the Midwest in 1994.
Elasticities
Elasticity values are calculated to measure the percent change in probability that an
individual will consume FF given a 1 percent change in income, age, or household size.
At the means of these variables, elasticity estimates for age, household size, and
income were calculated to be 20.33, 20.17, and 0.07. More specically a, 1 percent
increase in age will yield a 0.33 percent decrease in the probability an individual will
consume FF on a given day. Meanwhile, a 1 percent increase in household size will
yield a 0.17 percent decrease in the probability an individual will consume FF on
a given day. Finally, a 1 percent increase in income will yield a 0.07 percent increase in
the probability an individual will consume FF on a given day. McCracken and Brandt
(1987) reported an income elasticity of 0.04 and size elasticity of 0.26 for limited-menu
Determinants
of US fast food
consumption
11
expenditures. Byrne et al. (1998) also reported low income elasticity values for food
away from home and a similar range for size elasticities.
Predicted probability curves
Elasticity estimates are limited in that they provide information at only the value of
selected observations. For a different, and perhaps improved perspective, we also
present predicted probability curves from model results. These gures show the
probability that individuals consume FF on a particular day across age, household size,
and income levels by gender, urbanization, region, and origin. While our reported
results and subsequent discussion are based on an entire set of generated predicted
probability curves, to save space, we present only a select set of these curves.
Age, household size, and income. Figure 1 shows the predicted probability curves
that the RI consumes FF across different age levels. The likelihood of FF consumption
increases until 20-30 years of age and decreases thereafter. Figure 2 provides the
predicted probability curves that the RI consumes FF across household size. The
likelihood of FF decreases for larger households. Figure 3 shows the predicted
probability curves that the RI consumes FF across different household income levels.
The predicted probability curve increases as household income grows until about
$50,000-60,000 and then decreases thereafter.
Other variables. Here, we briey highlight the inuence of gender, urbanization,
region, and origin on the likelihood to consume FFs. Figures 1-3 show the predicted
probability curves that individuals will consume FF across age, household size, and
income levels by gender (male or female). The probability of FF consumption increases
to a maximum of 0.37 (0.30) for a male (female) at 30 (20) years of age (see Figure 1).
Figure 2 shows the likelihood of FF consumption across household size. Again, males
have a higher likelihood of consuming FFs relative to females. In Figure 3, the
likelihood of FF consumption increases to a maximum of the highest 0.37 for males and
0.28 for females for individuals with income levels ranging from $50,000 to $60,000.
Figure 1.
Predicted probabilities
of FF consumption across
age by gender
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0 10 20 30 40 50 60 70 80 90
Age
Male Female
BFJ
112,1
12
Figure 4 shows the impacts from the region an individual resides on the likelihood of
FF consumption. Across different age levels, individuals from the South and Midwest
have the highest likelihood of FF consumption followed by the Northeast and West
regions.
Further insights drawn from model results (but not presented here) indicate that the
probability of FF consumption increases for individuals in the South with more
Figure 2.
Predicted probabilities
of FF consumption across
household size by gender
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
1 3 5 7 9 11 13 15
Household size
Male Female
Figure 3.
Predicted probabilities
of FF consumption across
income by gender
0.2
0.22
0.24
0.26
0.28
0.3
0.32
0.34
0.36
0.38
0 10,000 20,000 30,000 40,000 50,000 60,000 70,000 80,000 90,000 1,00,000
Income
Male Female
Determinants
of US fast food
consumption
13
household members. In contrast, it decreases as households grow larger for the other
regions. Across income levels, individuals from the South and Midwest have higher
likelihoods of consuming FFs. Individuals outside central cities in MSA areas have the
highest likelihood of FF consumption, followed by those in non-MSA and central city
MSA areas.
Race and origin also have important impacts on the likelihood of FF consumption.
Blacks and then Asians have higher probability of consuming FFs across age levels
than do Whites. Across income levels, Asians and Blacks have higher likelihood of
consuming FFs than do Whites. Cuban, Mexican, and Puerto Ricans have higher
probability of consuming FFs across age, household size, and income.
Multinomial logit and results per eating occasion
Next, we consider modal eating experiences wherein there is a choice between FF,
FAH, and OFAFH. Let U
i1
1 if an individual consumes FF, U
i2
2 if an individual
consumes OFAFH and U
i3
3 if an individual consumes FAH, where the subscript i
represents the ith observation of the data set. The probability functions and marginal
effects are dened in equations (2) and (3), respectively. The income, household size,
and age variables X
i
are assumed to be second order with linear interaction terms as in
equation (6). The remaining socioeconomic and demographic variables are
incorporated as rst-order linear terms without interaction [regions of country,
urbanization status, gender, pregnancy, race, origin, rent, education, and temporal
variables (year, month, day of week, eating occasion)].
The multinomial logit regression model contained 51 variables made up of an
intercept term as well as linear, quadratic, and interaction terms (totaling 102 estimated
parameters for both the FF and OFAFH equations). Using day 1 data on a per eating
occasion basis yielded 135,277 observations. The null hypothesis that H
0
: b 0 is
rejected for each model with a p-value less than 0.0001 for the likelihood ratio, score,
and Wald tests. The regression models were formulated so that the RI is a white male
not of Spanish/Hispanic origin in the Midwest in 1994 but not living in a suburban
Figure 4.
Predicted probabilities
of FF consumption across
age by region
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0 10 20 30 40 50 60 70 80 90
Age
Midwest
Northeast
South
West
BFJ
112,1
14
area. The RI was also categorized as consuming breakfast on Sunday in January, not
being on any diet, and owning his home.
Predicted probability curves
In each gure below we present the predicted likelihood of consumption FF, OFAFH,
and FAH. This allows comparison to the previous gures generated from the logit
regression model and insight into the interactions between FF, OFAFH, and FAH.
Age. Figure 5 shows probability curves of FF, OFAFH, and FAH over age. The
likelihood of FF and OFAFH consumption increases until around middle age and then
decreases thereafter[8]. The likelihood of OFAFH is slightly larger than that of FF
across the age groups. In contrast the likelihood of FAH consumption decreases for
younger individuals, but increases for individuals past middle aged. Referring back to
Figure 1, the prole of the predicted likelihood for FF consumption across age is
consistent between the two sets of regression results.
Household size. Figure 6 shows probability curves of FF, OFAFH, and FAH across
household size. The likelihood of FF consumption decreases as household size
increases, which is consistent with the logit regression results in Figure 2. The
likelihood of OFAFH consumption initially decreases until a household has about eight
members then starts to increase, while the likelihood of FAH increases and then
decreases after about eight members. The increase (decrease) of OFAFH (FAH) may
reect reliance on public providers of food for the largest families, but only further
analysis would either support or reject the conjecture.
Income. Figure 7 shows probability curves of FF, OFAFH, and FAH across income
levels. The likelihood of FF consumption increases until the $50,000-70,000 range and
then decreases. This is very similar to the results from the logit regression model in
Figure 3, which predicted an increase until the $50,000-60,000 range and then a
decrease thereafter. The likelihood of OFAFH (FAH) consumption increases
(decreases) with higher income. This most likely reects increased consumption in
restaurants for those with high incomes, but again this merits further investigation.
Figure 5.
Predicted probabilities
of consumption across age
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 10 20 30 40 50 60 70 80 90 100
Age
FAH
FF
OFAFH
Determinants
of US fast food
consumption
15
Figure 6.
Predicted probabilities
of consumption across
household size
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1 3 5 7 9 11 13 15
Household size
FAH
FF
OFAFH
Figure 7.
Predicted probabilities
of consumption across
income
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 10,000 20,000 30,000 40,000 50,000 60,000 70,000 80,000 90,000 1,00,000
Income
FAH
FF
OFAFH
BFJ
112,1
16
Other variables. Pregnant females were less likely to consume FF and OFAFH.
Moreover, those individuals in rented households, living in a suburban area, or not
dieting were more likely to consume FF. Finally, temporal and seasonal variables were
also important. FF consumption was the highest during the year in the months of August
and December, during the week on Thursday and Friday, and during the day for lunch.
Discussion
It is interesting to compare the ndings from the binary and multinomial logit models
to previous research. Our results indicated that the West region has the lowest
likelihood of FF consumption, which is consistent with ndings reported in Jekanowski
et al. (2001). We found that the South and Midwest had the highest likelihood of FF
consumption. This result is consistent with Lin et al. (2001) nding reported that on
average the South (followed by the Midwest region) had the highest French fry
consumption on any given day. Similar to our ndings, Lin et al. (2001) also reported
that French fry consumption varies by age, region, urbanization, race, and ethnicity.
The reported elasticities for income, household size, and age in the Steward et al. (2005)
were consistent with the current study. Moreover, they also reported results
qualitatively similar to the current study for race and origin.
Both the present study and Steward et al. (2004) nd a statistically signicant link
between income and FF consumption/demand. McCracken and Brandt (1987) and Lin
et al. (2001) reported no signicant impact of income on FF expenditures. Jekanowski
et al. (2001) found income to be statistically signicant for only the 1982 equation
(insignicant in 1992). Byrne et al. (1996) also reported signicant income effects of
food away from home. Byrne et al. (1998) found a negative effect of income on
quick-serve facilities that diminished over time. As demonstrated in model results and
Figure 3, our evidence supports that income is a signicant factor in FF demand but
that the effects vary by income levels, gender, and region.
It is also interesting to compare the current ndings with regard to FF across the
binary and multinomial logit regression models. On average 26.9 percent of the
individuals consumed a FF item on a per day basis, while 4.4 percent of the individuals
consumed a FF item on a per eating occasion basis. In general, for both the binary and
multinomial logit regression model, the expected likelihood of FF consumption:
.
increased until around 20-30 years of age (contingent upon gender) and then
decreased;
.
increased as household income grew until about $50,000-70,000 and then
decreased; and
.
decreased as household size increased.
Further, males from the Midwest and South regions that live outside central cities in
MSA had the highest likelihood of consuming FF.
The multinomial regression model provided unique insight (not available from the
logit model) into important tradeoffs between FF, OFAFH, and FAHacross age, income,
and household size. The probability of FAH (OFAFH) consumption decreased
(increased) for the younger individual, but increased (decreased) for individuals past
middle aged. The likelihood of OFAFH consumption decreased until a household had
about eight members then started to increase, while the likelihood of FAHincreased and
then decreased after about eight members. The probability of OFAFH consumption
Determinants
of US fast food
consumption
17
increased and FAH consumption decreased with higher income. Further, because the
multinomial regression model was delineated by eating occasion, signicant differences
in FF consumption were found across eating occasions and seasons.
In all, the ndings of the current study are robust across model specications and
consistent with other research ndings. The predicted probability responses to
demographic and socioeconomic factors were predominately consistent across the two
empirical relationships, which reect robustness of model results. Bowman et al. (2004),
using the same data source as the current study, reported binary logit regression
results for ages 4-19 qualitatively consistent with our results. Income, gender, race,
urbanization, and region have signicant and qualitatively similar effects on FF
consumption for the younger and general population.
Concluding remarks
We specied a consumer choice model and analyzed the likelihood that an individual
consumed FF using the binary and multinomial logit regression. Data used in the
study were from the CSFII of 1994-1996 and the SCS of 1998.
Results of the empirical analysis indicate that 26.9 (4.4) percent of individuals
consumed FF on a per day (eating occasion) basis. Important regional and demographic
factors emerged from both sets of model results and predicted probability curves.
Consistent with previous research consumers in the South and Midwest were most likely
to consume FF. Individuals were more likely to consume FF until they reached
20-30 years of age at which point the likelihood that they consume FF decreases
throughout their life. In terms of gender, men were more likely to consume FF than were
women. Some minority respondents had a higher likely of consuming FFs than other
races. Larger households (especially those with more than four persons) were much less
likely to consume FF. Finally, income signicantly inuenced the likelihood of FF
consumption. Households with income levels under $50,000-60,000 are expected to
consume FF more often than those with higher incomes. The signicance of income as a
determinant of FF consumption during the 1990s is relevant given inconsistent ndings
from previous literature.
The results of this study also provided unique insight into important tradeoffs by
consumers between FF, OFAFH, and FAH. The probability of FAH (OFAFH)
consumption decreased (increased) for younger individuals, but increased (decreased)
for individuals past middle age. The likelihood of OFAFH consumption increased and
FAH consumption decreased with higher income. Further, we found differences in FF
consumption across eating occasions. Individuals have a higher likelihood of consuming
FF for lunch on Thursdays or Fridays in the months of August or December.
Over all, the above information should help researchers and policy makers concerned
with health and obesity to better understand the socioeconomic and demographic factors
that inuenced FF consumption and tradeoffs with OFAFH and FAH during the 1990s
(and relative to previous studies). Coupled with elasticity measures from the past and
current studies, the probability curves provide valuable insight into consumption of FF,
OFAFH, and FAHthat is useful in better understanding issues about structure of the food
industry and distribution sectors, nutritional intake, and demand for goods at different
marketing levels. This information is also relevant to retailers and wholesalers in the food
industry that respond directly to primary consumption. For example, the novelty of more
healthy menus at some FF facilities seemingly reects the response of the industry to
BFJ
112,1
18
address health and obesity concerns and target specic socio-demographic groups.
Finally, these results should be helpful to those structuring government programs that
directly or indirectly inuence consumer behavior or agricultural markets in general.
Notes
1. Smith (2004) also provides an interesting essay on dietary preferences, investigating the
consumers search problem subject to manipulation by food-producing rms. He
investigates evidence from medical and behavioral sciences to explain endogenous
determination of dietary preferences, including those for FF.
2. Bowman et al. (2004) also used day 1 observations because of the higher overall response rate.
3. Henceforth, we use FF to mean food items from FF establishments.
4. The well-known limitation of the multinomial logit model is the property called independence
fromirrelevant alternatives whereby the odds ratio between any two choices is independent of
the remaining choices. See Greene (2003) for further econometric details. Nevertheless, the
multinomial logit model can be viewed as a special case of a general model of utility
maximization. Hence, the results from the multinomial logit model estimated ahead can still
provide interesting and important insights about the choices between FF, FAH, and OFAFH.
5. Personal communication with Biin-Hwan Lin, USDA/ERS.
6. A reviewer pointed out alternatives specications (e.g. categorical variables) and limitations
(e.g. other potential interactions) to using a continuous variable for age when attempting to
quantify impacts on individual FF participation. Although interesting it is outside the scope
of this study and provides opportunity for future research. We direct the reader to Bowman
et al. (2004) for the effects of FF consumption and children.
7. Byrne et al. (1998) also specied linear and quadratic variables in the regression analysis.
8. As pointed out by a reviewer, larger families may have more school children and consume
more OFAFH in the form of school meals.
References
Bowman, S.A., Gortmaker, S.L., Ebbeling, C.B. and Pereira, M.A. (2004), Effects of fast-food
consumption on energy intake and diet quality among children in a national household
survey, Pediatrics, Vol. 113, pp. 112-18.
Byrne, P.J., Capps, O. Jr and Saha, A. (1996), Analysis of food-away-from-home expenditure
patterns for US households, 1982-1989, American Journal of Agricultural Economics,
Vol. 78 No. 3, pp. 614-27.
Byrne, P.J., Capps, O. Jr and Saha, A. (1998), Analysis of quick-serve, mid-scale, and up-scale
food away from home expenditures, The International Food and Agribusiness
Management Review, Vol. 1, pp. 51-72.
Deaton, A. and Muellbauer, J. (1980), Economics and Consumer Behavior, Cambridge University
Press, New York, NY.
Greene, W.H. (2003), Econometric Analysis, 5th ed., Prentice-Hall, Upper Saddle River, NJ.
Jekanowski, M.D. (1999), Causes and consequences of fast food sales growth, Food Review,
January/April, pp. 11-16.
Jekanowski, M.D., Binkley, J.K. and Eales, J. (2001), Convenience, accessibility, and the demand
for fast food, Journal of Agricultural and Resource Economics, July, pp. 58-74.
Judge, G.G., Hill, R.C., William, E.G., Lutkepohl, H. and Lee, T. (1988), Introduction to the Theory
and Practice of Econometrics, Wiley, New York, NY.
Determinants
of US fast food
consumption
19
Kuchler, F. and Lin, B.H. (2002), The inuence of individual choices and attitudes on adiposity,
International Journal of Obesity, Vol. 26 No. 7, pp. 1017-22.
Lin, B.H. and Franzao, E. (1999), Away-from-Home Foods Increasingly Important to Quality of
American Diet, available at: www.ers.usda.gov/Publications/aib749/
Lin, B.H., Lucier, G., Allhouse, J. and Kantor, L.S. (2001), Fast food growth boosts frozen potato
consumption, Food Review, Vol. 24, pp. 38-46.
McCracken, V.A. and Brandt, J.A. (1987), Household consumption of food-away-from-homejtotal
expenditure and by type of food facility, American Journal of Agricultural Economics,
Vol. 69 No. 2, pp. 274-84.
McFadden, D. (1981), Econometric models of probabilistic choice models, in Manski, C.F. and
McFadden, D. (Eds), Structural Analysis of Discrete Data with Applications, MIT Press,
Cambridge, MA.
McFadden, D. and Train, K. (2000), Mixed MNL models for discrete response, Journal Applied
Econometrics, Vol. 15, pp. 447-70.
Maddala, G.S. (1982), Limited Dependent and Qualitative Variables in Econometrics, Cambridge
University Press, New York, NY.
Marsh, T.L., Schroeder, T.C. and Mintert, J. (2004), Impacts of meat product recalls on consumer
demand in the USA, Applied Economics, Vol. 36, pp. 897-909.
Piggott, N.E. and Marsh, T.L. (2004), Does food safety information impact US meat demand?,
American Journal of Agricultural Economics, Vol. 86, pp. 154-74.
Schlosser, E. (2002), Fast Food Nation, Penguin Books, New York, NY.
Senauer, B. (1979), Effect of demographic shifts and changes in the income-distribution on
food-away-from-home expenditure, American Journal of Agricultural Economics, Vol. 61
No. 5, pp. 1046-57.
Smith, T.G. (2004), The McDonalds equilibrium: advertising, empty calories, and the endogenous
determination of dietary preferences, Social Choice and Welfare, Vol. 23, pp. 383-413.
Steward, H., Blisard, N., Bhuyan, S. and Nayga, R.M. Jr (2004), The demand for food away from
home: full-service or fast food, Agricultural Economic Report Number 829, ERS/USDA,
Washington, DC.
Stewart, H., Blisard, N., Jolliffe, D. and Bhuyan, S. (2005), The demand for
food-away-from-home: do other preferences compete with our desire to eat healthfully?,
Journal of Agricultural and Resource Economics, Vol. 30, pp. 520-36.
Yen, S. (1993), Working wives and food away form home: the Box-Cox double hurdle model,
American Journal of Agricultural Economics, Vol. 75, pp. 884-95.
Yen, S.T. and Jones, A.M. (1997), Household consumption of cheese: an inverse hyperbolic sine
double-hurdle model with dependent errors, American Journal of Agricultural Economics,
Vol. 79, pp. 246-51.
Further reading
USDA/ERS (2007), Food Expenditure Brieng Room: Food Expenditure Tables, US Department
of Agriculture, Economic Research Service, Washington, DC, available at: www.ers.usda.
gov/Brieng/CPIFoodAndExpenditures (accessed July 2007).
Corresponding author
Thomas Marsh can be contacted at: tl_marsh@wsu.edu
BFJ
112,1
20
To purchase reprints of this article please e-mail: reprints@emeraldinsight.com
Or visit our web site for further details: www.emeraldinsight.com/reprints

Das könnte Ihnen auch gefallen