372 THE JOURNAL OF CONSUMER AFFAIRS

JAMES K. BINKLEY
The Effect of Demographic, Economic, and Nutrition
Factors on the Frequency of Food Away from Home
Food away from home, especially fast food, is often cited as contributing
to obesity and other nutritional problems. This negative publicity can
affect demand. Models explaining visits to table service and fast food
restaurants are estimated, with nutrition variables added to standard
demographic measures. Demographic effects are similar to those in past
studies. Nutrition factors have little impact on table service, but nutrition-
orientated consumers tend to have lower fast food consumption.

One of the largest changes in American eating habits in recent decades

has been the increasing reliance on food eaten away from home (FAFH).
FAFH has increased from 33% of total food expenditures in 1970 to 47%
by 2003.1 Most of this is at table service and fast food restaurants. Much of
the growth is attributed to the rising value of household time, especially as
induced by more female labor force participation and rising household
incomes. The importance of these factors has been shown in numerous
studies (Byrne, Capps, and Saha 1998; Kinsey 1983; McCracken and
Brandt 1987; Prochaska and Schrimper 1973; Redman 1980; Yen
1993). In addition, studies have consistently found that FAFH declines with
household size, reflecting the scale economies associated with household
meal preparation, and that women and older individuals of either sex are
less likely to dine out. Separate analysis by type of facility has found dif-
ferent effects for some factors. For example, income is generally more
important for table service, while convenience and accessibility have rel-
atively greater influence for fast food (Jekanowski, Binkley, and Eales
2001; McCracken and Brandt 1987).
Recently, the growth in FAFH has generated concern about its possible
effect on dietary quality. Analysis of food consumption surveys has indi-
cated that meals eaten in restaurants are generally of lower nutritional qual-
ity than meals eaten at home, mainly due to higher fat and calorie content
(Lin and Frazao 1997). Furthermore, obesity is now one of the nations
leading health problems, and because its growth has paralleled the trend

James K. Binkley is a professor at the Department of Agricultural Economics, Purdue University,

West Lafayette, IN ([email protected]).

The Journal of Consumer Affairs, Vol. 40, No. 2, 2006

ISSN 0022-0078
Copyright 2006 by The American Council on Consumer Interests
WINTER 2006 VOLUME 40, NUMBER 2 373

in dining out, FAFH is often suggested as contributing to the energy imbal-

ance that brings it about. Although this has not been scientifically estab-
lished, the nature of restaurant food has become a policy issue. For
example, proposals mandating that chain restaurants provide nutritional
information on their menus have been introduced in both houses of Con-
gress (Burton and Creyer 2004). On another front, lawsuits have been filed
by diners alleging that their obesity resulted from restaurant meals.
This public scrutiny has caused some restaurant chains to adopt proac-
tive measures. New products geared to the nutrition-orientated consumer
have been introduced, particularly by fast food chains. The success of these
initiatives ultimately depends on acceptability by consumers. Although ini-
tial sales appear promising (Warner 2005), previous introductions of
healthy menu options have not been highly successful (Consumer Reports
1996, 2004). One possible reason for this is that individuals concerned with
nutrition are less inclined to dine out, perhaps due to the bad publicityef-
fectively negative advertisingdirected at restaurant food.
Because of these considerations, a potentially important question is the
extent to which nutrition concerns and attitudes affect the decision to have
an FAFH meal. Although there have been several studies of the impact of
nutrition factors on the demand for particular foods or nutrients (Brown and
Schrader 1990; Chern, Loehman, and Yen 1995; Ippolito and Mathios
1990), most have focused on nutrition information, and none has consid-
ered restaurant dining. In this paper, we do so using an econometric model
containing not only demographic measures and measures of convenience
but also measures of nutrition attitudes, concerns, and knowledge. The
issue addressed is the decision to dine out: we do not consider the subse-
quent issue of what is eaten when dining out occurs.
Separate equations are estimated for fast food and table service restau-
rants. This is not only because of the differing effects of economic and
demographic factors identified in previous work but also to permit differ-
ences for nutrition variables. Although FAFH is one of the most frequently
cited factors behind the obesity epidemic, it is fast food that receives most
of the criticism, especially in the popular media. Recent examples of this are
Eric Schlossers Fast Food Nation (2001) and in the well-received docu-
mentary Supersize Me. In a 2004 ABC NewsTime Magazine poll, 43% of
respondents thought that fast food bears a great deal of responsibility for
the obesity crisis. The attitude that fast food is particularly bad is also pres-
ent in the academic literature. Many studies of the dietary impacts of FAFH
are confined to fast food (e.g., Bowman et al. 2004; Paeratakul et al. 2003).
A likely reason for the different treatment is that most popular fast food
items tend to be relatively high in fat and calories. Although the same items
374 THE JOURNAL OF CONSUMER AFFAIRS

are available from table service restaurants, better-nutrition alternatives are

also likely to be on the menu. In any case, this emphasis on fast food in the
obesity and nutrition debate may have caused some consumers to avoid fast
food when making dining choices.

MODEL AND DATA

Most studies involving FAFH are grounded on household production

theory, and this study is no exception. As developed by Becker (1965),
household production theory views the household as both a consumer
and producer of final goods, so that both household time and market-
produced goods enter the utility maximization process. This view is clearly
relevant for food consumption since meals can either be produced in the
household using purchased inputs and household time or be purchased
ready-made at a restaurant. In addition, we recognize that health is an
important final good to most consumers and that they thus view food
and good nutrition as inputs into health. This interdependency between
food and health makes food choice dependent not only on prices, income,
and household time, but potentially also on measures of nutrition knowl-
edge and concern.
These considerations lead to a model of the following general form:

Yi f P, I, T, H, D 1

in which Yi is a measure of household or individual food choice, P is a set of

relevant prices, I represents household income, T involves measures of time
cost, H is measures of nutrition concerns and knowledge, and D represents
demographic and other factors. The latter can be viewed as proxies for taste
and perhaps factors not captured by the variables in H.
This study uses the individual consumer as the observational unit.
The data are from the 19941996 Continuing Survey of Food Intakes
by Individuals (CSFII) and the accompanying Diet and Health Knowledge
Survey (DHKS) (United States Department of Agriculture 2000). This is
a nationally representative sample of noninstutionalized persons living in
the United States. The CSFII involves 16,103 individuals, most of whom
supplied two nonconsecutive days of detailed dietary intake collected by
trained in-person interviewers using 24-hour recalls. The nutrient intake
lists the name of each food eaten, a detailed breakdown of its nutritional
content, and where it was obtained. The data also includes demographic
measures for the individuals surveyed and for their households. The DHKS
WINTER 2006 VOLUME 40, NUMBER 2 375

is a follow-up telephone survey of 5,765 individuals at least 20 years old

who participated in the CSFII. Its purpose is to assess their knowledge of
and attitudes toward nutrition and health.
The sample was limited to those participating in the DHKS. Beginning
with the 5,765 participants, we eliminated those who did not provide intake
data for both days. Additional observations were lost because some
respondents failed to provide values for all variables. This left a sample
of 4,361 individuals, 2,147 of whom had at least one FAFH visit during
the sample period.2
Given the interest in the decision to dine out, Yi in equation (1) is the
number of FAFH visits during the two days. This is a short period, and
a question might arise regarding how well it captures typical behavior.
For example, someone temporarily absent from home is more likely to dine
out, irrespective of demographics or attitudes concerning nutrition; con-
versely, during an illness dining out is less likely. This problem applies
in some degree to any analysis using the CSFII data. The survey design
attempts to lessen its impact by requiring that the survey days be noncon-
secutive, with at least three days between them. The possibility of being
away on both survey days is further reduced because interviews were con-
ducted in the respondents home on the day after the day of intake. Addi-
tionally, the large number of observations reduces the impact of any one of
them and generally helps to overcome the high degree of error associated
with individual behavior in a two-day period.
To reduce this error further, we limited the definition of a FAFH visit.
The CSFII data is based on occasions, defined as breakfast, brunch,
lunch, dinner, supper, or snack. We wished to confine attention to complete
meals. Therefore, snacks were excluded. A further requirement was that the
restaurant meal have at least two items, with no items obtained elsewhere.
The dependent variable in each equation is the total number of visits to the
restaurant type in question by each individual. The sample averages are
.371 for table service and .326 for fast food.
Because of the special nature of the dependent variable, ordinary least
squares is not an appropriate estimation procedure. A common way to
address this problem is to view the process as Poisson,

e2ki ki Yi
f Yi for Yi 1, 2, .
Yi !

The parameter ki is usually modeled as ln(ki) Xib, where X is a set of

explanatory variables affecting the probabilities for Y (Greene 2003, 880).
376 THE JOURNAL OF CONSUMER AFFAIRS

These models are easily estimated by maximum likelihood, based on the

log-likelihood function

X
n
ln L 2ki 1 Yi b0 Xi 2ln Yi !

i1

However, a disadvantage of the Poisson is that the variance and the mean
are equal. This restriction is likely to be unrealistic for many economic pro-
cesses for the variance often exceeds the mean, the overdispersion prob-
lem. Empirically, it generally appears in the form of more zeros and more
large values of Y than would be predicted by the Poisson process.
A popular alternative to the Poisson, which is not subject to overdisper-
sion, is the negative binomial model. The negative binomial can be viewed
as a Poisson model with specification error, i.e.

ki expXi b 1 gi

The error accounts for individual variation, and high individual variation
is the main consequence of having only two days of observation on the
survey respondents. Greene (2003, 886) illustrates that the distribution
of Y conditional on gi is again Poisson, and it can be straightforwardly
estimated by maximum likelihood. It is the method employed in this study.3
In Table 1 appear the sample frequencies of table service and fast food
visits, along with the expected Poisson frequencies with means equal to the
respective sample means. The expected and sample proportions are quite
similar, but the overdispersion in the sample is quite evident: there are too
many zeros and larger values.4

Independent Variables

In Table 2 are presented the independent variables used in the models,

along with summary statistics. The independent variables can be broadly
classed into two groups. One contains economic and demographic variables
similar to those used in prior studies. The second involves measures related
to nutrition and health. As can be seen, there is a large number of variables,
39 in total. While some might view this as excessive, we chose to err on the
side of overparameterization. The analysis is somewhat exploratory since
there is no exact theoretically correct model, at least with respect to the role
of nutrition. Furthermore, there is little multicollinearity in this data: the
largest variance inflation factor was 2.1. A larger number of included var-
iables also has the advantage of reducing the possibility of endogeneity
WINTER 2006 VOLUME 40, NUMBER 2 377

TABLE 1
Poisson-Predicted Frequencies and Sample Frequencies of Restaurant Visits
Table Service Fast Food
Visits Poisson Actual Poisson Actual
0 .690 .732 .722 .755
1 .256 .192 .235 .181
2 .047 .055 .038 .049
3 .006 .014 .004 .011
4 .000 .005 .000 .003
5 .000 .001 .000 .000
6 .000 .000 .000 .000

induced by neglected heterogeneity, i.e., omitted variable bias. This is elab-

orated below.
Among the group of economic variables are several related to money and
time costs. Income is measured as per-person household income, expected to
positively affect the number of FAFH occasions. Because table service pro-
vides more amenities and greater variety, it should be more responsive to
income, as found in previous work (e.g. Byrne, Capps and Saha 1998; Nayga
and Capps 1994). The CSFII database has no direct measures of prices. We
follow the usual practice of assuming that all respondents faced the same rel-
ative prices and to include regional variables to capture any remaining cross-
section differences. A dummy variable measuring whether the household is
receiving food stamps is included. Although food stamps could conceivably
increase FAFH due to an income effect, they effectively lower the food at
home price. The resulting substitution effect is likely to be more important,
reducing demand for FAFH. Higher time cost is always found to increase the
demand for FAFH; our measure of this is the number of hours per week usu-
ally worked by the respondent. An additional aspect of time cost is the avail-
ability and closeness of restaurants (Jekanowski, Binkley, and Eales 2001).
We include urban-suburban-rural indicators to capture this. Those living in
urban and suburban areas have lower accessibility costs than do rural resi-
dents, so FAFH usage should be higher. A variable also related to time cost
is the size of household. Because of scale economies in household food prep-
aration, the time cost per person by the meal preparer falls as household size
increases, reducing FAFH usage. This has also been found in most studies.
The DHKS provided two price and cost measures. PRICEIMP is a binary
variable measuring whether the respondent considers price very impor-
tant when buying food. Consumers particularly concerned with price,
the money cost, would be expected to make less use of the FAFH meal
option since it is usually more expensive than dining at home. The opposite
applies to CONVENIENCE, an indicator measuring the importance of
378 THE JOURNAL OF CONSUMER AFFAIRS

TABLE 2
Variables Used in the Analysis and Summary Statistics
Standard
Variable Description Mean Deviation Minimum Maximum
OCC2 Table service visits .371 .718 0 6
OCC3 Fast food visits .326 .651 0 4
INCOME Per capita household income (1000 dollars) 16.86 13.85 0 100
HOURS Usual hours worked/week 24.89 22.45 0 91
HHSIZE Household size 2.59 1.40 1 16
FOODSTAMPS 1 food stamp participant 0.07 .25 0 1
PRICEIMP 1 price very important when food shopping .45 .50 0 1
CONVENIENCE 1 convenience very important when .38 .49 0 1
food shopping
AGE Age in years 49.76 16.79 20 90
GENDER 1 female .51 .50 0 1
EDUCATION Years of education 12.96 2.86 0 17
AFRICAMER 1 African American .12 .32 0 1
HISPANIC 1 Hispanic .07 .25 0 1
TV Hours per day 2.61 2.10 0 19.5
FRIDAY Number of Fridays in the two interview days .28 .46 0 2
SATURDAY Number of Saturdays in the two interview days .20 .41 0 2
SUNDAY Number of Sundays in the two interview days .34 .48 0 2
RURAL 1 lives in rural areaa .26 .44 0 1
URBAN 1 lives in urban areaa .30 .46 0 1
EAST 1 East region .19 .39 0 1
SOUTH 1 South region .35 .48 0 1
MIDWEST 1 Midwest region .26 .44 0 1
BMI Body mass index 26.36 4.96 15.19 45.91
DIET 1 on any kind of diet .19 .39 0 1
VEGETARIAN 1 vegetarian .03 .17 0 1
PRODUCE Number of 23 types of fruits and vegetables 12.50 4.58 0 23
eaten in last year
DISFAT Larger values imply avoids discretionary fat 2.19 .71 1 4
SUBFAT Larger values imply substitutes low fat for 2.61 .74 1 4
regular foods
TASTE 1 taste is very important when .84 .36 0 1
food shopping
NUTRITION 1 nutrition is very important when food .65 .48 0 1
shopping
HLTHYWT 1 very important to maintain a healthy weight .75 .44 0 1
LOWFAT 1 very important for diet to be low in fat .60 .49 0 1
FRTVEG 1 very important to eat lots of fruit and .69 .46 0 1
vegetables
NUTSCORE Number correct of 14 nutrition questions 8.85 2.33 0 14
LABEL 1 frequently uses nutrition labels .32 .47 0 1
USEWELL 1 knows how to use labels for nutritious diet .23 .42 0 1
SENSE 1 healthy diet just requires knowing what is .40 .49 0 1
good and bad
NOCHANGE 1 my diet is healthy and requires no changes .18 .38 0 1
a
Suburban is reference.

preparation time. Individuals who regard ease of preparation as very impor-

tant have high time costs and thus are expected to dine out more frequently.
The model includes several demographic measures that can be broadly
classified as related to taste and preferences. These involve race and
WINTER 2006 VOLUME 40, NUMBER 2 379

ethnicity (African American and Hispanic), years of education, gender, and

age. We regard the effect of ethnicity somewhat as an empirical question,
although the balance of the evidence is that minorities make less use of
FAFH. It is reasonable that more highly educated individuals make greater
use of table service restaurants, due to food variety, but perhaps not fast
food. It has been found that FAFH declines after middle age (Dong et al.
2000; Nayga and Capps 1994), so a negative sign is expected for age. Re-
garding gender, a survey by the National Restaurant Association found that
men consume an average of 4.6 commercially prepared meals per week,
while for women the figure is 3.8 (Ebbin 2000).5
As a measure of lifestyle, TV, the hours per day spent watching TV, is
included in the model. People who watch a large amount of television are
likely to spend a lot of time at home, reducing demand for FAFH. So a neg-
ative sign is expected. It is reasonable to expect that people are more likely
to dine out on weekends than during the week (Dong et al. 2000). We al-
low for this with three additional dummies: FRIDAY, SATURDAY, and
SUNDAY. This allows separate effects for each of these, with a common
effect for the remaining four days.
A final variable included in this group is the body mass index (BMI) of
the respondent. If the main reason people become overweight is that they
obtain above normal enjoyment from eating, particularly energy dense
foods, we might expect overweight people to have a greater demand for
FAFH. Then being overweight can be regarded as a cause for dining
out. On the other hand, many people believe that one reason for the obesity
epidemic is the nations increased reliance on FAFH. To the extent this
view is valid, BMI is partly an endogenous variable. In spite of this,
BMI is included in the model, mainly because we found that results for
other variables were insensitive to whether it was present.6
Most of the nutrition variables are binary in nature. Variable choice was
based on factors deemed likely to affect the dining out decision and meas-
ures potentially related to policy issues. In many cases, the direction of
effect is not necessarily obvious a priori. Some ambiguity is inevitable,
given the exploratory nature of the study and the fact that these variables
are survey responses, not observational measures. Nevertheless, under
a maintained hypothesis that, relative to food at home, restaurant meals
are high in fat and calories, it is reasonable to expect that variables asso-
ciated with good dietary practice and a higher regard for nutrition and
health would have negative signs in our models.
That said, however, an additional complexity is the possibility of endo-
geneity due to joint relations between the error, the decision to dine out,
and nutrition measures. This has been a concern in studies specifically
380 THE JOURNAL OF CONSUMER AFFAIRS

examining the role of nutrition knowledge on diet. Presumably, the problem

is that consumers with high nutrition concernsan omitted variablehave
better diets and greater levels of knowledge, leading to an overestimate of
the latters effect. However, the model herein allows for this by including
measures of nutrition concern, reducing the likelihood of endogeneity prob-
lems from that source. Of course it could arise for other, less apparent
reasons (a possibility with any study), with indeterminate consequences.
They are likely to be small relative to the high levels of statistical noise
inevitably present in nonexperimental data on individuals. Park and Davis
(2001) have an illuminating discussion of the difficulty of dealing with
endogeneity in household survey data, where instrumental variables are
particularly difficult to obtain.
Of the nutrition variables, six describe current dietary behavior. DIET
indicates whether the respondent is on any kind of diet. Because dining
away from home reduces the individuals control of food ingredients, a neg-
ative sign is expected. The same is true of VEGETARIAN, since meatless
FAFH options, while certainly available, are usually somewhat limited,
especially for fast food. PRODUCE, taken from the CSFII, is a measure
of fruit and vegetable consumption. Respondents were asked 23 questions
of the form In the last twelve months, did you eat ___?, where the blank
contained a fruit or vegetable.7 PRODUCE is the number of affirmatives. It
is a measure of taste for fruits and vegetables, variety, and preference for
nutritious foods, so a negative sign is expected.
DISFAT and SUBFAT indicate whether respondents avoid adding dis-
cretionary fat and whether they substitute low-fat versions of foods, respec-
tively. Each of these is the average of questions coded from 1 to 4, with larger
values indicating an inclination to avoid fat and substitute low-fat items.8
Low-fat substitution possibilities in restaurants may be limited, and the dis-
cretionary fat may be added before the food is served. Thus, individuals who
normally restrict discretionary fat and/or make low-fat food substitutions
may be less inclined to dine out. Hence, negative signs are expected.
The model includes five variables measuring whether certain aspects of
diet/health are very important. These are NUTRITION (nutrition when
food shopping), HLTHYWT (maintaining a healthy weight), LOWFAT
(maintaining a diet low in fat), FRTVEG (getting adequate amounts of fruit
and vegetables), and TASTE (taste when food shopping). The first four are
positively related to nutrition and thus are expected to have negative signs;
TASTE is expected to be positive.
The remaining variables are measures associated with nutrition knowl-
edge. LABEL measures whether the respondent currently uses food labels
with frequency. If this is found to be related to FAFH demand, it can help
WINTER 2006 VOLUME 40, NUMBER 2 381

assess the potential value of menu labels. Perhaps the most reasonable
expectation is a negative effect. Presumably, label users desire a healthy
diet, and with the general belief that FAFH is substandard in nutrition, they
might avoid restaurants. The same applies to a related variable, USEWELL,
which indicates the respondent has high confidence in her/his ability to use
labels to choose a healthy diet.
NUTSCORE is the number of correct answers to 14 specific questions
about nutritional characteristics of foods (a typical question: Which has
more saturated fat: butter or margarine?). A consumer with high nutrition
knowledge (and who desires a nutritious diet) may avoid FAFH because it
tends to be less healthy, thus generating a negative effect. But such an indi-
vidual may have greater ability to navigate the menu and avoid nutrition
pitfalls, reducing concern that FAFH will lower diet quality. Then, dining
out may be more likely.
SENSE is an indicator variable equaling 1 when the individual strongly
agrees with the statement Choosing a healthy diet is just a matter of know-
ing what is good and what is bad. Although this is essentially a truism, we
interpret it as expressing the sentiment that what is needed to have a healthy
diet are common sense rules, such as avoid fat and eat lots of fruit and
vegetables, not detailed nutrition knowledge, such as that measured by
NUTSCORE.9 Given the negative publicity about the nutrition of FAFH,
one might expect a negative effect. On the other hand, such a viewpoint
may simply be a rationalization for not making the effort to obtain specific
information.
A similar variable is NOCHANGE, which has value 1 if the individual
strongly believes that their current diet is healthy and requires no change.
We interpret this as indicating the respondent believes they make no serious
nutritional missteps, not that their current diet has achieved perfection. If
this self-assessment is accurate, then an expectation of reduced likelihood
of dining out is reasonable. However, Variyam, Shim, and Blaylock (2001)
found that people believing their diet needed no improvement were often
mistaken.

RESULTS

The results for the two estimated models appear in Table 3. The effect of
the six variables associated with the money or time cost of FAFH relative to
home mealsINCOME, HHSIZE, HOURS, FOODSTAMPS, PRICE-
IMP, and CONVENIENCEdiffers in the two cases. All except HHSIZE
are significant at 10% or better in the table service model, with expected
signs; estimates for fast food are similar in direction, but INCOME and
382 THE JOURNAL OF CONSUMER AFFAIRS

TABLE 3
Poisson Regression Results
Table Service Fast Food

Estimate Chi-Square Estimate Chi-Square

INTERCEPT 22.264 35.22*** 0.455 1.49
INCOME 0.292 33.93*** 0.054 1.25
HOURS 0.004 5.44** 0.005 10.03***
HHSIZE 20.041 2.45 20.024 0.98
FOODSTAMPS 20.379 3.77* 20.131 0.82
PRICEIMP 20.152 5.58** 20.139 4.54**
CONVENIENCE 0.119 3.76* 0.134 4.60**
AGE 20.001 0.08 20.022 88.62***
GENDER 20.192 9.57*** 20.225 12.19***
EDUCATION 0.011 0.69 20.008 0.32
AFRIAMER 20.339 7.92*** 0.140 1.96
HISPANIC 20.068 0.26 0.169 2.06
TV 20.102 32.31*** 20.057 11.48***
FRIDAY 0.209 11.10*** 0.083 1.61
SATURDAY 20.009 0.02 0.091 1.64
SUNDAY 0.030 0.21 20.036 0.29
RURAL 0.015 0.04 20.142 3.35*
URBAN 0.102 2.09 20.037 0.26
EAST 20.035 0.15 20.101 0.95
SOUTH 20.092 1.27 0.238 7.73***
MIDWEST 20.050 0.34 0.078 0.70
BMI 0.019 9.47*** 0.015 6.16**
DIET 20.040 0.26 20.081 0.83
VEGETARIAN 0.034 0.04 20.549 5.26**
PRODUCE 0.015 4.35** 20.025 12.64***
DISFAT 20.041 0.83 20.149 9.93***
SUBFAT 0.003 0.00 0.048 0.86
TASTE 0.095 1.41 0.031 0.14
NUTRITION 20.034 0.26 20.034 0.25
HLTHYWT 0.055 0.53 0.064 0.70
LOWFAT 20.013 0.03 20.007 0.01
FRTVEG 20.034 0.23 20.132 3.43*
NUTSCORE 0.027 3.78* 20.005 0.13
LABELUSE 0.067 0.87 20.014 0.04
USEWELL 0.016 0.05 20.162 4.17**
SENSE 20.108 2.98* 20.101 2.49
NOCHANGE 20.088 1.04 20.192 3.92**
R2 .076 .094
Note: R2 is calculated as the squared correlation between the actual and predicted value of the dependent
variable.
*Significant at 10% level; **significant at 5% level; ***significant at 1% level.

FOODSTAMPS are insignificant, while HOURS and CONVENIENCE

are somewhat more important. Generally, financial cost seems more impor-
tant for table service, while time cost is the driver of fast food. This is not
surprising.
WINTER 2006 VOLUME 40, NUMBER 2 383

Gender has a similar negative and highly significant effect for both res-
taurant types, indicating that women dine out less than men. Age also has
a negative effect for both, but it is not significant for table service but very
highly significant for fast food. Years of education affects neither. The only
strong racial effect in these models is that blacks are estimated to be less
likely to dine at table service facilities. Our view that the TV variable is
a measure of a stay-at-home lifestyle is borne out by the results: the coef-
ficient is negative and highly significant in both models.
Like earlier studies, we find a weekend effect but of a very limited type.
According to the results, a table service meal is much more likely to occur
on a Friday than on any other day. Fast food usage displays no such effect,
evidently more or less evenly distributed through the week. We find that
rural consumers are significantly less likely to use fast food than are those in
suburban areas (the omitted class), while there is a vague indication that
urban consumers make greater use of table service (the probability value
is .15). These likely reflect the importance of facility availability. The only
significant regional effect is the clear preference for fast food in the South.
In each case, the coefficient on BMI is positive and significant, which is
evidence that overweight individuals are more likely to dine out than are
others. Based on coefficient size and level of significance, they are estimated
to be more likely to choose table service. Assuming that one reason people
become overweight is an above average liking for eating, this could reflect
the greater variety and perhaps palatability found at table service restaurants
relative to fast food. It could also be that, fast food supersizing notwithstand-
ing, buffet style and frequent all-you-can-eat offers at table service outlets
provide a better value for anyone interested in eating large meals. To the
extent there is reverse causality, the result also suggests that dining in table
service restaurants is a greater source of excess weight. In any case, it is
evident that fast food is not the only factor in the FAFHobesity question.

Nutrition Variables

We now consider the nutrition-related variables, beginning with those

describing the current diet of the respondent. DIET, measuring whether
the respondent reports being on any kind of diet, is not significant in either
equation. Although this is somewhat surprising, Nayga and Capps (1994)
obtained the same result in their study. However, VEGETARIAN, which is
certainly a type of diet, is significantly negative in the fast food model.
Most fast food menus are built around a small number of meat-based items,
making them of limited interest to vegetarians. Table service, which often
384 THE JOURNAL OF CONSUMER AFFAIRS

provides vegetarian entrees, is not affected. For fast food, PRODUCE, the
measure of fruit and vegetable consumption, also has a negative effect, yet
more significant. But the coefficient for table service is significantly pos-
itive. A possible reason for this somewhat unexpected sign is that PRO-
DUCE is measuring not only a desire for fruits and vegetables but
a preference for variety as well, which we would expect to be positively
associated with dining in table service restaurants.
DISFAT and SUBFAT measure the degree to which the respondent
makes an effort to avoid fat, by either not adding discretionary fat in
the first place and/or by substituting low-fat foods for standard counter-
parts. Neither is significant for table service. For fast food, SUBFAT is
not significant, while DISFAT has a very highly significant negative effect,
indicating that those who avoid discretionary fat tend to also avoid fast
food. This difference for these apparently related variables suggests they
are distinguishing between two consumer types. One is those who avoid
discretionary fat not only for nutrition considerations but also because they
simply do not care for foods higher in fat. Since this seems to characterize
fast food, they would dine out less. The second is the group who enjoys
high-fat foodsand thus FAFHbut also worries about nutrition and so is
willing to make substitutions when the sacrifice is not too onerous.
Results for the group of variables measuring the importance respondents
claimed for aspects of nutrition were generally disappointing. NUTRITION
and TASTE indicate whether the characteristics in question are deemed
very important. We expected consumers ranking food taste as very
important to be more inclined to dine out, with the opposite effect for nutri-
tion. However, these expectations failed to materialize, for both variables
are estimated to have no effect whatsoever in either equation. Similarly,
neither LOWFAT nor HLTHYWT, measuring the importance of a low-
fat diet and maintaining a healthy weight, is significant in either model.
Of course, actual behavior does not always accord with what consumers
say they regard as important. Note from Table 2 that 75% of the sample
strongly agreed with the statement that healthy weight is very important.
Nevertheless, according to the Centers for Disease Control, in 19992000,
65% of adults older than 20 were overweight (Hedley et al. 2004). Further-
more, some people may become more aware of the importance of a healthy
weight when they themselves become overweight.
FRTVEG, measuring the importance of fruits and vegetables, is the only
member of this group whose effect is as expected. It is negative in both equa-
tions, and modestly significant for fast food. Given similar results for PRO-
DUCE and VEGETARIAN, we conclude that the unavailability of fruit and
vegetable items restricts the customer base of fast food outlets. This agrees
WINTER 2006 VOLUME 40, NUMBER 2 385

with other evidence. A recent survey found that more than twice as many fast
food than table service customers said they would eat out more often if more
fruits and vegetables were offered (Odesser-Torpe 2005).
Of the information variables, NUTSCORE is the most direct measure,
being the score on a nutrition test. With an insignificant effect for fast food
and a modestly significant positive effect for table service, we conclude that
more knowledgeable consumers are likely to choose table service over fast
food. Accepting that higher knowledge implies a desire for a healthy diet,
this makes sense. Someone alert to the nutritional properties of various
foods will be able to find reasonably healthy items among the variety of
table service dishes available, certainly more so than with the limited selec-
tions at a fast food outlet.
LABELUSE is insignificant in both models, meaning that food label
users are neither more nor less likely to dine out than nonusers. However,
those who believe they know how to use labels to choose healthy foods,
measured by USEWELL, are significantly less likely to eat fast food. This
is additional evidence of a negative correlation between nutrition concern
and fast food use.
We regard SENSE as indicating that nutrition choices are based on broad
rules of thumb rather than punctilious attention to details. In view of the
negative publicity regarding FAFH nutrition, one such rule is likely to
be other things the same, avoid discretionary dining out. The results
somewhat support this: the coefficient on SENSE is negative for both
FAFH types and significant at p .10 for table service, slightly less so
for fast food. Much the same argument applies to NOCHANGE, the
indicator that the respondent believes his/her current diet is healthy, which
also has a negative effect in both cases, but only significantly so for fast food.
Although our interpretation of these variables can be questioned, less dining
out would probably be viewed as evidence of good nutrition behavior.

Practical Effects

The coefficients in Table 3 are not direct effects because they refer to the
nonlinear equation for the expected number of FAFH visits in equation (1).
The ith marginal effect is bieXb, which depends on the values of all the

However, rather than
variables. A typical point of evaluation is the point X.
marginal effects, it is more interesting to consider selected discrete changes
in each of the k variables, using the difference formula

Di eXc b 2eXb b , i 1 to k
386 THE JOURNAL OF CONSUMER AFFAIRS

For continuous variables, Xb is the vector of sample means (including

sample means of binary variables) and Xc is Xb with the ith position
increased by 1 standard deviation of xi. That is, we predicted the effect
of a 1 standard deviation increase in xi on predicted visits taken at the mean
of other variables. For binary variables, Xb was again the vector of means
except that the ith position was replaced with a zero. For Xc, the same vector
was used, except the ith position was 1. This is the predicted difference in
visits when the characteristic is present versus when it is not, again taken at
the means. Since these effects refer to a period of two days, they were mul-
tiplied by 15, making them monthly differences.
These appear in Table 4. From this we see, for example, that being on
food stamps is associated with two fewer FAFH visits per month, 1.5 table
service and .5 fast food, while price-conscious food shoppers make at least
one less. Increasing age by 1 standard deviation from the mean (i.e., from
about 50 to over 66 years [Table 2]) has a similar effect, mostly due to less
fast food. Given our somewhat stringent definition of a restaurant meal, the
predicted malefemale difference of 2.2 per month compares reasonably
well with the .8 weekly difference found by the National Restaurant Asso-
ciation Survey noted previously.

Fast Food versus Table Service

A purpose of the study was to test whether nutrition factors have a greater
impact for fast food than for table service. Based on the number of signif-
icant coefficients, they do. On balance, the pattern suggests that consumers
with better dietary practices are less likely to dine at fast food outlets. There
is little evidence of any similar effect for table service demand.
To address this more formally, the variables were classified into two sets:
14 nutrition and diet variables, and the remaining 22 variables. ONDIET
and VEGETARIAN were included in the latter group, the first because it is
often not a choice variable and the second because of its low prevalence in
the sample. We then conducted a likelihood ratio test of the significance of
each group in each equation. These appear in Table 5, which shows that the
likelihood ratio statistics for the other group do not greatly differ for the
two restaurant types, with fast food slightly larger. Both are highly signif-
icant. In either case, it is considerably larger than the statistic for the nutri-
tion group, indicating that economic and demographic factors are more
important than nutrition measures in explaining differences in FAFH
demand. However, it is evident that the difference in the fast food equation
is much smaller, suggesting a larger role for nutrition factors in the fast food
WINTER 2006 VOLUME 40, NUMBER 2 387

TABLE 4
Effect of a 1-Standard Deviation Change in Indicated Variable on Monthly FAFH Visits
Table Service Fast Food
INCOME 1.40 0.20
HOURS 0.42 0.51
HHSIZE 20.26 20.14
FOODSTAMPS 21.53 20.50
PRICEIMP 20.72 20.55
CONVENIENCE 0.57 0.55
AGE 20.05 21.25
FEMALE 21.11 21.13
EDUCATION 0.15 20.09
AFRICAMER 21.41 0.59
HISPANIC 20.31 0.73
TV 20.91 20.45
FRIDAY 1.03 0.34
SATURDAY 20.04 0.38
SUNDAY 0.14 20.14
RURAL 0.07 20.55
URBAN 0.49 20.15
EAST 20.17 20.39
SOUTH 20.43 0.99
MIDWEST 20.24 0.32
BMI 0.48 0.32
DIET 20.19 20.32
VEGETARIAN 0.16 21.72
PRODUCE 0.34 20.44
DISFAT 20.14 20.41
SUBFAT 0.01 0.15
TASTE 0.44 0.12
NUTRITION 20.16 20.14
HLTHYWT 0.26 0.25
LOWFAT 20.06 20.03
FRTVEG 20.16 20.54
NUTSCORE 0.31 20.05
LABELUSE 0.32 20.06
USEWELL 0.08 20.62
SENSE 20.51 20.40
NOCHANGE 20.41 20.73
Note: Bold indicates coefficient significant at 10% level or better.

decision. For table service, the nutrition variables as a group are not sig-
nificant at even .10; for fast food, they are highly significant.
To illustrate the potential consequences of this difference, we used the
estimated equation to compare the predicted mean outcomes for two hypo-
thetical consumers, one with high and one with low nutrition concerns/
interest/behavior, where these are defined by values of the 14 nutrition-
related variables. In the case of the high consumer, all binaries are
388 THE JOURNAL OF CONSUMER AFFAIRS

TABLE 5
Results of Chi-Square Tests (Probability Values in Parentheses)
Table Service Fast Food
Economic and demographic 258.26 (.000) 284.13 (.000)
Nutrition 18.60 (.136) 51.07 (.000)

set at 1, except TASTE and PORTION set at 0; nonbinary nutrition var-

iables are set at their 75th percentile value. For the low consumer, all
binaries are reversed, and the continuous measures set at the 25th percen-
tile. In both cases, the remaining variables are at their sample means. Based
on equation (1), a 2standard deviation confidence interval was constructed
for mean visits by each consumer type to each type of restaurant in a 30-day
period.
These appear in Table 6. For table service, there is virtually no differ-
ence between the consumer types, with the high-nutrition interval entirely
contained with the low-nutrition interval. It is quite the contrary for fast
food: the intervals have no overlap, with much smaller values for high
nutrition. The means indicate that the typical low-nutrition consumer makes
at least two more fast food visits per month than does the high-nutrition
consumer. For table service, there is virtually no difference. Clearly, those
with higher nutrition concerns are more likely to choose a table service
restaurant when they dine out, while those without these concerns are some-
what more inclined to fast food.

Concluding Remarks

In this paper, we have examined a question that has been studied several
times over the past three decades: the factors behind the decision to dine
away from home. The study differs from previous work by including var-
iables measuring nutrition attitudes, behavior, and knowledge as well as
demographic and economic factors. Some people believe that restaurant
meals have lower nutritional value than meals prepared at home, and res-
taurant food is often linked to the growing obesity problem, especially fast

TABLE 6
Two-Standard Deviation Confidence Intervals for Mean 30-Day Visits by
Low-Nutrition and High-Nutrition Consumers
Table Service Fast Food
Consumer type Lower Mean Upper Lower Mean Upper
Low nutrition 3.99 4.77 5.69 4.59 5.47 6.53
High nutrition 3.62 4.57 5.78 2.57 3.28 4.18
WINTER 2006 VOLUME 40, NUMBER 2 389

food. Thus, the hypothesis examined is that consumers concerned with

nutrition are less likely to dine out, with fast food particularly affected.
We found support for this. Although the variables that have been found
most important in prior studies, such as income, time value, age, and gen-
der, continue to play the primary role in FAFH demand, our results show it
is influenced by nutrition concerns as well. The effect is much more pro-
nounced for fast food, and more consistent in direction. Nutrition-focused
consumers make fewer visits to fast food outlets. In the case of table ser-
vice, nutrition is less of a factor, and the direction of effect is ambiguous: if
anything it appears to be more positive than negative.
Our conclusion is that negative publicity regarding the nutritional effects
of FAFH has adversely affected the demand for fast food, but the effect on
table service has been inconsequential. This is not a surprising result,
because fast food has become a symbol of high-fat, low-nutrition dining.
While this may be justified, that consumers obtain better nutrition at table
service restaurants has not been demonstrated. Indeed, the limited informa-
tion available suggests little difference, with table service possibly worse
(Binkley 2003; Lin and Frazao 1997). Certainly, the issue needs clarifica-
tion. If table service is no better, consumers may mistakenly believe that as
long as they avoid fast food, they need not be greatly concerned with their
diet when dining out.
The last point is important because, according to a recent United States
Department of Agriculture study, demographic trends favor the table ser-
vice sector. Because of the aging of the population, rising incomes, and the
continuing decline in household size, demand for table service meals will
grow faster than fast food demand (Stewart et al. 2004). Our results for
these variables support this. In addition, if nutritional concerns continue
to grow, which they seem likely to do, the results of this study suggest
the shift to table service may be even stronger. However, our results also
suggest that the recent addition of fruit and salad items to fast food menus is
likely to counteract these trends. Fast food firms may serve themselves and
consumers by continuing in this direction.

ENDNOTES

1. http://www.ers.usda.gov/briefing/CPIFoodAndExpenditures/.
2. The CSFII often surveyed more than one person in a household. However, the DHKS was lim-
ited to one person in any household. Thus, our sample did not include any individuals from the same
household.
3. For more details on the method, see Dong et al. (2000).
4. We tested for overdispersion using the regression-based test presented in Greene and rejected the
null of no overdispersion (Greene 2003, 884885).
390 THE JOURNAL OF CONSUMER AFFAIRS

5. This is more than implied by the sample means presented in Table 2. Reasons for this include our
exclusion of snacks and meals from other sources, such as cafeterias.
6. Also, all individuals who dine out frequently do not become overweight. This implies that any-
one who does gain weight due to dining out is doing something different from others, e.g., choosing
fattier foods or eating more. That is, ultimately it is due to diners choice. We also note that BMI has
been used in similar studies (e.g., Variyam, Blaylock, and Smallwood 1996; Wilde, McNamara, and
Ranney 1999).
7. The 23 are artichokes, asparagus, broccoli, brussels sprouts, cauliflower, eggplant, kale, swiss
chard, okra, spinach, summer squash, winter squash, yams, turnips, avocado, grapefruit, cantaloupe,
honeydew, watermelon, nectarines, pears, plums, and rhubarb.
8. DISFAT is the average of three questions asking how often fat (e.g., butter, sour cream) is added
to baked potatoes, cooked vegetables, and breads. 1 is never and 4 is almost always. SUBFAT is the
average of six questions, asking how often the respondent substitutes low fat for regular versions of lunch
meat, milk, cheese, ice cream, and salad dressing, and whether fruit is substituted for regular versions.
9. This attitude is exemplified by a respondent in a recent New York Times survey on food label
usage. I dont need to read nutrition labels closely to know doughnuts are bad for me .. I just sort of
know what would be good and what wouldnt (Burros 2004).

REFERENCES

Americans Dining-Out Habits. 2000. Restaurants, USA, November. http://www.restaurant.org/rusa/

magArticle.cfm?ArticleID=138.
ABC News-Time Poll, May 30. http://www.abcnews.go.com/images/pdf/954a1ObesityinAmerica.pdf.
Becker, Gary S. 1965. A Theory of the Allocation of Time. Economic Journal, 75 (September):
493517.
Binkley, James K. 2003. A Day without FAFH Is a Day with Better Nutrition. Paper presented at the
Meeting of the American Agricultural Association, Montreal, Quebec.
Bowman, S.A., S.L. Gortmaker, C.B. Ebbeling, M.A. Pereira, and D.S. Ludwig. 2004. Effects of Fast
Food Intake and Diet Quality among Children in a National Household Survey. Pediatrics, 113 (1):
112118.
Brown, Debra J. and Lee F. Schrader. 1990. Cholesterol Information and Shell Egg Consumption.
American Journal of Agriculture Economics, 72 (3): 548555.
Burros, Marian. 2004. Read Any Good Nutrition Labels Lately? New York Times, September 1, (14).
Burton, Scott and Elizabeth H. Creyer. 2004. What Consumers Dont Know Can Hurt Them: Consumer
Evaluations and Disease Risk Perceptions of Restaurant Menu Items. The Journal of Consumer
Affairs, 38 (1): 121145.
Byrne, Patrick J., Oral Capps, Jr. and Atanu Saha. 1998. Analysis of Quick-Serve, Mid-Scale, and Up-
Scale Food Away from Home Expenditures. International Food and Agribusiness Management
Review, 1 (1): 5172.
Chern, Wen E., Edna T. Loehman, and Steven T. Yen. 1995. Information, Health Risk Beliefs, and the
Demand for Fats and Oils. Review of Economics and Statistics, 77 (3): 555564.
Dong, Diansheng, Patrick J. Byrne, Atanu Saha, and Oral Capps, Jr. 2000. Determinants of Food-Away-
From-Home (FAFH) Visit Frequency: A Count-Data Approach. Journal of Restaurant & Food-
service Marketing, 4 (1): 3146.
Ebbin, Robert. (2000). Americans Dining-Out Habits. Restaurants, USA, November, http://
www.restaurant.org/rusa/magArticle.cfm?ArticleID=138.
Fatter Than Ever. 1996. Consumer Reports, 8 (August): 8588.
Greene, William H. 2003. Econometric Analysis. 4th edition. New York: Prentice-Hall.
Hedley, Allison A., Cynthia L. Ogden, Clifford L. Johnson, Margaret D. Carroll, Lester R. Curtin, and
Katherine M. Flegal. 2004. Prevalence of Overweight and Obesity among US Children, Adoles-
cents, and Adults, 19992002. Journal of the American Medical Association, 291 (23):
28472850.
WINTER 2006 VOLUME 40, NUMBER 2 391

Ippolito, Pauline M. and Alan D. Mathios. 1990. Information, Advertising, and Health: A Study of the
Cereal Market. Rand Journal of Economics, 21 (3): 459480.
Jekanowski, Mark D., James K. Binkley, and James S. Eales. 2001. Convenience, Accessibility, and the
Demand for Fast Food. Journal of Agricultural and Resource Economics, 26 (1): 5874.
Kinsey, Jean. 1983. Working Wives and the Marginal Propensity to Consume Food Away from Home.
American Journal of Agricultural Economics, 65 (1): 1019.
Lin, Biing-Hwan and Elizabeth Frazao. 1997. Nutritional Quality of Foods at and Away from Home.
Food Review, 2: 3340.
McCracken, Vicki and Jon Brandt. 1987. Household Consumption of Food Away from Home: Total
Expenditure and by Type of Food Facility. American Journal of Agricultural Economics, 69 (1):
274284.
Nayga, Rodolfo M., Jr. and Oral Capps, Jr. 1994. Impact of Socio-Economic and Demographic Factors
on Food Away from Home Consumption: Number of Meals and Type of Facility. Journal of Res-
taurant & Foodservice Marketing, 1 (2): 4569.
Odesser-Torpe, Marilyn. (2005). Cause Its Good for You. QSR Magazine, 73 (February). http://
www.qsrmagazine.com/issue/73/fruit.phtml.
Paeratakul, Sahasporn, Daphne P. Ferdinand, Catherine M. Champagne, Donna H. Ryan, and George
A. Bray. 2003. Fast-Food Consumption among US Adults and Children: Dietary and Nutrient
Intake Profile. Journal of the American Dietetic Association, 103 (10): 13321339.
Park, Jeahong and George C. Davis. 2001. The Theory and Econometrics of Health Information in
Cross Sectional Nutrient Demand Analysis. American Journal of Agricultural Economics,
83 (4): 840851.
Prochaska, F. and R. Schrimper. 1973. Opportunity Cost of Time and Other Socioeconomic Effects
on Away from Home Food Consumption. American Journal of Agricultural Economics, 55 (3):
595603.
Read Any Good Nutrition Labels Lately? 2004. New York Times, September 1.
Redman, Barbara. 1980. The Impact of Womens Time Allocation on Expenditure for Meals Away
from Home and Prepared Foods. American Journal of Agriculture Economics, 62 (2): 234237.
Schlosser, Eric. 2001. Fast Food Nation. New York: Houghton Mifflin.
Stewart, Hayden, Noel Blisard, Sanjib Bhuyan, and Rodolfo M. Nayga, Jr. 2004. The Demand for Food
Away from Home: Table Service or Fast Food? Agricultural Economic Report Number 829. Wash-
ington, DC: Economic Research Service, USDA.
The New, Healthier Fast Food. 2004. Consumer Reports, 6 (April): 89.
US Department of Agriculture, Agricultural Research Service. 2000. Continuing Survey of Food Intakes
by Individuals 199496. Springfield, VA: National Technical Information Service. CD-ROM.
Variyam, Jayachandran N., James Blaylock, and David Smallwood. 1996. A Probit Latent Variable
Model of Nutrition Information and Dietary Fiber Intake. American Journal of Agriculture Eco-
nomics, 78 (3): 628663.
Variyam, J.N., Y. Shim, and J. Blaylock. 2001. Consumer Misperceptions of Diet Quality. Journal of
Nutrition Education, 33 (6): 314321.
Warner, Melanie. You Want Any Fruit with That Big Mac? 2005. New York Times, February 20.
Wilde, Parke, Paul McNamara, and Christine K. Ranney. 1999. The Effect of Income and Food Pro-
grams on Dietary Quality: A Seemingly Unrelated Regression Analysis with Error Components.
American Journal of Agricultural Economics, 81 (4): 959971.
Yen, S. 1993. Working Wives and Food Away from Home: The Box-Cox Double Hurdle Model. Amer-
ican Journal of Agricultural Economics, 75 (4): 884895.