Breakfast, lunch, and dinner expenditures away from home in the United States

Food Policy 38 (2013) 156–164

Contents lists available at SciVerse ScienceDirect

Food Policy journal homepage: www.elsevier.com/locate/foodpol

Breakfast, lunch, and dinner expenditures away from home in the United States Miaoru Liu a, Panagiotis Kasteridis b, Steven T. Yen c,⇑ a

School of Economic Sciences, Washington State University, Pullman, WA 99164-6210, USA Centre for Health Economics, University of York, Heslington, York YO10 5DD, UK c Department of Agricultural and Resource Economics, The University of Tennessee, Knoxville, TN 37996-4518, USA b

a r t i c l e

i n f o

Article history: Received 28 June 2011 Received in revised form 26 September 2012 Accepted 28 November 2012 Available online 5 January 2013 Keywords: Food away from home Censoring Sample selection system

a b s t r a c t This study investigates the differentiated effects of economic and socio-demographic variables on food away from home (FAFH) expenditures by type of meal among different types of households in the United States. Using data from the 2008 and 2009 Consumer Expenditure Surveys, the systems of expenditures on breakfast, lunch, and dinner are estimated with a multivariate sample selection procedure. Statistical significance of error correlations among equations justifies estimation of the sample selection systems. Income, work hours, race, education, geographic region, and household composition are important determinants of FAFH expenditures. Income contributes to FAFH expenditures of all meal types implying that the future of FAFH industry is tied to macroeconomic conditions. More education is associated with increased expenditures for FAFH lunch and dinner. Effects of the Supplemental Nutrition Assistance Program (SNAP) are negligible. Ó 2012 Elsevier Ltd. All rights reserved.

Introduction Americans now spend nearly half of their food dollars on food away from home (FAFH). Total spending on FAFH at eating and drinking places amounted to $433.5 billion in 2010. As a share of total food expenditure, FAFH for all families and individuals rose from 32.0% in 1980 to 41.3% in 2010. In real dollars, however, expenditure reached a peak during 2007 (at $410.5 billion in 2007 dollars) but has declined since (USDA-ERS, 2011). The food consumption trends in many other countries are also converging to the pattern in the US, due to rapid income growth, urbanization and globalization. Pingali (2007) summarizes the transformation of Asian diets and food systems and indicates the rising popularity of eating outside the home in many Asian countries. Bai et al. (2012) suggest that rising income and demographic factors have contributed to the tremendous increase in FAFH consumption by households in China. Continuous upward trends in FAFH consumption are also found in many other countries, such as Malaysia (Tan, 2010), Spain (Angulo et al., 2007; Mutlu and Gracia, 2006), Greece (Minhalopoulos and Demoussis, 2001), and Turkey (Gäl et al., 2007). The literature has identified a variety of economic and sociodemographic factors that are potential determinants of FAFH consumption including household income, household size and structure, and household manager’s characteristics such as working ⇑ Corresponding author. Tel.: +1 865 974 7474; fax: +1 865 974 4829. E-mail addresses: [email protected] (M. Liu), [email protected] (P. Kasteridis), [email protected] (S.T. Yen). 0306-9192/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.foodpol.2012.11.010

hours, age, education level, race, ethnicity, and region of residence (McCracken and Brandt, 1987; Byrne et al., 1998; Stewart et al., 2004). Food policies implemented by government agencies, such as the Supplemental Nutrition Assistance Program (SNAP), may also have an impact on FAFH. During the 2007–2009 recession, which encompassed falling incomes, significant increase in unemployment, relatively high food prices, and high participation rates in the federal food and nutrition programs, Americans spent relatively less on eating out in their food expenditures (Kumcu and Kaufman, 2011). A recent study of demographic trends in the US showed that several of the demographic factors triggering FAFH changed remarkably in the 2000s (Cherlin, 2010). For instance, increasing proportions of single parents, Hispanic and Asian immigrants, and the elderly in the population marked significant changes in family and ethnic composition. Continuation of rapid demographic changes may result in a new upward trend in FAFH consumption in the next few years. This possibility escalates the concerns about the public health implications of dining out. Empirical literature has suggested that FAFH is less healthy than food at home (e.g., Mancino et al., 2009), and many policies have been implemented in order to raise the public awareness of the benefits of a healthy diet. Therefore, there remains great interest among economists, academicians, away-from-home foods sector participants, and policy makers in FAFH expenditures. The effects of economic and demographic factors on FAFH may vary with different types of meals away from home, because certain foods are more likely to be consumed at particular times than others. Jensen and Yen (1996) posit that some food items

M. Liu et al. / Food Policy 38 (2013) 156–164

are consumed at certain meals and times of the day but not traditionally at others and, therefore, differences might be expected. A good understanding of the factors that influence FAFH by type of meal is both timely and important, for explaining changes in eating patterns in the US, for creating successful marketing and promotional campaigns, for the design and implementation of policy intervention programs, and for making predictions about the future of the foodservice industry. We explore the differential impacts of economic and demographic characteristics on household FAFH expenditures by type of meal, viz., breakfast, lunch and dinner, among different household types under current economic conditions.1 The vehicle of our analysis is the sample selection system (Stewart and Yen, 2004; Yen, 2005) which allows an investigation of the expenditures jointly. We consider the logarithmic transformation in the dependent variables which, as a variance-stabilizing transformation, accommodates potential non-normality and heteroscedasticity in the error terms and facilitates estimation of the system (Yen and Rosinski, 2008). Theoretical framework Our empirical model is derived by extending the discrete random utility theory (Pudney, 1989). Each individual is assumed to maximize the random utility subject to a fixed budget m:

maxfUðDq; c; hÞjp0 q þ c ¼ mg q;c

ð1Þ

where q = [q1, . . . , qn]0 is the quantity vector with positive prices p = [p1, . . . , pn]0 , c is a composite commodity for other goods with price normalized at unity, h is a vector of demographic variables, and D = diag(d1, . . . , dn) is a diagonal matrix with each binary variable di indicating if an individual is a potential consumer of qi. Assume the utility function U(Dq, c; h) is regular strictly quasiconcave and has positive first partial derivatives with respect to positive elements of Dq and c. Then, solving Eq. (1) yields the notional demand q for FAFH, which is optimal demand without a non-negativity constraint. This constrained utility maximization problem motivates specifications for the demand functions. When an individual can be a potential non-consumer of qi, optimum qi occurs in the interior of the choice set that corresponds to di = 1 and qi = 0 when di = 0 since price pi is assumed positive. In this case, censoring in qi is governed by a sample selection mechanism. Express the notional demand with a latent consumption equation, and denote as x the vector of income and demographic variables (with corresponding parameter vector b) affecting quantity demanded. Since prices do not vary in the sample used so they are absorbed in the constant term and, further, we also consider the demand equations in expenditure forms. Let random errors vi reflect the unobservable. Then, the first-order (linear) approximation to latent expenditures yi are expressed by the latent equations

yi ¼ x0 bi þ v i ;

i ¼ 1; . . . ; n:

ð2Þ

Income is a key determinant affecting FAFH consumption positively for all groups, since households with higher incomes face looser budget constraints. Demographic variables include age, race, education, and working hours, household composition, home ownership,2 season, geographical regions, and participation status in SNAP. 1 Snacks are not included in breakfast, lunch, or dinner. Information on snack expenditures is collected separately. With the growing importance of snacks in American diet (Piernas and Popkin, 2010; Sebastian et al., 2011), snacks could have been included in the analysis as an additional type of meal. However, given the complexity of the econometric framework, we focus on a smaller system of the three major meal types. 2 Individual characteristics are those of the reference person for a single-person household and of the husband for a husband-wife household. Working hours for husband and wife are both included for a husband-wife household.

157

Since time value plays an important role in food consumption decisions (Becker, 1965), work hours of both husband and wife (or of the only person in single-person households) are included. For parameter identification, we use work hours as a unique variable(s) in the selection equation. Byrne et al. (1998) argue that market labor hours are assumed to have a varying and positive effect on the decision to consume FAFH, due to the limited time available for household production. However, once a decision is made to consume, the number of hours worked plays little role in determining expenditure level. A dummy variable for SNAP participation status is included to reflect the impact of the program on FAFH. Many state agencies provide nutrition education as a part of the SNAP to assist recipients in making healthy food and active lifestyles choices. Thus, SNAP participation might decrease consumption of FAFH as participants opt for healthier diets. Also, since SNAP benefits cannot be used for the purchase of FAFH and further, since children from SNAP-eligible households would qualify for free school meals, participation in the program may reduce consumption of FAFH for the recipients. On the other hand, food benefits may free up resources to spend on FAFH as well as other goods. The net effect of SNAP on FAFH is thus unclear. Household composition variables are important determinants in demand analysis, the effects of which vary with different types of meals (Jensen and Yen, 1996). Urban residency can also play a role, because urban families have been found to consume more FAFH (e.g., McCracken and Brandt, 1987; Yen, 1993), due to the metropolitan life style and better access to dining facilities. Homeowners, on the one hand, may consume more food away from home because of greater financial stability. On the other hand, they may have less cash flow which diminishes FAFH expenditure (Soberon-Ferrer and Dardis, 1991; Yen, 1993). Tastes and eating habits may differ by race. Because food preferences and other unobserved characteristics may differ across geographical regions and seasons, dummy variables indicating regions and seasons will be included to account for these differences (e.g., Jensen and Yen, 1996; Stewart and Yen, 2004). Due to the absence of prices in a single cross section, these regional and seasonal dummy variables will also accommodate regional and seasonal price variations. Gender and age are expected to influence FAFH expenditures to different extents by meal. Finally, education is expected to affect FAFH expenditures, as it is correlated with labor market participation, income and likely different preferences for location of eating.

Data and sample Data are drawn from the 2008 and 2009 Consumer Expenditure Surveys (CES) (US BLS, 2009, 2010), which provide consecutive 2week information on FAFH expenditures. We focus on a system of three main meals – breakfast, lunch, and dinner, each of which includes expenditures by sources, viz., at fast food, full service, vending machine, employer, board, and catered affairs. Also included in the survey are economic and demographic characteristics of the households. After removing households with missing values for important information such as household type, the full sample consists of 11,674 observations. In order to address the fact that household expenditure reflects spending by any member of the household and not just one member, the full sample is segmented into three sub-samples by household types: (i) 4592 households with a single person or a single parent with children; (ii) 3950 husband–wife households without children (at home); and (iii) 3132 husband–wife households with children. All other households such as single parents with all children older than 17 and families of unrelated adults are excluded. Among the first group, there are 728 households with a single parent, for which a dummy variable is included to capture the

158

M. Liu et al. / Food Policy 38 (2013) 156–164

effects of single parenthood. The dependent variables are biweekly per capita expenditures on breakfast, lunch, and dinner, classified by place of expenditure, not eating. The percentage of households consuming lunch away from home is the highest among the three types of meals for every household type, but the total per capita amount spent on dinner exceeds those on other meals for all household types. Appendix Table A1 presents sample statistics of household per capita FAFH expenditures by type of meal, and Table A2 provides the definitions and sample statistics of explanatory variables. Econometric procedure The large proportions of zero expenditures in our samples mandate a proper treatment for censoring of the dependent variables. Earlier studies employed models such as the Tobit model (McCracken and Brandt, 1987), single-hurdle model (Yen, 1993), double-hurdle model (Jensen and Yen, 1996; Mutlu and Gracia, 2006), and sample selection system (Stewart and Yen, 2004) to accommodate the censored data. All these models, except Stewart and Yen (2004), are estimated with single-commodity equations. In this study, the system of expenditures is estimated with a system procedure to accommodate censoring in the dependent variables to improve statistical efficiency of parameter estimates and to capture the interaction among different types of meals. One system approach is the Tobit system. However, the Tobit model is undesirable in coping with zero observations because the same parameters and variables determining probability of zero and positive outcomes also determine the level. In addition, the relative effects of two continuous explanatory variables on the probability, conditional mean, and unconditional mean of the dependent variable are identical and equal to the ratio of the corresponding coefficients (Yen, 2009). Following Yen (2005) and Stewart and Yen (2004), we use a sample selection system. Consider a three-outcome system where each outcome variable yi (FAFH expenditure) is governed by a binary sample-selection rule

log yi ¼ x0 bi þ v i

if z0 ai þ ui > 0

ð3Þ

if z0 ai þ ui 6 0

yi ¼ 0

where i = 1, 2, 3 for breakfast, lunch, and dinner, respectively, z and x are column vectors of explanatory variables, ai and bi are conformable parameter vectors, and ui and vi are random errors. Assume the concatenated error vector [u0 , v0 ]0 = [u1, u2, u3, v1, v2, v3]0 is distributed as six-dimensional normal with zero means, standard deviations [1, 1, 1, r1, r2, r3]0 , and correlation matrix R = [qij] In contrast to the linear functional form in Stewart and Yen (2004), each dependent variable yi is transformed by natural logarithm. Such transformation is common in estimation of endogenous selection and switching regression model (Heckman and Honoré, 1990; Yen, 2005), which ameliorates potential nonnormality and heteroscedasticity of the error term (Yen and Rosinski, 2008). The log-transformed model is also found to perform better in fitting our data than the untransformed model (discussed below). The sample likelihood contribution of an all-positive regime is (Yen, 2005, Eq. (4)) 3 Y

L¼

! y1 i

i¼1



Z

1

z0 a3

1 i

r

/3 ðw1 ; w2 ; w3 ; RÞ

Z

Z

1

z0 a

1

1

z0 a2

hðu1 ; u2 ; u3 jv 1 ; v 2 ; v 3 Þ du3 du2 du1

ð4Þ

where wi ¼ ðlog yi  x0 bi Þ=ri ; /3 ðw1 ; w2 ; w3 ; RÞ is trivariate standard normal probability density function with random variates Q w1 ; w2 ; w3 and correlation matrix R, and ð 3i¼1 y1 i Þ is the Jacobian 0 of the transformations from ½v 1 ; v 2 ; v 3  to ½log y1 ; log y2 ; log y3 0 : In (4), the conditional density h(u1, u2, u3|v1, v2, v3) is trivariate normal

(Kotz et al., 2000, pp. 111–112) which can be integrated by usual means (Yen, 2005). The sample selection system nests two restricted specifications: (i) an independent model which corresponds to parametric restrictions qij = 0 for all i–j viz., with all error correlations equal to zeros; (ii) a pairwise selection system which corresponds to qij = 0 for all i–j except q41 –0; q52 –0 and q63 –0 These restricted models can be estimated by imposing the above parametric restrictions. In addition, the independent model can be estimated by the probit model based on the binary (0/1) outcome related to each yi using the whole sample, and ordinary least squares for each log(yi) using the truncated sample (conditional on yi > 0). The pairwise selection system consists of three bivariate sample selection models (Heckman, 1979) for the three expenditures which can be estimated as such separately. Tests of the sample selection system against the two nested models can be done with the Wald, likelihood ratio (LR), and Lagrange multiplier (LM) tests (Engle, 1984). Marginal effects of probabilities, conditional levels, and unconditional levels are calculated to facilitate interpretation of the effects of explanatory variables. For each good i, the probability, conditional mean and unconditional mean of yi are (Yen and Rosinski, 2008, p. 5)

Prðyi > 0Þ ¼ U1 ðz0 ai Þ

ð5Þ

Eðyi jyi > 0Þ ¼ expðx0 bi þ r2i =2Þ U1 ðz0 ai þ quiiv ri Þ=U1 ðz0 ai Þ

ð6Þ

Eðyi Þ ¼ expðx0 bi þ r2i =2Þ U1 ðz0 ai þ quiiv ri Þ

ð7Þ

where U1() is cumulative distribution function of the unit normal. Differentiating (differencing) Eqs. (5)–(7) with respect to explanatory variables x and z gives the marginal (discrete) effects of continuous (discrete) explanatory variables (Yen and Rosinski, 2008, pp. 5), which can be evaluated for all observations and averaged over the sample. Standard errors of marginal effects are calculated by a mathematical approximation procedure known as the delta method. Estimation results and specification tests During preliminary analysis, the untransformed model could not be estimated (failure to converge). We use Vuong’s (1989) nonnested specification test to compare the independent versions (discussed above) of the log-transformed and untransformed models. The log-transformed model performs better in fitting the data, with standard normal statistic z > 15, for all three samples considered, lending support for the logarithmic transformation.3 Given the preferred log-transformed specification, we then compare the Tobit system and the sample selection system, also with Vuong’s test. The results, with standard normal statistic z = 3.29 for husband–wife households with children and z > 10 for the other two samples, suggest that the sample selection system performs better than the Tobit system. The Tobit system estimates are not presented due to space consideration and we focus on results of the sample selection system. The next empirical task is to test for equality of parameters across household groups and determine poolability of samples. This task is accomplished with a likelihood ratio (LR) test, similar to Chow test in linear regression models. The hypothesis of equal slope coefficients is rejected (LR = 552, df = 246, p-value < 0.0001), which justifies estimation of the model by segmented samples.4 3 The untransformed model is estimated with two-step procedure (Shonkwiler and Yen, 1999), a less efficient alternative to the ML procedure, for single-person households and produced notably different marginal effects. Two-step results are available upon request. 4 Household composition variables are unavailable while gender and single parenthood are unique in single-person households. These variables are excluded for the purpose of this test.

M. Liu et al. / Food Policy 38 (2013) 156–164

In addition, marginal effects (discussed below) differ greatly across household groups, which can be masked by the use of a pooled sample. We then explore the appropriateness of the sample selection system vis-à-vis the nested independent models and pairwise selection system. The error correlations between selection and corresponding level equations are significant at the 10% levels of significance or lower for two of the three meals for all types of households. In general, among the fifteen error correlations, at least ten of them are significant at the 10% significance level or lower. These error correlations are also jointly significant by formal statistical tests. For husband–wife households with children, results of the tests suggest rejection of the independent system (Wald = 1898.63, LR = 2116.22, LM = 1167.68, df = 15) and pairwise selection system (Wald = 1560.83, LR = 1179.10, LM = 821.48, df = 12), all with p-values < 0.0001. Both nested specifications are also rejected for the other two samples.

Marginal effects Marginal effects of explanatory on the probability, conditional level, and unconditional level for the sample selection system are presented in Tables 1–3.5 The marginal effects for most variables differ substantially among meal occasions and across household types. Some of the results, such as the effects of income, gender, and work hours, are in agreement with previous studies. Some disagreements are also found due to the empirical procedures and methods, the definition of dependent variables (e.g., confounded by categorization of snacks) and explanatory variables and the treatment of the data. The impacts of age on FAFH differ notably among types of meal and across household types. Among single-person households, age has negative effects on the probability of consuming all meals among single-person households; it increases the level of expenditure on breakfast but decreases the level of lunch conditional on consumption. Overall, for these single-person households, the effects of age on unconditional levels of lunch and dinner are both negative.6 These negative effects likely reflect the consuming activity of single elderly, who generally are less likely to eat away from home than their younger cohort. Negative effects of age are also seen on the probability and unconditional level of dinner among husband–wife households without children, while positive effects are found on the conditional and unconditional levels of lunch among husband–wife households with children. As expected, per capita income has positive effects on probabilities of FAFH, conditional levels, and unconditional levels of expenditures on all meals at the 1% level of significance for all groups. Income elasticities reported for most other studies, though not directly comparable with marginal effects of per capita income, also highlight the importance of household income in FAFH expenditure. Overall, the marginal effects of income on FAFH are relatively small, with a $10,000 increase in per-capita income increasing any meal by less than $2 per person for 2 weeks. These positive and small effects are in agreement with findings from previous studies for the US (e.g., McCracken and Brandt, 1987; Jensen and Yen, 1996). Larger effects of income are reported for Spain (Angulo et al., 2007). Supporting our hypothesis on the role of time spent at work, the number of hours worked has mostly positive impacts on FAFH. Husbands’ work hours generally have greater effects than wives’. 5 The Tobit system produces notably different marginal effects, obviously caused by its restrictive parameterization. Complete results for the Tobit system are available upon request. 6 Note that the effects of age in single person households are estimated while holding constant being the parent of a child less than 17 years old.

159

Similar results are found in previous studies, though different variables are used. For instance, Stewart and Yen (2004) find positive effects of work hours of household managers on the probabilities of consuming FAFH. Yen (1993) find that wife’s work hours have positive effects on lunch and dinner away from home. The positive and significant effects of wife’s opportunity cost of time are consistent with studies for China (Bai et al., 2012) and Spain (Mutlu and Gracia, 2006). However, few studies take into account the impact of husbands’ work hours. Race has differentiated effects across household groups. In general, households of white and other races consume more FAFH than black households, especially for dinner away from home. The effects of race are most notable among husband–wife households with children. For instance, compared their black counterparts, husband–wife households of the white (other) race are 10.32% (12.71%) more likely to have dinner away from home, spend $10.10 ($24.04) more conditional on consuming, and $9.69 ($23.29) more overall per 2 weeks. Whereas education generally does not affect breakfast away from home, especially for single-person households and husband–wife households with children, higher education has positive effects on lunch and dinner away from home for all three household groups. SNAP has no effect on any meal for husband–wife households without children. For both single-person households and husband–wife households with children, SNAP has negative effects on the probabilities and levels of lunch and dinner away from home. The negative effect of SNAP on lunch is expected for households with children as those on SNAP would qualify for free school meals in the US. SNAP also decreases the conditional level of expenditure on breakfast among single-person households. By segmenting the sample, differentiated effects of home ownership are found across household types. Home ownership increases probabilities of consuming breakfast, lunch, and dinner away from home among husband–wife households without and with children; it also increases expenditures on lunch for husband–wife households with children. Our positive effects of home ownership are generally consistent with finding by Jensen and Yen (1996). Home ownership does not affect any FAFH expenditure by single-person households. Geographic variables have mixed effects. Compared with those residing in the West, single-person households in the Northeast spend more on dinner; those in the Midwest have lower expenditures on lunch and dinner, while those in the South spend less on breakfast. Husband–wife households without children from the Midwest and South spend less on breakfast, while those from the Northeast and South spend more on dinner. Geographical differences are found in husband–wife households with children as well, which may reflect price differences across regions. The effects of urbanization are fairly scant. Relative to their rural counterparts, single-person households in urban areas spend more on lunch, while positive effects are found in dinner among husband–wife households without children, and in lunch among husband–wife households with children. Byrne et al. (1996) also found urban households consume more FAFH than their rural counterparts. For the single-person sample, the other two significant variables are dummy variables indicating gender and single parenthood. Compared with single women, single men have higher expenditures on FAFH than women, except the probability of consuming lunch away from home (insignificant). The effects are most notable for dinner among the three meal occasions, with a male spending $6.27 more per 2 weeks than females. Compared with single persons without children, single parents are more likely to consume FAFH but they spend less conditional on consumption. Raising children can take a large part of time and income, tightening both time and budget constraints. Therefore, single parents tend to dine out

160

M. Liu et al. / Food Policy 38 (2013) 156–164

Table 1 Marginal effects of explanatory variables on probabilities, conditional levels, and unconditional levels by type of meal: sample selection system for single-person households.a Variable

Breakfast Prob. (100)

Continuous explanatory variables Age/10 0.914* (0.494) Income/10 0.949*** (0.224) Hour/10 2.360*** (0.386) Binary explanatory variables Northeast 1.424 (2.318) Midwest 0.416 (2.095) South 1.861 (2.047) Spring 1.909 (1.992) Summer 2.627 (2.076) Fall 2.611 (2.018) White 1.133 (2.069) Other race 3.793 (3.820)
Lunch Cond. level

Uncond. level

Prob. (100)

Dinner Cond. level

Uncond. level

Prob. (100)

Cond. level

Uncond. level

0.715** (0.282) 0.269** (0.118) 0.574*** (0.126)

0.120 (0.133) 0.243*** (0.060) 0.569*** (0.097)

2.307*** (0.486) 1.159*** (0.220) 2.083*** (0.403)

0.695* (0.390) 1.144*** (0.161) 0.053 (0.198)

1.082*** (0.280) 1.027*** (0.120) 0.564*** (0.162)

3.914*** (0.488) 1.395*** (0.218) 1.479*** (0.410)

0.558 (0.613) 2.045*** (0.254) 0.049 (0.266)

1.957*** (0.414) 1.740*** (0.175) 0.592*** (0.219)

0.975 (1.284) 2.889*** (1.078) 2.213** (1.105) 0.301 (1.142) 1.455 (1.246) 0.346 (1.137) 1.079 (1.122) 3.320* (1.861) 1.322 (1.371) 0.129 (0.995) 1.539 (1.572) 1.506 (2.198) 0.126 (0.927) 5.630*** (0.898) 2.841** (1.371) 9.569*** (0.747) 0.104 (0.801)

0.580 (0.611) 1.109** (0.496) 1.075** (0.504) 0.404 (0.530) 0.951 (0.584) 0.529 (0.536) 0.557 (0.507) 1.671** (0.785) 0.882 (0.591) 0.192 (0.455) 1.040 (0.772) 0.583 (0.981) 0.015 (0.418) 2.510*** (0.410) 1.010 (0.624) 2.991*** (0.353) 0.002 (0.366)

0.030 (2.336) 0.721 (2.144) 1.252 (2.085) 4.921** (1.983) 0.039 (2.071) 0.081 (2.017) 6.339*** (2.136) 8.004** (3.687) 6.457*** (2.480) 2.745 (1.732) 7.343*** (2.671) 0.663 (3.201) 0.879 (1.625) 0.092 (1.547) 1.770 (2.846) 9.734*** (2.184) 1.235 (1.427)

1.012 (1.867) 2.747* (1.613) 2.233 (1.722) 1.305 (1.614) 1.328 (1.710) 0.128 (1.633) 4.742*** (1.573) 8.497** (4.095) 2.430 (2.256) 2.893** (1.373) 8.235*** (2.610) 4.054* (2.337) 1.368 (1.319) 6.042*** (1.233) 13.182*** (1.466) 13.452*** (1.156) 0.112 (1.120)

0.624 (1.326) 1.866* (1.137) 1.723 (1.215) 2.227* (1.172) 0.796 (1.207) 0.055 (1.146) 4.507*** (1.058) 8.049** (3.145) 0.495 (1.473) 2.538*** (0.975) 7.536*** (1.970) 2.296 (1.660) 1.082 (0.920) 3.629*** (0.875) 8.281*** (0.998) 6.435*** (0.866) 0.422 (0.795)

1.048 (2.344) 0.337 (2.151) 0.587 (2.089) 2.470 (1.990) 3.538* (2.092) 0.244 (2.008) 4.031* (2.125) 0.371 (4.020) 5.504** (2.473) 4.030** (1.758) 5.387** (2.699) 1.816 (3.344) 0.104 (1.622) 4.227*** (1.565) 6.837** (2.834) 8.528*** (2.215) 1.172 (1.429)

10.140*** (3.242) 0.047 (2.595) 4.265* (2.577) 1.722 (2.600) 2.882 (2.678) 1.857 (2.450) 8.297*** (2.423) 7.923 (5.305) 2.001 (3.492) 7.945*** (2.216) 13.747*** (4.113) 5.610 (3.794) 1.828 (2.045) 7.958*** (1.897) 16.033*** (2.500) 20.434*** (1.804) 4.294** (1.759)

5.220** (2.126) 0.115 (1.718) 2.154 (1.704) 2.026 (1.723) 0.088 (1.725) 0.948 (1.608) 6.159*** (1.538) 4.653 (3.590) 1.273 (2.119) 6.148*** (1.458) 10.526*** (2.823) 3.836 (2.434) 0.989 (1.345) 6.270*** (1.274) 10.929*** (1.517) 9.329*** (1.253) 1.931* (1.162)

a

Asymptotic standard errors in parentheses. Indicate significance level at 10%. ** Indicate significance level at 5%. *** Indicate significance level at 1%. *

more often but spend less money on food, compared with singleperson households without children. Also, children eat less than adults, so the average expenditure is lower.7 In the case of husband–wife households, household composition also influences all meals away from home. For husband–wife households without children, an additional household member increases the probabilities of consuming FAFH but decreases the expenditures. For husband–wife households with children, an additional household member does not affect the probabilities of consuming FAFH, but decreases the expenditure levels significantly. The number of elderlies (age > 64) has no effect on breakfast away from home. Jensen and Yen (1996) find that the number of family members age 19–64 has no effect on the probability to consume breakfast or lunch, but a negative effect on the probability to consume dinner away from home. Our results for 7 There may be interaction, as one reviewer suggests, between working status and age for single individuals as singles with children at home are at working age and the probability of FAFH would likely increase as a result of working outside of the home. Work hour was interacted with age but was found jointly insignificant in the selection equations (p-value = 0.19).

husband–wife households with children are similar to those reported by Jensen and Yen (1996). However, the effects of the numbers of adults on the probability of consuming FAFH are all positive for husband–wife households without children, which are different from findings reported by Jensen and Yen (1996). Surprisingly, for husband–wife households with children, an additional member age < 18 does not have a significant effect on the probability of consuming lunch but has a negative effect on the lunch expenditure. The negative effect on lunch expenditure may be because these children tend to spend less than adults when they consume away from home.

Concluding remarks With the constantly evolving away-from-home food market and continued interest in consumer diet and health, information on the factors that determine FAFH in a continuously changing socio-economic environment is both timely and important. This paper examines how socio-economic variables affect household expenditures on FAFH by type of meal. The analysis is carried out

161

M. Liu et al. / Food Policy 38 (2013) 156–164

Table 2 Marginal effects of explanatory variables on probabilities, conditional levels, and unconditional levels by type of meal: sample selection system for husband–wife households without children.a Variable

Breakfast Prob. (100)

Lunch Cond. level

Uncond. level

Continuous explanatory variables Age/10 1.011 (0.892) Income/10 1.226*** (0.297) Age 18–64 2.064 (1.308) Age > 64 3.305* (2.000) Hour/10 1.722*** (0.462) SP hour/10 1.179*** (0.450)

0.260 (0.323) 0.424*** (0.105) 3.230*** (0.455) 2.557*** (0.694) 0.053 (0.112) 0.036 (0.077)

0.006 (0.178) 0.327*** (0.061) 1.236*** (0.257) 0.794** (0.382) 0.214*** (0.074) 0.146** (0.064)

Binary explanatory variables Northeast 0.045 (2.503) Midwest 1.262 (2.405) South 0.898 (2.233) Spring 4.462** (2.211) Summer 3.201 (2.244) Fall 3.368 (2.250) White 5.271 (3.366) Other race 8.894* (4.900)
0.054 (0.874) 1.916** (0.809) 1.550** (0.780) 0.353 (0.812) 0.183 (0.836) 0.056 (0.818) 0.155 (1.368) 1.220 (1.966) 1.252 (1.019) 1.224* (0.707) 2.513*** (0.819) 0.851 (1.058) 0.240 (0.919) 0.769 (2.438) 1.284** (0.569)

0.019 (0.487) 0.994** (0.439) 0.797* (0.425) 0.659 (0.460) 0.438 (0.463) 0.397 (0.457) 0.517 (0.667) 1.628 (1.199) 1.000* (0.517) 0.391 (0.389) 1.417*** (0.432) 0.420 (0.570) 1.164*** (0.443) 0.551 (1.202) 0.880*** (0.313)

Prob. (100)

Dinner Cond. level

Uncond. level

0.710 (0.820) 1.757*** (0.287) 4.414*** (1.237) 6.106*** (1.817) 1.553*** (0.412) 0.701* (0.407)

0.388 (0.473) 0.792*** (0.136) 4.177*** (0.721) 4.422*** (1.088) 0.108 (0.085) 0.049 (0.046)

0.419 (0.370) 0.919*** (0.115) 1.976*** (0.562) 1.791** (0.840) 0.401*** (0.119) 0.181* (0.107)

5.177** (2.390) 3.568 (2.292) 0.910 (2.100) 2.751 (2.000) 3.248 (2.012) 3.923* (2.008) 11.880*** (3.290) 10.456*** (3.659) 10.985*** (2.726) 0.328 (1.831) 2.664 (2.496) 2.803 (2.854) 5.067** (2.339) 0.790 (5.810) 1.758 (1.462)

0.620 (1.332) 1.103 (1.269) 0.773 (1.188) 0.642 (1.226) 0.522 (1.230) 0.018 (1.202) 1.691 (1.763) 2.060 (2.793) 2.394 (1.543) 2.236** (1.050) 3.682** (1.574) 2.176 (1.478) 1.803 (1.325) 0.066 (3.880) 0.876 (0.855)

1.501 (1.018) 1.496 (0.986) 0.344 (0.945) 1.031 (0.975) 1.052 (0.975) 0.836 (0.963) 3.527*** (1.259) 3.801 (2.320) 3.776*** (1.088) 1.620* (0.831) 1.930 (1.205) 0.976 (1.208) 2.254** (0.977) 0.120 (2.954) 0.240 (0.671)

Prob. (100)

Cond. level

Uncond. level

1.539* (0.850) 1.828*** (0.270) 2.983** (1.237) 2.670 (1.835) 1.492*** (0.437) 0.971** (0.415)

0.774 (0.787) 1.718*** (0.223) 8.035*** (1.227) 9.113*** (1.832) 0.008 (0.193) 0.005 (0.124)

1.055* (0.613) 1.796*** (0.183) 4.432*** (0.927) 5.272*** (1.379) 0.518*** (0.197) 0.338** (0.165)

0.854 (2.336) 1.188 (2.223) 4.322** (2.073) 3.898* (2.021) 2.522 (2.066) 2.312 (2.075) 10.317*** (3.267) 12.709*** (3.660) 10.958*** (2.779) 0.544 (1.832) 3.592 (2.568) 0.553 (2.919) 4.179* (2.420) 5.888 (6.730) 2.398 (1.482)

6.805*** (2.449) 0.037 (2.083) 3.727* (2.044) 0.272 (2.011) 2.913 (1.921) 1.448 (1.962) 10.099*** (2.591) 24.042*** (6.992) 7.885*** (2.432) 6.596*** (1.750) 6.991*** (2.475) 8.726*** (2.222) 0.773 (2.307) 1.729 (5.987) 0.014 (1.407)

4.301** (1.893) 0.383 (1.622) 4.070** (1.607) 1.150 (1.564) 1.149 (1.488) 0.205 (1.519) 9.686*** (1.824) 23.292*** (5.863) 8.445*** (1.655) 4.654*** (1.349) 3.341* (1.914) 5.775*** (1.701) 1.944 (1.758) 3.114 (4.498) 0.815 (1.081)

a

Asymptotic standard errors in parentheses. Indicate significance level at 10%. Indicate significance level at 5%. *** Indicate significance level at 1%. *

**

by estimating the sample selection system, which accommodates zero expenditures in the expenditure variables. A multivariate extension to the bivariate sample selection model (Heckman, 1979), the sample selection system is found to perform better than the Tobit system. Joint statistical significance of error correlations among equations is found, which justifies estimation of the sample selection systems versus the bivariate and independent alternatives. Marginal effects of explanatory variables on probabilities and conditional levels of many variables show opposite signs and different levels of significance, which highlight the importance of the sample selection system over the more restrictive Tobit system. In addition, the hypothesis of equal parameters across household types is rejected, supporting segmentation of the sample by household types in the analysis. The differentiated marginal effects of most variables on FAFH expenditures further justify the analysis by household types. Studies of FAFH expenditures have been conducted in many different countries. Although not directly comparable due to the use

of different econometric approaches and, more importantly, variables (e.g., per capita versus household expenditures), these studies generally find income and socio-demographic characteristics playing key roles in FAFH expenditures. The findings of this study can inform policy deliberations by federal and state governments. Our estimates identify segments of the population that should be targeted for nutrition education. Among single-person households, men spend more on FAFH than women. For husband–wife households without children, work hours of both husband and wife increase the probability of having a meal away from home. Work hours of the reference person have similar effect among husband–wife households with children. In single-person households, work hours play an even more important role increasing not only the probability of having a meal away from home but also the conditional expenditures on FAFH breakfast. College and graduate education has strong and positive effects among all types of households, particularly on lunch and dinner. Therefore, nutrition educators should focus their efforts on single

162

M. Liu et al. / Food Policy 38 (2013) 156–164

Table 3 Marginal effects of explanatory variables on probabilities, conditional levels, and unconditional levels by type of meal: sample selection system for husband–wife households with children.a Variable

Breakfast Prob. (100)

Lunch

Dinner

Cond. level

Uncond. level

Prob. (100)

Cond. level

Uncond. level

Continuous explanatory variables Age/10 0.473 (1.115) Income/10 1.586*** (0.534) Age < 18 0.391 (0.972) Age 18–64 1.684 (1.555) Age > 64 2.624 (3.936) Hour/10 1.177* (0.634) SP hour/10 0.811* (0.483)

0.149 (0.228) 0.408*** (0.109) 1.513*** (0.229) 0.931*** (0.321) 1.195 (0.796) 0.002 (0.071) 0.001 (0.048)

0.046 (0.138) 0.312*** (0.068) 0.752*** (0.136) 0.586*** (0.189) 0.782 (0.489) 0.077 (0.054) 0.053 (0.039)

0.035 (0.866) 2.721*** (0.499) 0.212 (0.765) 0.372 (1.206) 0.169 (2.833) 1.647*** (0.487) 0.712* (0.389)

0.820** (0.408) 1.349*** (0.206) 2.730*** (0.365) 2.429*** (0.538) 3.197** (1.478) 0.050 (0.069) 0.022 (0.031)

0.668* (0.361) 1.495*** (0.182) 2.176*** (0.324) 1.908*** (0.475) 2.610** (1.267) 0.204** (0.081) 0.088* (0.053)

Binary explanatory variables Northeast 1.523 (2.904) Midwest 0.664 (2.639) South 4.122 (2.520) Spring 1.193 (2.537) Summer 0.032 (2.569) Fall 2.215 (2.533) White 2.948 (3.666) Other race 3.767 (5.020)
0.208 (0.592) 1.712*** (0.486) 0.443 (0.511) 0.406 (0.545) 0.656 (0.496) 0.035 (0.537) 0.226 (0.721) 1.258 (0.895) 1.729** (0.872) 0.023 (0.481) 0.267 (0.650) 0.367 (0.726) 0.087 (0.511) 0.417 (0.879) 0.408 (0.373)

0.011 (0.361) 0.915*** (0.289) 0.033 (0.308) 0.129 (0.323) 0.339 (0.297) 0.160 (0.315) 0.301 (0.417) 0.847* (0.507) 0.498 (0.485) 0.125 (0.286) 0.161 (0.394) 0.124 (0.464) 0.311 (0.294) 0.293 (0.505) 0.217 (0.224)

3.889 (2.442) 0.141 (2.111) 4.985** (1.947) 0.328 (2.102) 4.624** (2.202) 2.273 (2.135) 2.771 (3.070) 0.470 (3.831) 4.637 (2.674) 3.865** (1.832) 3.875 (2.549) 0.579 (3.349) 5.963*** (2.056) 4.744 (3.527) 1.639 (1.453)

0.050 (1.108) 1.399 (0.938) 1.925** (0.965) 0.991 (0.909) 1.695* (0.898) 1.379 (0.891) 4.634*** (1.064) 8.053*** (2.602) 0.025 (1.241) 1.993** (0.847) 1.688 (1.351) 2.942** (1.367) 2.616*** (0.813) 5.470*** (1.150) 0.845 (0.671)

0.618 (0.952) 1.110 (0.823) 2.323*** (0.857) 0.849 (0.799) 2.022*** (0.777) 1.438* (0.780) 4.077*** (0.927) 6.613*** (2.288) 0.714 (1.053) 2.184*** (0.745) 1.981* (1.198) 2.450** (1.199) 2.917*** (0.705) 4.900*** (0.973) 0.440 (0.587)

Prob. (100)

Cond. level

Uncond. level

0.090 (0.942) 2.713*** (0.513) 1.365* (0.831) 1.390 (1.252) 5.412* (2.995) 1.412*** (0.533) 0.019 (0.424)

0.756 (0.566) 1.783*** (0.253) 3.946*** (0.524) 2.842*** (0.769) 5.066*** (1.962) 0.093 (0.109) 0.001 (0.028)

0.590 (0.481) 1.903*** (0.224) 3.261*** (0.438) 2.431*** (0.648) 4.935*** (1.662) 0.219* (0.115) 0.003 (0.066)

6.196** (2.636) 0.572 (2.277) 2.087 (2.166) 1.242 (2.233) 1.571 (2.280) 4.131* (2.280) 3.462 (3.137) 2.332 (4.315) 7.879*** (2.960) 4.437** (1.976) 4.537 (2.803) 0.721 (3.670) 8.363*** (2.230) 11.086*** (3.860) 3.322** (1.579)

3.065* (1.586) 1.761 (1.226) 1.807 (1.308) 1.114 (1.303) 0.592 (1.314) 1.350 (1.202) 4.768*** (1.567) 8.484** (3.347) 2.063 (1.640) 2.569** (1.164) 3.259* (1.747) 0.323 (2.528) 1.243 (1.243) 3.128 (2.450) 1.262 (0.923)

0.928 (1.293) 1.219 (1.041) 1.804 (1.100) 0.581 (1.093) 0.120 (1.094) 1.842* (0.999) 4.204*** (1.285) 5.775** (2.751) 3.061** (1.294) 2.846*** (0.974) 3.485** (1.516) 0.094 (2.081) 2.606*** (0.988) 4.357** (1.783) 0.274 (0.772)

a

Asymptotic standard errors in parentheses. Indicate significance level at 10%. Indicate significance level at 5%. *** Indicate significance level at 1%. *

**

men and households with busy working schedules, as well as those with college or higher education. Such efforts include warnings about the relatively higher levels of sodium, cholesterol, and saturated fats in FAFH meals, recommendations about healthy FAFH choices such as fruits, vegetables, milk, and oils, and educational messages about moderating consumption of fats, added sugars, and alcohol (You et al., 2009). Our estimate of the effect of SNAP participation on FAFH is also informative for policymakers whose goal is more nutritious and adequate food. Our finding suggests that the concern about SNAP promoting consumption of less healthy FAFH is groundless since effects of the program are negligible. The results of the study can also assist marketing strategies by foodservice firms. The strong impact of household composition changes on FAFH spending provides valuable information for the foodservice industry. For instance, whereas an additional house-

hold member has no effects on the probabilities to consume breakfast and lunch among households with children and positive effects on probabilities of consuming all three meals among households without children, the negative effects on the conditional and unconditional levels spent are unambiguously negative for both types of households. The overarching message is that when these households do eat out they tend to spend less on a per capita basis. Thus, promotional campaigns offering quantity discounts can be an effective tool for restaurants to attract large families (with and without children). Senior citizen discounts could also be used as age reduces the probability to have lunch or dinner away from home in single-person households. Some conclusions about future trends of FAFH consumption can also be drawn. Income contributes to FAFH expenditures of all meal types implying that the future of FAFH industry is tied to macroeconomic conditions. Some research has stated that the

163

M. Liu et al. / Food Policy 38 (2013) 156–164 Table A1 Sample statistics of household per capita FAFH expenditures by type of meal (biweekly). Expenditure

% Consuming

Full sample

Consuming sample

Mean ($)

SD ($)

Mean ($)

SD ($)

4.59 14.43 20.80

12.21 22.83 36.99

10.79 21.48 33.22

16.84 24.99 42.11

Single-person households (n = 4592) Breakfast 35.17 5.60 Lunch 58.89 16.73 Dinner 53.46 23.12

16.50 28.66 46.85

15.91 28.40 43.24

24.71 32.60 56.88

Husband–wife households (without children) (n = 3950) Breakfast 44.86 4.66 9.62 Lunch 68.23 14.05 20.46 Dinner 65.49 22.72 33.74

10.38 20.59 34.69

12.12 21.88 36.37

Husband–wife households (with children) (n = 3132) Breakfast 50.48 3.04 6.02 Lunch 78.00 11.54 13.95 Dinner 72.45 15.00 20.07

6.02 14.79 20.70

7.33 14.19 20.93

Full sample (n = 11,674) Breakfast 42.56 Lunch 67.17 Dinner 62.62

growing trend of an aging population is working against the FAFH industry, especially on fast food (e.g., Stewart et al., 2004; Stewart and Yen, 2004). Our results echo these findings as age has negative effects on dinner among husband–wife households without children and on both lunch and dinner among single-person households. While this study provides useful insights regarding FAFH consumption in the US and its policy and marketing implications, a few caveats pertain. First, the important role of snacks in today’s FAFH market is not addressed. Snacking patterns are likely to have differential effects on meals by type of meal and for different demographic groups. As a result, differences across meal occasions or in comparison to previous studies may be due to the larger share of calories (food) coming from snacking occasions. Second, the line between snack and meal is much less discernible today than previously (Piernas and Popkin, 2010; Sebastian et al., 2011). Piernas and Popkin (2010), for instance, argue that the increased frequency of snacking might be the result of people reporting snacks in cases where consumption would typically be considered a meal. Thus, there may be a fair amount of mis-reporting or measurement errors in our analysis. To address these limitations, future studies might consider adding snacks in the analysis and using more

Table A2 Definitions and sample statistics of explanatory variables. Variable

Definitions

Full sample

Single-person households

Husband–wife households without children

Husband–wife households with children

Mean

SD

Mean

SD

Mean

SD

1.14 0.89 31.06

1.93 2.25 0.05 23.55

0.98 0.62 0.27 20.07

Mean

SD

Continuous explanatory variables Household characteristics Age < 18 Number of children age < 18 Age 18–64 Number of adults age 18–64 Age > 64 Number of adults age > 64 Income After-tax income per capita in past 12 months (imputed mean, unit = $1000)

0.63 1.45 0.34 31.11

1.06 1.00 0.65 31.53

35.37

38.79

1.62 0.66 36.19

Household managers’ characteristicsa Age Age in years Working hours Hours usually worked per week by household manager

51.11 30.59

16.89 21.39

51.14 26.16

19.23 21.49

58.64 27.08

14.60 22.38

41.56 41.52

9.49 15.27

21.63

20.65

25.54

19.35

Spouses’ characteristics (wife for married households) Working hours Hours usually worked per week by spouse Dummy variables (1 = yes; no = 0) Household characteristics Homeowner Owns a home SNAP Any members received SNAP during past year Urban Resides in an urban areaa Northeast Resides in the Northeast Midwest Resides in the Midwest South Resides in the South West Resides in the West (reference) Spring Diary (survey) date occurred during spring Summer Diary date during summer Fall Diary date during fall Winter Diary date during winter (Ref.) Household managers’ characteristicsb White Race is White Black Race is Black (Ref.) Other race Race is of other race Male Gender is male
0.70 0.06 0.94 0.19 0.26 0.35 0.20 0.27 0.24 0.25 0.24

0.52 0.09 0.95 0.19 0.27 0.35 0.19 0.27 0.23 0.26 0.24

0.86 0.02 0.93 0.20 0.24 0.35 0.21 0.26 0.24 0.24 0.26

0.78 0.06 0.95 0.17 0.25 0.34 0.24 0.26 0.24 0.25 0.25

0.85 0.09 0.06

0.81 0.14 0.05 0.40 0.13 0.27 0.50 0.10 0.16 0.51 4592

0.88 0.07 0.05

0.85 0.08 0.07

0.11 0.28 0.46 0.15

0.12 0.25 0.49 0.14

0.51 3950

0.50 3132

0.12 0.27 0.48 0.13 0.06 0.51

a Urban households refer to all households living in Metropolitan Statistical Areas (MSAs) and in urbanized areas and urban places of 2500 or more persons outside of MSAs. Urban, defined in this survey, includes the rural populations within an MSA (US BLS, 2009, 2010). b Household manager is defined as the husband for a married household, and as the reference person for a single-person household.

164

M. Liu et al. / Food Policy 38 (2013) 156–164

accurate measurements of FAFH. Third, in the present study, employment of the household head and the spouse are treated as exogenous. Previous studies reported evidence of endogeneity in the analysis of FAFH and other time-saving goods and services (e.g., Bryant, 1988; Yen, 1993). While due to the lack of viable instruments we cannot address endogeneity of employment, future studies might explore the role of employment in greater depth. Acknowledgements This paper draws on Liu’s MS thesis at The University of Tennessee. Comments by Kimberly L. Jensen and Christopher D. Clark and assistance by Steven M. Graves, California State University, Northridge, are gratefully acknowledged. Research was supported in part by USDA-ERS Cooperative Agreements Nos. 58-5000-7-0123 and 43-3AEM-2-80063. The views in this paper are those of the authors and do not necessarily reflect the views or policies of the US Department of Agriculture. Appendix A. Appendix See Tables A1 and A2. References Angulo, A.M., Gil, J.M., Mur, J., 2007. Spanish demand for food away from home: analysis of panel data. Journal of Agricultural Economics 58 (2), 289–307. Bai, J., Zhang, C., Qiao, F., Wahl, T., 2012. Disaggregating household expenditures on food away from home in Beijing by type of food facility and type of meal. China Agricultural Economic Review 4 (1), 18–35. Becker, G.S., 1965. A theory of the allocation of time. Economic Journal 75, 493–517. Bryant, W.K., 1988. Durables and wives’ employment yet again. Journal of Consumer Research 15 (1), 37–47. Byrne, P.J., Capps Jr., O., Saha, A., 1996. Analysis of food-away-from-home expenditure patterns for U.S. households, 1982–89. American Journal of Agricultural Economics 78 (3), 614–627. Byrne, P.J., Capps Jr., O., Saha, A., 1998. Analysis of quick-serve, mid-scale, and upscale food away from home expenditures. International Food and Agribusiness Management Review 1 (1), 51–72. Cherlin, J.A., 2010. Demographic trends in the United States: a review of research in the 2000s. Journal of Marriage and Family 72 (3), 403–419. Engle, R.F., 1984. Wald, likelihood ratio, and Lagrange multiplier tests in econometrics. In: Griliches, Z., Intriligator, M.D. (Eds.), Handbook of Econometrics. Elsevier, Amsterdam, pp. 775–826. Gäl, A., Akbay, C., Özcicek, C., Özel, R., Akbay, A.O., 2007. Expenditure pattern for food away from home consumption in Turkey. Journal of International Food and Agribusiness Marketing 19 (4), 31–43. Heckman, J.J., 1979. Sample selection bias as a specification error. Econometrica 47 (1), 153–161. Heckman, J.J., Honoré, B.E., 1990. The empirical content of the Roy model. Econometrica 58 (5), 1121–1149. Jensen, H.H., Yen, S.T., 1996. Food expenditures away from home by type of meal. Canadian Journal of Agricultural Economics 44 (1), 67–80.

Kotz, S., Balakrishnan, N., Johnson, N.L., 2000, . Continuous Multivariate Distributions: Models and Applications, second ed., vol. 1. John Wiley & Sons, New York. Kumcu, A., Kaufman, P., 2011. Food spending adjustments during recessionary times. USDA economic research service. Amber Waves 9(3). (accessed 31.03.12). Mancino, L., Todd, J., Lin, B.H., 2009. Separating what we eat from where: measuring the effect of food away from home on diet quality. Food Policy 34 (6), 557–562. McCracken, V.A., Brandt, J.A., 1987. Household consumption of food-away-fromhome: total expenditure and by type of food facility. American Journal of Agricultural Economics 69 (2), 274–284. Minhalopoulos, V.G., Demoussis, M.P., 2001. Greek household consumption of food away from home: a microeconometric approach. European Review of Agricultural Economics 28 (4), 421–432. Mutlu, S., Gracia, A., 2006. Spanish food expenditure away from home (FAFH): by type of meal. Applied Economics 38 (9), 1037–1047. Piernas, C., Popkin, B.M., 2010. Snacking increased among U.S. adults between 1977 and 2006. The Journal of Nutrition 140 (2), 325–332. Pingali, P., 2007. Westernization of Asian diets and the transformation of food systems: implications for research and policy. Food Policy 32 (3), 281–298. Pudney, S., 1989. Modelling Individual Choice. The Econometrics of Corners, Kinks, and Holes. Blackwell Publishers, Cambridge, UK. Sebastian, R.S., Enns, C.W., Goldman, J.D., 2011. Snacking Patterns of U.S. Adults: What we Eat in America, NHANES 2007–2008. Food Surveys Research Group Dietary Data Brief No. 4. June. (accessed 23.09.12). Shonkwiler, J.S., Yen, S.T., 1999. Two-step estimation of a censored system of equations. American Journal of Agricultural Economics 81 (4), 972–982. Soberon-Ferrer, H., Dardis, R., 1991. Determinants of household expenditures for services. Journal of Consumer Research 17 (4), 385–397. Stewart, H., Yen, S.T., 2004. Changing household characteristics and the away-fromhome food market: a censored equation system approach. Food Policy 29 (6), 643–658. Stewart, H., Blisard, N., Bhuyan, S., Nayga Jr., R.M., 2004. The Demand for Food Away from Home: Full-Service or Fast Food? USDA Economic Research Service Report AER-829. (accessed 01.03.12). Tan, A.K.G., 2010. Demand for food-away-from-home in Malaysia: a sample selection analysis by ethnicity and gender. Journal of Foodservice Business Research 13 (3), 252–267. US BLS (US Bureau of Labor Statistics), 2009. Consumer Expenditure Survey, 2008. Washington, DC, December. US BLS, 2010. Consumer Expenditure Survey, 2009. Washington, DC, October. USDA-ERS (US Department of Agriculture, Economic Research Service), 2011. Briefing Room: Food CPI and Expenditures: Food Expenditure Tables. (accessed 01.01.12). –>. Vuong, Q.H., 1989. Likelihood ratio tests for model selection and non-nested hypotheses. Econometrica 57 (2), 307–333. Yen, S.T., 1993. Working wives and food away from home: the box-cox double hurdle model. American Journal of Agricultural Economics 75 (4), 884–895. Yen, S.T., 2005. A multivariate sample-selection model: estimation cigarette and alcohol demands with zero observations. American Journal of Agricultural Economics 87 (2), 453–466. Yen, S.T., 2009. Lecture Notes in Applied Microeconometrics. The University of Tennessee, Knoxville. Yen, S.T., Rosinski, J., 2008. On the marginal effects of variables in the logtransformed sample selection models. Economics Letters 100 (1), 4–8. You, W., Zhang, G., Davy, B.M., Carlson, A., Lin, B.-H., 2009. Food consumed away from home can be a part of a healthy and affordable diet. The Journal of Nutrition 139 (10), 1994–1999.