Do the school nutrition programs supplement household food expenditures?
The objectives of the National School Lunch Program (NSLP) and the School Breakfast Program (SBP) are to provide nutritious meals to school children and to create expanded outlets for domestic agricultural products. A critical factor in determining whether the objectives of the programs are satisfied is the extent to which the school nutrition programs supplement the normal food consumption of households, as measured by the household' expenditures on food. Supplementation of households food expenditures occurs when the value of the benefits received through the NSLP and SBP does not lead to a completely offsetting reduction in the household's expenditures on food from other sources.
Two earlier studies have examined the supplementation of household food expenditures by the NSLP and SBP. However, the findings from those studies cannot be considered conclusive because of the use of a specialized sample - households of 8-12 year-old school children in Washington State (West and Price 1976) - and an unusual model specification - the exclusion of expenditures for food at school from the total food expenditures dependent variable in the model (Wellisch et al. 1983a,b).(1)
This paper uses data from the 1980-81 National Evaluation of School Nutrition Programs (NESNP) to examine the impact of the NSLP and SBP on household food expenditures, controlling for possible selection bias that may arise because of unmeasured differences between program participants and nonparticipants in their preferences for food. The paper is organized as follows. Section II provides a brief description of the school nutrition programs and describes the NESNP data. Section III presents the statistical model for the analysis of the impact of the NSLP and SBP on food expenditures. The estimation results are reported in Section IV, and Section V presents the summary and conclusions.
II. The School Nutrition Programs
Under the NSLP and SBP, there are two levels of program participation.(2) First, educational institutions - public and private nonprofit schools(3) and public or licensed nonprofit residential childcare institutions - are eligible to receive cash and commodity assistance under the NSLP and the SBP for providing meals to school children. Each eligible institution may choose to provide the NSlP, the SBP, both programs, or neither. In FY 1981, the NSLP was provided in 81 percent of all schools, serving approximately 92 percent of all school children. In contract, the SBP is provided in many fewer schools than the NSLP. Only about 33 percent of all schools, serving 39 percent of all school children, provided the SBP in FY 1981. With few exceptions, the schools that provide SBP meals also provide the NSLP.
At the second level of program participation, students enrolled in schools that provide the NSLP and/or SBP are eligible to receive or purchase a program meal. for students with family incomes at or below 130 percent of the poverty threshold, the school meals are available without charge. Reduced-price meals are available to students with family income that falls between 130 and 185 percent of the poverty threshold, while students from families with incomes above 185 percent of poverty pay full price for the school meals.(4) Of the meals served under the NSLP in FY 1981, 41 percent were provided free and 7 percent were provided at reduced price. The remaining 52 percent of the meals were served at full price.
Free meals comprised a much larger component of the SBP meals (81 percent), while the proportion of reduced-price meals was about the same as under the NSLP. Only 13 percent of the SBP participants paid full price for their meals in FY 1981.(5) The greater proportion of free meals served under the SBP reflects the high concentration of schools participating in the SBP in poorer school districts. Under the federal reimbursement structure of the SBP, an additional per meal payment is made to participating schools that are designated as in "severe need" - schools in which 40 percent or more of the lunches served are free or at a reduced price. Approximately 44 percent of the schools providing meals under the SBP in FY 1981 qualified for the severe need reimbursement rates.
Overall, student participation in the NSLP during a typical week was about 80 percent in Fall 1980. During the same period, participation in the smaller SBP was about 28 percent of all eligible students (Wellisch et al. 1983a).
III. Statistical Model
The model that is estimated consists of a food expenditure equation based on a linear Engel relationship: (1) F = [X.sub.i alpha] + [beta.sub.1 LS.sub.i + beta.sub.2 BS.sub.i + epsilon.sub.i] i = 1, 2, . . . , N
where [F.sub.i] food expenditures of the ith household: [X.sub.i] is a vector of the ith household's characteristics (including income); [alpha] is a vector of parameters to be estimated; [LS.sub.i] and [BS.sub.i] are the value of the subsidies received by the ith households from the NSLP and SBP, respectively; [BETA.sub.1] and [BETA.sub.2] represent the marginal effects of the benefits from the NSLP and the SBP, respectively, on household food expenditures; and [xi.sub.i] is a random disturbance term. Since the households that choose to participate in the NSLP or the SBP are a self-selected group of households that may have greater food expenditures than otherwise similar eligible households even in the absence of the programs, the model also includes two program participation equations: (2) [L.sub.j] = 1 if [Z.sub.1j delta.sub.1] + [mu.sub.1j] [is greater than or equal to] 0] j = 1, 2, . . . , N
= 0 if [Z.sub.1j delta.sub.1] + [mu.sub.1j < 0] (3) [B.sub.k] = 1 if [Z.sub.2k delta.sub.2] + [mu.sub.2k] [is greater than or eq ual to] 0] k = 1,
2, . . . , NB
= 0 if [Z.sub.2k delta.sub.2] + [mu.sub.2k < 0] where [L.sub.j] and [B.sub.k] are binary variables indicating the household's NSLP and SBP participation status, respectively (1 = participant, 0 = nonparticipant); [Z.sub.1j] and [Z.sub.2k] are vectors of households characteristics that affect the NSLP and SBP participation decisions, respectively: [delta.sub.1] and [delta.sub.2] are vectors of parameters to be estimated; and [mu.sub.1j] and [mu.sub.2k] are random disturbance terms. The disturbance terms [xi.sub.1], and [mu.sub.2k] have a multivariate normal distribution with mean zero and covariance matrix:
[Mathematical Expression Omitted]
In order to obtain consistent estimates of [BETA.sub.1] and [BETA.sub.2] and their standard errors, the correlation between the error terms of Equation (1) and Equations (2) and (3) need to be accounted for in the estimation procedure. The major complication of the model arises because some of the households have access to the NSLP but not the SBP (i.e., NB < N, where NB households have access to the SBP and N households have access to the NSLP).
The estimation procedure that is used is a modification of the two-stage sample selection procedure proposed by Heckman (1978) to include two selection equations (Maddala 1983). Terms corresponding to Heckman's "lambda" in the single equation context are derived from the estimation of the two participation equations. These terms are included in the second-stage food expenditure equation to reflect the components of the food expenditure disturbance term that are correlated with participation in each of the programs (and, consequently, are the source of the selection bias).
The first stage involves estimating the NSLP and SBP participation equations [Equations (2) and (3)]. Since the dependent variables in these equations - whether or not any child in the household participates at least once in the particular school nutrition program during a typical week - can take on only two values, the participation equations are estimated using probit. Because of the correlation between the disturbance terms of the two participation equations, those equations are estimated jointly using bivariate probit for those households with access to both programs.
Since the households who have the option of participating in the SBP (that is, the households with access to the SBP) are a subset of those who have access to the NSLP, I estimated two versions of the participation equation. A bivariate probit model of the NSLP and SBP participation decisions is estimated for those households that have access to both programs and thus choose whether to participate in each of the programs. The NSLP- and SBP-lambdas are derived from those equations.
A Univariate probit is estimated for all households that have access to the NSLP but do not have access to the SBP (and, consequently, do not choose whether to participate in the SBP). The univariate probit provides the estimates of the NSLP-lambda for those households that have access to the NSLP but not the SBP. For such households, the SBP-lambda is set equal to zero since the SBP participation decision is not relevant.
The two versions of the participation equations are needed since estimating a single bivariate probit model of the SBP and NSLP participation decisions for all households would imply that all households were choosing whether to participate in the SBP, when, in fact, only about half the households face that choice. Including households that are not eligible for the SBP in the population for which the SBP participation equation is estimated would introduce noise in the estimates that are obtained.
In the second stage of the procedure, the food expenditure equation is estimated using generalized least squares, with the NSLP and SBP lambdas control for the selection bias.
In this section, the data that are used and the specification of the multivariate model are described, followed by the presentation of the estimation results.
A. The Data
The data used for this analysis are from the household survey conducted during the 1980-81 school year under the National Evaluation of School Nutrition Programs. That survey collected data on the households of a nationally representative sample of 5,977 students who attended schools that provided the NSLP.(6) The interview included information on participation in the NSLP and SBP, family food expenditures, and background information, such as family income, family size and composition.(7)
B. The Specification of the Model
The specification of the food expenditure equation and the program participant equations are discussed in turn.
1. The Food Expenditures Equation
The dependent and explanatory variables of the food expenditure equation are presented in Table 1. Appendix Table A.1 provides mean values for these variables.
In interpreting the coefficient estimates on the NSLP and SBP subsidy measures, it is important to note that the dependent variable - the households' total expenditures on food during the past week - does not include the money value of the NSLP and/or SBP subsidies or the money value of home grown food. Consequently, the complete substitution of the households' normal food expenditures by the program benefits would be indicated by a coefficient estimate of - 1, while complete supplementation of normal food expenditures would be indicated by a coefficient estimate of zero. Coefficient estimates between - 1 and zero would imply partial supplementation of the households' normal food expenditures by the school nutrition program subsidies.
The explanatory variables included in the final specification of the food expenditure equation are relatively straightforward and can be divided into following categories:
* Measures of household resources (including the value of the benefits
received under the NSLP and SBP) adjusted for household size
and composition.(8 9)
(1.) By excluding food purchases at school, the household food expenditure measure used by Wellisch et al. includes all of the food expenditures of some households (those with children who receive free school meals or bring a meal from home) and only aprt of the food expenditures of other households (those with children who purchase meals at school). (2.) This section draws on U.S. Senate, Committee on Agriculture, Nutrition, and Forestry (1983). (3.) Private schools which charge average tuition of $1,500 or more per year are not eligible for the programs. (4.) Although the students with family incomes above 185 percent of poverty pay "full-price" for their school meals, those school meals, like the school meals provided for free or at reduced price, are subsidized through the federal commodity assistance program. (5.) The distribution of free, reduced price, and full price meals under the NSLP and SBP has changed little since FY 1981, when the NESNP-1 data were collected. In FY 1987, the distribution of meals by price status under the school nutrition programs was as follows (FNS, USDA, unpublished statistics): (6.) To collect the information on students and their families a three-stage stratified sample designed was used. First, school districts were stratified by poverty level and then sampled within strata; second, public schools were stratified by grade level and then sampled within the selected district; and finally, students were sampled within the chosen public schools. (7.) A second National Evaluation of School Nutrition Programs was conducted during the 1983-84 school year: information on household food expenditures was not collected under that evaluation. (8.) A consistent finding of previous research on household food expenditures is that household size and composition have important effects on food expenditures and must be controlled in such analyses (e.g., Pollack and Wales 1980, 1981; Barnes and Gillingham 1984). A common approach to controlling for household size and composition in food expenditure analyses entails the scaling of household food expenditures (and the income and program benefit explanatory variable) by "equivalent person" units, where the units are constructed by weighting each household member by the expenditure or nutritional requirements of an arbitary household member, generally an adult male. Household size is then defined in adult-male-equivalent units as the sum of the weights applied to each household member. (For examples of studies using this approach, see Devaney and Fraker 1989, Smallwood and Blaylock 1984, Basiotis et al. 1983, Brown and Johnson 1983, Buse and Salathe 1978, and Hymans and Shapiro 1976). In this study, I use relative cost of a nutritionally adequate diet for each household member to obtain an adult-male-equivalent-adjusted measure of household food expenditures. The food plan used as the basis for the AME-adjustment in this study is based on the moderate-cost food plan (MFP) developed by the Human Nutritional Information Service of the U.S. Department of Agriculture. This food plan, which is one of four plans (thrifty, low-cost, moderate-cost, and liberal-cost), suggests the amounts of foods that could be consumed by individuals of different genders and ages to meet dietary standards for a moderate expenditure. (9.) The benefits from the NSLP and the SBP are valued at the amoutn of the federal subsidy for the programs - the per meal dollar value of the cash and commodities contributed to the relevant program minus the amount paid by the child for the meal. Benefits from the Food Stamp Program and WIC are assigned their market value-the dollar value of the coupons or vouchers received by the household.
Definition of Dependent and Explanatory Variables for the Total Household Food E xpenditure Equation Variable Definition Dependent Variable Total food expenditures Total expenditures on food at home and away from home during the past week per AME ($/AME) Explanatory Variables Earned income Average weekly earned income for the past month per AME ($/AME) Earned income x SBP availability The interaction of earned income and a dummy variable for whether the SBP is available to the children in the household (1 = yes, 0 = no) Other cash income Average weekly cash income from other sources (including assistance programs) for the past month per AME ($/AME) Other case income x SBP availability The interaction of other cash income and SBP availability NSLP benefits Average weekly subsidy value of NSLP benefits received in a typical week per AME ($/AME) SBP benefits Average weekly subsidy value of SBP benefits received in a typical week per AME ($/AME) FSP benefits Average weekly subsidy value of FSP benefits received in the past month per AME ($/AME) WIC benefits Average weekly subsidy value of WIC benefits currently received per AME ($/AME) Ate home grown food Dummy variable indicating whether home grown foods were consumed during the past week (1 = yes, 0 = no) Expenditures unusually high Dummy variable indicating whether amount spent on food during the past week was more than normally spent (1 = yes, 0 = no) Expenditures unusually low Dummy variable indicating whether amount spent on food during the past week was less than normally spent (1 = yes, 0 = no) Household size in AME Number of household members (AMEs) Respondent is Black Dummy variable indicating whether the respondent is black (non-Hispanic) (1 = yes, 0 = no) Respondent is Hispanic Dummy variable indicating whether the respondent is Hispanic (1 = yes, 0 = no) Single household head Dummy variable indicating whether the household is headed by a single individual (1 = yes, 0 = no) Meal planner aged 35 years or more Dummy variable indicating whether the meal planner, assumed to be the female head (or male head, if no female head is present), is aged 35 years or more (1 = yes, 0 = no) Meal planner completed high school Dummy variable indicating whether the meal planner, assumed to be the female head (or male head, if no female head present), completed high school, but not college (1 = yes, 0 = no) Meal planner completed college Dummy variable indicating whether the meal planner, assumed to be the female head (or male head, if no female head present), completed college (1 = yes, 0 = no) Meal planner employed Dummy variable indicating whether the meal planner, assumed to be the female head (or male head, if no female head present) was employed during the previous week (1 = yes, 0 = no) Region: North Central, South or West Dummy variables indicating whether the household is located in the North Central, South, or West region, respectively (1 = yes, 0 = no) North Central x SBP availability The interaction of the region variables and SBP availability South x SBP availability West x SBP availability Urban area Dummy variables indicating whether the household is located in an urban area or Suburban area a suburban area, respectively (1 = yes, 0 = no) Urban area x SBP availability The interaction of the urbanicity variables Suburban area x SBP availability and the SBP availability variable School District: Low poverty or Dummy variables Moderate poverty indicating whether the level of poverty in the school district was low (0-11.9 percent of children below the poverty line) or moderate (12-24.9) percent of the children below the poverty line), respectively (1 = yes, 0 = no) Low poverty x SBP availability The interaction of the poverty status variables and SBP availability Moderate poverty x SBP availability Selection-bias correction terms The inverse of the Mill's ratio (lambda) derived from the NSLP NSLP participation and SBP participation equations, respectively SBP participation Constant Constant term
* A measure of household size of adjust for the economies of scale in
food purchase and preparation associated with larger households.
* Measures of whether the household's food expenditures for the
past week were unusually high or low. These variables are included
as proxies for any unusual events which may have occurred during
the week, such as bulk food purchases, major shopping trips (i.e.,
food purchases that are intended to provide supplies for more than
one week), purchases for a party, or guests eating from the household
* Measures of the characteristics of the household member who
makes the majority of the meal planning and food purchase decisions
within the household. This person is assumed to be the female
household head so long as one is present in the household. For
households headed by a single male, the male head is assumed to
be making the meal planning and food purchase decisions.
* Measures of the social and demographic characteristics of the
* Measures of the geographic location of the household.
* Selection-bias correction terms that are derived from the program
participation equations, described below.
To allow for differences in the food expenditure behavior of households with access to only the NSLP and those with access to both the NSLP and the SBP, several of the explanatory variables were created by interacting a household characteristic variable (e.g., earned income) with a dummy variable indicating whether the SBP was available to that household. A significant coefficient estimate for an interaction variable indicates that there is a structural difference in the food expenditure behavior of the two household groups.
2. The NSLP and SBP Participation Equations
The dependent variables and explanatory variables of the NSLP and SBP participation equations are presented in Table 2. Appendix Table A.2 provides mean values for these variables.
Table 2 Definition of Dependent and Explanatory Variables for the NSLP and SBP Participa tion Equations Variable Definition Dependent Variables NSLP participation Dummy variable indicating whether any household members participate in the NSLP during a typical week (1 = yes, 0 = no) SBP participation Dummy variable indicating whether any household members participate in the SBP during a typical week (1 = yes, 0 = no) Explanatory Variables Total income Average weekly earned income and income from other sources (including assistance programs) for the past month ($) Potential NSLP benefit NSLP only: Value of the household's potential weekly benefits from the NSLP, defined as the NSLP subsidy value per meal for the household) x (number of school-aged children in the household) x (5 meals per week) ($) Potential SBP benefit SBP only: Value of the household's potential weekly benefits from the SBP, defined as the (SBP subsidy value per meal for the household) x (number of school-aged children in the household) x (5 meals per week) ($) Household size Number of members of the household Proportion of household members Number of household members aged 6 to 13 years divided by household size Proportion of household members Number of household aged 14 to 17 years members aged 14 to 17 years divided by household size Respondent in nonwhite Dummy variable indicating whether the respondent is non-white (1 = yes, 0 = no) Single household head Dummy variable indicating whether the household is headed by a single individual (1 = yes, 0 = no) Meal planner completed college NSLP only: Dummy variable indicating whether the meal planner, assumed to be the female head (or male head, if no female head present), completed college (1 = yes, 0 = no) Meal planner employed Dummy variable indicating whether the meal planner, assumed to be the female head (or male head, if no female head head present), was employed during the previous week (1 = yes, 0 = no) Adult at home for breakfast SBP only: Dummy variable indicating whether there is an adult at home during breakfast time or school days (1 = yes, 0 = no) Parent decides on meal Dummy variable indicating whether the parent decides where the target child eats the meal on school days (1 = yes, 0 = no) Region: North Central, South, or Dummy variables indicating whether the household is located in the North Central, South, or West region, respectively (1 = yes, 0 = no) Urban or suburban area NSLP only: Dummy variable indicating whether the household is located in a nonrural area (1 = yes, 0 = no) Urban area SBP only: Dummy Suburban area variables indicating whether the household is located in urban area of a suburban area, respectively (1 = yes, 0 = no) Poverty status of school district Dummy variables Low level of poverty indicating whether the Moderate level of poverty level of poverty in the school district in which the household is located is low (0-11.9 percent of children below the poverty line) or moderate (12-24.9 percent of the children below the poverty line), respectively (1 = yes, 0 = no) Constant Constant term
The measures of program participation used in this study are based on whether any member of the household participates in the relevant school nutrition program at least once during a typical week.(11) The explanatory variables included in the program participation equations are very similar to those of the food expenditure equation. As in that equation, I include measures of household resources, household size, and household composition, as well as variables reflecting the characteristics of the individual responsible for meal planning and preparation within the household.
Because bivariate probit is a complex maximum likelihood estimation procedure, obtaining parameter estimates can be difficult. In order to successfully estimate the NSLP/SBP bivariate probit model it was necessary to be relatively parsimonious in the model specification, particularly for NSLP participation where the majority of the sample was participating in the program. Consequently, some of the detailed variables in the food expenditure equation have been combined in one or both of the participation equations to form more general measures (e.g., "Earned income" and "Other cash income" are combined in a single "Total income" variable for the participation equations). And some variables that were not found to be significant in initial univariate probit models estimated for the sample with access to the program were excluded from the relevant participation equation (e.g., "Meal planner completed college" was excluded from the SBP participation equation).(12)
The participation equations also include measures that are intended to capture the household's preferences for meals at school. Those variables are: measures of the presence of the school-age children in the household, a measure indicating the presence of an adult at home during breakfast meal time on school days, and a measure indicating whether the parent decides where the target child eats on school days. Finally, a measure of the potential benefits that the household could receive if it chose to participate in the NSLP is included in the NSLP participation equation and a measure of potential SBP benefits is included in the SBP participation equation.(13) The latter variables in particular address an important consideration in the specification of the program participation equations - the conditions needed to identify the effect of program participation independent of food expenditure behavior. In particular, there is no conceptual basis for believing that the potential program benefits affect food expenditures; however, there is a strong basis for believing that such potential benefits do affect the participation decision. It is worth noting that the estimates of program impacts obtained from the food expenditure equation are not sensitive to marginal changes in the set of variables selected to identify the model.
C. The Estimation Results
As noted above, the focus of the analysis is on obtaining estimates of the impact of the programs on food expenditures for those households that have the programs available. Thus, this discussion focuses on the estimation of the food expenditure equation. The results of the estimation of the two school nutrition program participation equations are presented in Appendix Table A.3.
1. Program Impacts
The results of the analysis of food expenditures for households are presented in Table 3. The coefficient estimate for the NSLP subsidy suggests that each additional dollar of NSLP benefits reduces normal household food expenditures by about 61 cents.(14) This implies that less than one-half of the NSLP subsidy is used by the households to supplement normal food expenditures.
In contrast, the coefficient estimate for the SBP benefit is 0.357, although the estimate is not significantly different from zero in a statistical sense.(15) Nevertheless, the point estimate of .357 implies that household food expenditures increase by more than one dollar for each additional
(10.) It is important to control for out-of-the-ordinary events since the goal of the analysis is to examine the impact of the NSLP and the SBP programs on usual food expenditures. Nineteen percent of the households reported spending "a lot more" on food during the past week than they normally spend in a week, while 35 percent reported spending "a lot less." Unfortunately, the households were not asked to explain why their food expenditures during the past week were unusually high or low. However, data from the Nationwide Food Consumption Survey on frequency of major food shopping trips suggests that infrequent shopping trips may be an important factor since many households undertake such shopping trips on a biweekly or monthly basis. The coefficient estimates for the program benefit variables and for the majority of the other variables in the model are virtually unchanged with the inclusion of the unusual expenditure variables. (11.) Because of problems with missing data on program participation for children in the household other than the sample target child, it was not possible to define the participation variables for the same time period as was used for the food expenditure measure (i.e., the previous week). It is likely that the relationship between school nutrition program participation in a typical week and household food expenditures during the prior week is less strong than the relationship between program participation and food expenditures during the same week. (12.) In subsequent runs testing the sensitivity of the program impact estimates from the food expenditure equation to marginal changes in the specification of the bivarate probit model, the inclusion/exclusion of these variables had little impact on the program-impact estimates. (13.) Because of the strong collinearity of the two potential benefit variables, I did not include the value of the potential NSLP benefits in the SBP participation equation. (14.) Because of the definition of total food expenditures used in this study, the value of the marginal propensity to consume (MPC) for food out of the school nutrition program benefits is defined as one plus the relevant coefficient estimate. (15.) Given the large standard error for the coefficient estimate for the SBP benefits, it is not surprising that the impact of the SBP benefit is not significantly different from that for the NSLP benefit. dollar of SBP benefits. Such super-supplementation of household food expenditures could reflect changes in household behavior in response to the SBP, for example, an increased likelihood that SBP-participant children eat any breakfast. Evidence on this issue is mixed. Wellisch et al. (1983a,b) found that a minor nutritional benefit of the SBP is a higher likelihood that a student would eat any breakfast, while Devaney et al. (1986) found no evidence of an impact of the SBP on the probability of eating breakfast.
It appears that the food expenditures of the households are not reduced by SBP benefits; rather, the SBP subsidy acts as a complete supplement to the households' normal food expenditures and may encourage an increase in food expenditures in excess of the value of the program benefits.
A finding of complete supplementation of food expenditures by SBP benefits is consistent with the results obtained by Wellisch et al. (1983a,b), while the finding of some substitution of food expenditures by NSLP benefits is quite different from their finding of complete supplementation. However, the estimate of 39 cents out of every dollar of NSLP subsidy supplementing household food expenditures is comparable to the estimate of .60 reported in the study of West and Price (1976).
Although estimates of the impact of WIC and FSP benefits on food expenditures can be obtained from this study, care should be taken in the interpretation of such estimates. Since the NESNP-1 sample population does not correspond to either the WIC-eligible or FSP-eligible populations, the coefficient estimates obtained for those programs should not be interpreted as measures of the impacts of the programs on their respective target populations. Rather, the coefficient estimates reflect the impacts of the FSP and WIC on the population of households with the NSLP available. Furthermore, the adjustment for possible selection bias due to program participation decisions has been limited to the programs of interest in this study - the NSLP and the SBP. No effort has been made to purge the estimates of the impact of the FSP and WIC on food expenditures of such selection bias.
2. Other Findings
In addition to estimating the effects of the benefits from the school nutrition programs on food expenditures, the impacts of earned income and other cash income (including public assistance/welfare) on the household's food expenditures are derived. As can be seen in Table 3, the proportions of each additional dollar of earned income and unearned income allocated to food expenditures are .04 and .02 respectively, for the households with access to the NSLP only. For the households with access to both the NSLP and SBP, the estimated proportions for earned and unearned income are both about .06.
Table 3 Estimation Results for Total Household Food Expenditures, Fall 1980 (weighted) Coefficient Standard Explanatory Variables Estimate Error Earned income .037** .003 Earned income x SBP availability .020** .005 Other cash income .016** .003 Other cash income x SBP availability .039** .009 NSLP benefit -.607** .171 SBP benefits .357 .497 FSP benefits -.125 .069 WIC benefits -.373 .405 Ate home grown food -3.752** .412 Expenditures unusually high 6.269 .480 Expenditures unusually low -6.898** .400 Household size in AME -1.609** -53 Respondent is Black -1.565** .591 Respondent is Hispanic -.763 .693 Single head of household 1.190* .596 Meal planner aged 35 years or more .829* .398 Meal planner completed high school -.533 .464 Meal planner completed college -2.799** .684 Meal planner employed -.309 .386 Region North Central -1.623* .630 North Central x SBP availability -4.996** 1.240 West -3.722** .761 West X SBP availability -.870 1.097 South -1.480* .679 South X SBP availability -3.413 .886 Urban area 1.757 .617 Urban area X SBP availability -2.425 .948 Suburban area .029 .647 Suburban area X SBP availability 1.524 1.059 Poverty status of school district Low level of poverty -1.780* .836 Low level X SBP availability 1.261 1.066 Moderate level of poverty -1.584 .880 Moderate level X SBP availability 4.203** 1.068 Selection-bias correction terms NSLP participation .983* .467 SBP participation 1.226** .394 Constant 35.461** 1.182 Chi-squared (df) 1,681.5 (35) Sample size 5,778 Mean of dependent variable 27,455 Source: Data are from the National Evaluation of School Nutrition Programs-I, Fa ll 1980. * Significant at the .05 level, two-tailed test. ** Significant at the .01 level, two-tailed test.
Other findings of interest include the effect of socioeconomic characteristics of the household and characteristics of the household's meal planner on food expenditures. As reported in Table 3, food expenditures per AME are significantly lower of households that consumed home grown food during the past week and for larger households, all else equal. Since food expenditures are measured on a per household member basis,the latter finding reflects the presence of economic of scale in the purchase and preparation of food products.
Households in which the survey respondent is black and households with better educated meal planners have lower food expenditures, while households headed by a single parent and households with older meal planners have higher food expenditures, all else equal. The lower spending on food per adult-made-equivalent household member by the better educated meal planners may reflect efficient food purchase (under the assumption that the AME adjustment is an appropriate scaling for household size).
Somewhat surprisingly, the employment status of the meal planner does not have a significant impact on household food expenditures, after controlling for the impact of earnings on those purchases.(18) One might expect that food expenditures would be higher for households in which the meal planner was employed because of a greater use of more expensive convenience foods and food away from home relative to households in which the meal planner does not work. That hypothesis is not supported by our findings.
Finally, the significance of the coefficients for the selection-bias correction terms indicate the presence of systematic differences in the food expenditures of program participants and nonparticipants that are controlled through the two-stage estimations procedure.
V. Summary and Conclusions
In this study, the extend to which the school nutrition programs supplement the normal food expenditures of households that have access to the NSLP and the SBP is examined. The results that are obtained from the estimation of the food expenditure equation indicate that somewhat less than one-half of each additional dollar of NSLP benefits is used by the households to supplement food expenditures, while all of each additional dollar of SBP benefits is allocated to households food expenditures. The finding that there is some substitution of food expenditures by the NSLP benefits and complete supplementation of food expenditures by the SBP benefits, suggest that school nutritional program benefits do supplement the food expenditures of the households. Furthermore, the benefits that are targeted through the SBP provide for greater supplementation of household food expenditures, perhaps because of changes in household behavior in response to the program (such as an increased probability that children eat breakfast).
Table A1 Mean Values for the Explanatory Variables that are Included in the Total Household Food Expenditures Equation, Fall 1980 (weighted) Standard Explanatory Variable Mean Error Earned income $98.01 80.60 Earned income x SBP availability $37.53 65.33 Other cash income $21.23 58.85 Other cash income x SBP availability $ 7.86 25.08 NSLP benefit $ 1.54 1.54 SBP benefit $ 0.19 .52 FSP benefit $ 1.06 3.20 WIC benefit $ 0.06 .45 Ate home grown food .33 .47 Expenditures unusually high .19 .40 Expenditures unusually low .35 .48 Household size in AME 4.06 1.37 Respondent is Black .16 .37 Respondent is Hispanic .09 .29 Single head of household .18 .39 Meal planner aged 35 years or more .66 .48 Meal planner completed high school .62 .49 Meal planner completed college .12 .33 Meal planner employed .55 .50 North Central .23 .42 North Central x SBP availability .04 .20 West .18 .39 West x SBP availability .09 .29 South .38 .49 South x SBP availability .23 .42 Urban area .39 .49 Urban area x SBP availability .19 .39 Suburban area .30 .46 Suburban area x SBP availability .14 .34 Low level of poverty .51 .50 Low level x SBP availability .14 .35 Moderate level of poverty .35 .48 Moderate level x SBP availability .21 .40 Constant 1.00 .00 Sample size 5,778 Source: Data are from the National Evaluation of School Nutrition Programs-I, Fall 1980
[TABULAR DATA OMITTED]
Table A3 Estimation Results for the Bivariate Probit Model of NSLP and SBP Participation for Households with Access to the NSLP and the SBP, Fall 1980 (weighted; standard errors in parentheses) Coefficient Estimates for NSLP SBP Explanatory Variables Participation Participation Total income -.0002 -.0004(*) (.0003) (.0001) Potential NSLP benefit -.123(**) -- (.041) Potential SBP benefit -- .134(**) (.117) Household size -.104 -.047 (.816) (.253) Proportion household members -.484 .065 aged 6 to 13 years (.681) (.223) Proportion household members .173 -.782(**) aged 14 to 17 years (.796) (.232) Respondent is nonwhite .115 .407(**) (.244) (.065) Single household head -.482 .103 (.253) (.083) Meal planner completed college -.517(*) -- (.208) Meal planner employed -.071 .134(*) (.174) (.059) Adult at home for breakfast -- -.393(**) (.107) Parents decide on meal .089 -.482(**) (.178) (.057) Region North Central .484 .320(**) (.342) (1.118) West .107 .005 (.311) (.097) South .237 -.018 (.266) (.098) Urban or suburban area -.474(*) -- -(.236) Urban area -- -.670(**) (.080) Suburban area -- -.860(**) (.096) Poverty status of school district Low level of poverty .639(*) .188(*) (.270) (.095) Moderate level of poverty .110 -.138 (.211) (.082) Constant 2.173(**) .958(**) (.462) (.208) Rho .523(**) (.209) Chi-squared (df) 878.40 (37) Sample size 2,710 Source: Data are from the National Evaluation of School Nutrition Programs-I, Fa ll 1980. Note: These equations were estimated for the sample of households with both the NSLP and SBP available using bivariate probit. (*) Significant at the .05 level, two-tailed test. (**) Significant at the .01 level, two-tailed test. Table A4 Estimation Results for the Univariate Probit Model of NSLP Participation for Households with Access to the NSLP Only, Fall 1980 (weighted; standard errors in parentheses) Explanatory Variable Coefficient Estimate Total income .0002 (.0001) Potential NSLP benefits .117(**) (.029) Household size -.091(*) (.039) Proportion of household members -.387 aged 6 to 13 years (.398) Proportion of household members .665 aged 14 to 17 years (.440) Respondent is nonwhite .399 (.204) Single household head .154 (.171) Meal planner completed college -.120 (.114) Meal planner employed (.041) (.092) Parent decides on meal -.014 (.094) Region North Central .238 (.123) West .042 (.137) South -.164 (.130) Urban or suburban area -.955(**) (.141) School district poverty status Low poverty -.311 (.248) Moderate poverty -.036 (.279) Constant 2.615(**) (.353) Chi-squared (df) 107.60 (16) Sample size 3,068 Mean of dependent variable .956 Source: Data are from the National Evaluation of School Nutrition Programs-I, Fall 1980. (*) Significant at the .05 level, two-tailed test. (**) Significant at the 0.01 level, two-tailed test.
Barnes, R., and R. Gillingham. 1984. "Demographic Effects in Demand
Analysis: Estimation of the Quadratic Expenditure System Using
Microdata." Review of Economics and Statistics 66(4):591-601. Basiotis, P., M. Brown, S. R. Johnson, and K. J. Morgan. 1983. "Nutrient
Availability, Food Costs, and Food Stamps." American Journal of
Agricultural Economics 65(1):685-93. Brown, M., and S. R. Johnson. 1983. "Food Stamps: Program Parameters and
Standards of Living for Low-Income Households." Southern Journal of
Agricultural Economics 15(4):43-49. Buse, R. C., and L. E. Salathe. 1978. "Adult Equivalent Scales: An
Alternative Approach." American Journal of Agricultural Economics
60(1):460-68. Devaney, B., and T. Fraker. 1986. The Dietary Impacts of the School
Breakfast Program. Princeton, N.J.: Mathematica Policy Research, Inc. ______. 1989. "The Effect of Food Stamps on Food Expenditures: An
Assessment of Findings from the Nationwide Food Consumption Survey."
American Journal of Agricultural Economics 71(2):99-104. Heckman, J. J. 1978. "Dummy Endogenous Variables in a Simultaneous
Equation System." Econometrica 46(2):931-59. Hymans, S. H., and H. T. Shapiro. 1976. "The Allocation of Household
Income to Food Consumption." Journal of Econometrics 4(2):167-88. Maddala, G. S. 1983. Limited Dependent and Qualitative Variables in
Econometrics. New York, N.Y.: Cambridge University Press. Pollack, R. A., and T. J. Wales. 1981. "Demographic Variables in Demand
Analysis." Econometrica 49(6):1533-51. _______. 1980. "Comparison of the Quadratic Expenditure System and Translog
Demand Systems with Alternative Specifications of Demographic Effects."
Econometrica 48(3):592-612. Smallwood, D. M., and J. R. Blaylock. 1984. "Scaling Household Nutrient
Availability Data." Agricultural Economics Research 36(2):12-22. U.S. Senate, Committee on Agriculture, Nutrition, and Forestry. 1983. "Child
Nutrition Programs: Description, History, Issues, and Options." Committee
Print. Washington, D.C.: U.S. GPO. Wellisch, J. B., S. D. Hanes, L. A. Jordan, K. M. Maurer, and J. A.
Vermeersch. 1983a. "The National Evaluation of School Nutrition Programs:
Final Report, Volume I - Overview and Presentation of Findings." Santa
Monica, Ca.: Systems Development Corporation. _______. 1983b. "The National Evaluation of School Nutrition Programs: Final
Report, Volume 2 - Technical Appendices." Santa Monica, Ca.: Systems
Development Corporation. West, D. A. and D. W. Price. 1976. "The Effects of Income, Assets, Food
Programs, and Household Size on Food Consumption." American Journal of
Agricultural Economics 58(1):725-30.
|Printer friendly Cite/link Email Feedback|
|Author:||Long, Sharon K.|
|Publication:||Journal of Human Resources|
|Date:||Sep 22, 1991|
|Previous Article:||Longitudinal patterns of compliance with Occupational Safety and Health Administration health and safety regulations in the manufacturing sector.|
|Next Article:||Explaining pension dynamics.|