The effect of the food stamp program on nutrient intake.

Butler, J.S. ; Raymond, Jennie E.

I. INTRODUCTION

One of the goals of the Food Stamp Program, as stated in the preamble of the law, is to ensure that low-income households have the opportunity to attain a nutritionally adequate diet. Previous economic studies evaluating the effects of food stamp income on nutrient intake have typically found a positive, usually small, correlation between the receipt of food stamps and nutrition. Two recent papers by Devaney and Moffitt [1991] and Horton and Campbell [1991] reach very positive conclusions concerning the effect of income in general. Papers from the nutrition literature generally find positive but weak and often insignificant effects of income on nutrition.

We employ information on two groups of low-income people and find slightly different results for the two groups, but very different results from many of the other studies. For a sample of rural families, taken from data from the Rural Income Maintenance Experiment, we find almost no effect of food stamp income or evidence of selection bias.(1) However, for a sample of elderly people, from the Food Stamp Cashout Project, we find strong evidence that self selection into the Food Stamp Program is highly correlated with attaining nutritional adequacy and that, controlling for the self selection, food stamp income and other income have a negative effect on nutrition.

We re-examine the relationship between food stamp income and the nutrient intake of the poor and improve upon the existing literature in several important ways. First, it is possible that the positive correlation between the receipt of food stamps and nutrition is not a direct causal link, but rather is the result of other factors affecting both. For example, in one of the samples examined in this paper, only 50 percent of those eligible actually apply for and receive food stamps. It could be the case that those who care more about nutrition are at the same time more likely to apply for and receive food stamps and more likely to maintain a nutritionally adequate diet. These possible self-selection problems can be disentangled from possible direct effects of food stamp income on nutrition using estimation procedures which correct for selection bias.

Another issue which our research addresses is that the level of one nutrient may be correlated with the levels of other nutrients. This may occur either because individuals who care about nutrition may be more likely to attain adequate levels of all nutrients or because combinations of nutrients occur naturally in foods. We posit a multi-equation model of the levels of nutrients and allow the disturbances in each equation to be correlated. None of the previous studies considered this correlation.

Finally, we recognize the fact that, above some level, extra nutrients are superfluous. If an individual has attained a nutritionally adequate diet, then whether or not food stamp income allows him or her to consume more nutrients is irrelevant. We are primarily interested in, therefore, whether or not an individual attains an adequate level of a nutrient, rather than the absolute amount of a nutrient consumed. In previous studies, Devaney and Moffitt [1991] had addressed simultaneously the issues of selection bias and the superfluity of nutrients above some level.

The finding that food stamp income actually lowers nutrient intake, ceteris paribus, is a novel result for studies of the Food Stamp Program, but it would not be novel in the literature on nutrition in less developed countries (Pitt and Rosenzweig [1985], Behrman and Deolalikar [1987]). It is reasonable to believe that, as income increases, households employ relatively more money than time in household production (this is Silberberg's [1985] explanation for decreased efficiency in producing pure nutrients); in other words, as income increases, people spend less time preparing meals and more money on foods which are easier to prepare. If, as is often the case, foods which are more convenient to prepare are less nutritious than foods which require a lot of time to prepare, then our result is not surprising.(2)

It is important to keep in mind that although nutrition may suffer as a result of higher income, participants in the Food Stamp Program still may maintain nutritionally adequate diets. Whether or not a diet is adequate is a separate issue from whether the absolute amount of a nutrient increases or decreases with income. To address this second issue, we use the results of our estimation to predict the probability that individuals will attain nutritionally adequate diets. Our results enable us to identify those factors which appear to be crucial to nutritional adequacy. We find that the factors which have the largest effect on the probability of achieving an adequate diet are, in general, the composition of the household (i.e., whether there are children in the family or whether the individual lives alone), education, and measures of an individual's knowledge of nutrition. (This point is also made by Clarkson [1975].) The effects of income are substantial for the sample of elderly persons and insignificant for the sample of rural families. Our conclusion is that, at least for the elderly households, other factors contribute more to the probability of attaining adequate levels of nutrition than does income. Further, the results are consistent with the theory that once adequate or near-adequate levels of nutrients are attained, individuals may prefer to substitute money-intensive foods, which may be lower in nutritional value, for time-intensive foods, as Silberberg [1985] shows.(3)

The remainder of the paper is organized as follows: section II presents a literature review, section III discusses the data, section IV discusses the econometric methodology and the results from the estimation, section V contains our estimates of the probabilities of achieving a nutritionally adequate diet, and section VI contains concluding remarks.

II. LITERATURE REVIEW

We briefly review here the wide variety of papers that study the relationship between income and food stamp support on nutrition of the poor.(4) Most earlier studies find a positive correlation between food stamp income and nutrition.(5) Clarkson [1975] shows that people with more income generally have higher nutrient intake, but that up to 1975 gains in aggregate income accompanied declining nutrition. He summarizes four studies, including Madden and Yoder [1972]: "Recent nutritional outcome of the Food Stamp Program appears to be equally poor. When the outcome is adjusted for other variables, the only factor that systematically appears to improve diets is a program of nutrition education (income is sometimes a significant variable but is highly correlated with education)." He goes on to attribute the results to the substitution of convenient, packaged or processed food as a result of food stamp support.

Only two early studies control for endogenous or self-selected participation in food stamp programs. Akin et al. [1985] report positive effects based on the estimated coefficients from their equations for participants and, usually, negative effects in their equations for eligible nonparticipants and ineligibles. Butler, Ohls, and Posner [1985] found positive results but coded participation as a dummy variable, thus ignoring the value of food stamps received.

Devaney and Moffitt [1991], studying the effects of various types of income on the consumption of ten nutrients on a national sample of low-income households, find significant positive effects on nutrition of food stamps and of other income with stronger, (three to seven times as high) effects for food stamps. They correct for selection bias in the level of nutrients and in the effect of income. The correction makes little difference. Devaney and Moffitt [1991] had no measure of general education or knowledge of nutrition. Also they do not correct for nutritionally recommended consumption vs. nutrients in excess of the recommended levels. Nevertheless, their results suggest very positive nutrient effects from food stamp programs.

Several studies that address this issue for other countries report small or negative effects of income on nutrition. Devaney and Fraker [1986] study Puerto Rico and find nonsignificant differences in nutrition between food stamps and cash. Pitt and Rosenzweig [1985], examining nutrition in farm households in Indonesia, report small income (wage and farm profit) elasticities for nutrients. In a study of families in rural south India, Behrman and Deolalikar [1987] report that income increases food expenditures but not nutrient intake.

Horton and Campbell [1991], who study the effect of various factors, especially the employment of the wife in a household, on food expenditure and apparent nutrient intake in Canada, report that income has a significant positive effect on all nutrients but with a declining marginal effect.

Finally, there are studies in the nutrition literature that relate income and nutrient intake.(6) None of these papers controls for selection bias, but they provide support for the observation that effects of income on dietary quality are weak and variable, whereas education has measurable effects on dietary quality. They also provide some support for the use of two-thirds of the Recommended Dietary Allowance as a criterion for an adequate diet.

Two studies, Kohrs, Czajka-Narins, and Nordstrom [1989, 308] focused on the elderly and concluded that income has a positive correlation with energy and iron intake. Davis [1981, 298] concluded that "studies have shown that the lower the income level, the greater the risk that nutritional intake will fall below the recommended dietary allowances."

We conclude this review with two excerpts from Guthrie [1986].

The poor have been identified in many nutritional surveys as a group with generally less than adequate diets. This is attributed in part to their limited resources for all necessities of life, including food, and in part to the fact that low-income families generally have less education and less sound nutritional knowledge on which to base their food choices. Their problem is compounded by the fact that the cost of less expensive foods eaten by the poor, who spend 37 percent of their income on food, is rising faster than that of more expensive foods usually consumed by the more affluent. Interestingly, the poor get more nutrients per dollar spent on food than do those with more money.

The nutritional problems of the elderly stem from psychological and social factors such as low income, long-standing food habits, loneliness, poor housing, lack of adequate storage and preparation facilities, lack of transportation to stores, and indifference to or ignorance of adequate food habits. Physiologically they suffer from decreased ability to absorb and transport nutrients, increased excretions of nutrients, and thus relatively increased need. [pp. 634-35]

Inadequate income is an obvious factor in undernutrition.... Although money is no guarantee of an adequate diet, when income falls below a certain point, the chances of obtaining enough nutrients decrease. [p. 645]

III. THE DATA

The two groups analyzed here are families who participated in the Rural Income Maintenance Experiment and a group of elderly people from the Supplemental Social Insurance/Elderly Food Stamp Cashout Project. The Rural Income Maintenance Experiment was one of four experiments in negative income taxes. It collected data in Manson, Iowa and Warsaw, North Carolina from the summer of 1969 to September 1973.(7) The Food Stamp Cashout Project operated from April 1980 to August 1981 in six sites, one demonstration site and one comparison site each in New York, South Carolina, and Oregon.(8) In the demonstration sites, cash was substituted for food stamps in payments to the elderly and to persons receiving Supplemental Social Insurance. Food stamps were provided as usual in the comparison sites. Our analysis is restricted to the elderly participants in the project, some of whom received food stamps or cash and others of whom did not.

Nutrient intake is measured by determining the foods consumed by an individual and then converting those foods into nutrients using computer programs which can analyze almost 5000 categories of food [U.S. Department of Agriculture 1975]. We have measures of ten nutrients for the participants in the Rural Income program and for nine nutrients for the Food Stamp Cashout participants.(9) The level of a nutrient is measured as the percentage of the USDA's Recommended Dietary Allowance for that nutrient consumed. This measure of nutrition, the Nutrient Adequacy Ratio, is calculated for individuals in the Food Stamp Cashout project. However, for the Rural Income program participants, the Nutrient Adequency Ratio is computed for the entire family: the nutrient intake of the family was added together and divided by the Recommended Dietary Allowance for all members of the family.

In both samples, the information on foods consumed was collected using a twenty-four hour recall survey.(10) For the participants in the Rural Income program, nutrient data were collected twice, in Family Management Interviews V (September 1970) and XIII (September 1972). Unfortunately, nutrition data from Iowa for the first interview are no longer available. The sample consists of 799 observations on families in North Carolina (there may be many families for which there are two observations) and 254 observations on families in Iowa.

For the participants in the Food Stamp Cashout project, a long and complex interview was undertaken to determine the eligibility and the sources of income of the persons in the sample. A subset of those who were eligible was re-interviewed to determine their food consumption and a few other items, such as psychological scales and height and weight. Altogether, 2033 individuals were interviewed to the end of the food intake survey. However, we restrict our analysis to those 1542 individuals for whom we have complete data.

IV. THE EMPIRICAL MODEL AND RESULTS

Nutrients differ from other commodities in that they are usually not consumed directly, but rather indirectly through food. Foods have attributes other than nutrients, such as taste, appearance, and ease of preparation, which individuals may also value. Nutrient intake will be highest among people who value the nutritional characteristics of food and who may be willing to sacrifice other attributes to obtain nutrients. We hypothesize that nutrient intake depends on individual characteristics which affect the willingness and ability to consume nutrients.

Clarkson [1975, 75] wrote: "There are differences in household diets as the result of nutritional information, homemaker education, geographic location, climate, population density, ethnic background, age distribution, activity levels, income, and other variables." We control for all of these factors in our model.

Physical factors, such as height and weight, may affect the level of nutrients required. People who are taller or heavier may require higher levels of nutrients to maintain health or to perform physical tasks. We include the height of an individual in our model, but, since weight is endogenous, we omit it. We include other individual characteristics - age, sex, and race - which may have an effect on nutrient intake, although the effect may be due primarily to social rather than physical, reasons.

The composition of the household may also affect the willingness to consume nutrients. In particular, those who live alone may be less eager to prepare healthy meals due to the possible technical inefficiency of cooking for one. Elderly men who live alone may be uninterested in domestic tasks and even less likely to maintain adequate nutrition. If the household includes children, there may be an added incentive to prepare nutritious meals, or, conversely, the presence of children may reduce the ability of a family to provide adequate nutrition for everyone. We include variables to control for the composition of the household, and, in the sample of elderly people, a variable for those households where an elderly man lives alone.

Diets vary in different regions of the country, and those who live in rural areas may have access to different foods from those who live in urban areas. We control for the region of the country where the households are located and for those who live in rural areas.

Those individuals who possess more human capital may be better able to plan and prepare nutritious meals. We include two measures of human capital, years of formal education and a measure of an individual's knowledge of elementary nutrition. Respondents in the Rural Income experiment were asked the following question:

Next I would like to ask your ideas about what food and drinks a person should have to keep in good health. What things to you think a healthy person should eat or drink daily to be healthy? [3rd quarterly interview, 1970, Part B, page 29.]

A similar question is asked in the Food Stamp Cashout Survey. In each case the respondent could name any number of foods and the result was coded by counting the number of the four basic food groups represented.(11)

Behavior may be different on the weekends from that on weekdays, and nutrient intake may be different as well. We include a dummy variable to control for whether the nutrients reported were consumed on a weekend, although we cannot predict whether there will be a positive or negative effect on nutrition.

The ability to acquire nutrients depends on the resources available to purchase food. Both income and the amount of the food stamp bonus have long been known to have a positive effect on the amount spent on food (Holbrook and Stafford [1971], Benus, Kmenta, and Shapiro [1976], Blanchard et al. [1982]). But whether the increased expenditure on food translates into an increase in the nutrient value of the food purchased depends on the value to the individual and the cost of the other attributes of the food relative to the value and cost of nutrients. We include both the amount of the food stamp bonus and other income in our equation, but do not predict the signs of the coefficients. If, as income increases, foods which taste better or which are easier to prepare are more likely to be purchased, then the effect of income on nutrient intake may be negative.

We have already discussed the justification for the inclusion of the amount of the food stamp bonus in the equations for nutrients. However, the amount of the food stamp bonus is not exogenous; it is observed only if an individual participates in the Food Stamp Program. Thus, if those individuals who are more likely to participate in the Food Stamp Program are also more likely to value nutrition, then including food stamp bonus amount in the nutrient equations introduces selection bias

We can model this more formally as a variant on an endogenous switching model:(12)

(1a) [y.sub.ij] = [[Beta].sub.11][x.sub.1] + [[Beta].sub.12][x.sub.2] + [u.sub.1], - [Gamma][Z.sub.i] [greater than or equal to] [u.sub.3].

(1b) [y.sub.ij] = [[Beta].sub.21][x.sub.1] + [[Beta].sub.22][x.sub.2] + [u.sub.2], - [Gamma][Z.sub.i] [less than] [u.sub.3].

(2) [Mathematical Expression Omitted],

where [y.sub.ij] is the amount of nutrient j consumed by individual i, [x.sub.1] is the vector of explanatory variables common to both equations (1a) and (1b), [x.sub.2] is the amount of food stamp income received, [[Beta].sub.11], [[Beta].sub.12], [[Beta].sub.21], and [[Beta].sub.22] are the coefficients on the explanatory variables, [P.sup.*] is the propensity to participate in the Food Stamp Program and receive food stamp income, [z.sub.i] is a vector of coefficients which affect the probability of participating, [Gamma] is a vector of coefficients and [u.sub.1], [u.sub.2], and [u.sub.3] are disturbance terms. The switching is endogenous if [u.sub.3] is correlated with [u.sub.1] or [u.sub.2]. Since we believe that it is plausible that such a correlation exists, we need to control for the endogeneity of the actual food stamp bonus amount received.

Heckman [1979] suggests what is the standard correction for selection bias in this framework. Equation (2) is estimated using probit, and the inverse Mills ratios ([Lambda]'s) are formed from the results of the probit estimation. The [Lambda]'s are included in equations (1a) and (1b) which are then estimated by OLS.(13) The estimated coefficients are the covariances between [u.sub.1] and [u.sub.3] and between [u.sub.2] and [u.sub.3] respectively.

There are additional considerations which simplify the model somewhat and allow us to collapse equations (1a) and (1b) into one equation. First, since [x.sub.1] appears in both equations, if we assume that the coefficients on [x.sub.1] are the same in both equations ([[Beta].sub.11] = [[Beta].sub.21]), we can reduce the number of parameters to be estimated. Further, for those who do not participate in the Food Stamp Program, the amount of food stamp bonus, [x.sub.2], has no effect on nutrition ([[Beta].sub.22] = 0) and since the value of [x.sub.2] is zero, we can eliminate that regressor from the equations. Finally, if we assume that [[Sigma].sub.13] = [[Sigma].sub.23], we can include the individual's value of [Lambda] as one regressor. Therefore, for one nutrient, we have collapsed the endogenous switching model into the following equations:

(1[prime]) [y.sub.ij] = [[Beta].sub.1][x.sub.1] + [[Beta].sub.2][x.sub.2] + [[Sigma].sub.1] *[[Lambda].sub.1] + [v.sub.1],

(2) [Mathematical Expression Omitted].

We hypothesize that, in general, those factors which affect nutrient intake also affect the probability that an individual will participate in the Food Stamp Program. However, the value of assets held by a family may also affect the participation decision but not have an effect on nutrition. Assets can affect the flow of income which the household receives; however, we control for that. Assets which do not produce a cash flow probably do not improve the diet. Finally, assets can affect the perception by the participants themselves and others viewing them of how appropriate their behavior is, i.e. stigma is created by assets.

We estimate the model two ways: using the Heckman structure and by entering the food stamp bonus amount into the equation and using the potential food stamp bonus amount as an instrumental variable for it. Signs and magnitudes of the estimated coefficients were very similar, but the standard errors were smaller with the Heckman procedure, as they should be, since it uses more information in constructing the estimates.

A final econometric consideration is that we believe that the level of nutrient intake is correlated across nutrients: an individual who consumes large amounts of one nutrient may also tend to consume large amounts of other nutrients. We have an equation (1[prime]) for each of several nutrients and we estimate those as a seemingly unrelated regression model. The equations have identical regressors, but the identity of OLS and Generalized Least Squares requires linear equations without endogenous variables, so the system must be estimated explicitly.

TABLE I

Summary Statistics: RIME Participants


Standard
 Mean
Deviation

Sex of Head of Household

(1 = male) .7893
.4077
Race (1 = Black) .4686
.4990
Age 47.7466
13.4726
Live Alone 0.0692
0.2538
Children 1.8140
1.8402
Adults 2.4876
1.4072
Elderly 0.2552
0.5530
Education 8.0692
3.2504
Knowledge 2.8519
1.0540
Non-Food Stamp Income 153.8246
160.0486
Actual Food Stamp Bonus 9.9506
46.5388
Assets 707.3066
1351.4292
North Carolina 0.7580
0.4282
Year = 1970 0.3538
0.4781
Home Produced Food 0.4582
0.4982
Weekend 0.1299
0.3362
Number of Children Sick 0.0635
0.2553
Doctor 3.3795
0.9828
Hospital 2.6660
9.0851
Outpatient 0.2618
1.3794
Pharmacy 0.2552
1.2543

Measures of Nutrition

Calories 1.0548
0.5204
Protein 1.5682
0.7831
Calcium 0.7273
0.5259
Iron 1.1590
0.8055
Vitamin A 1.2862
1.4396
Thiamine 1.6719
1.0982
Riboflavin 1.4181
0.9415
Niacin 1.5373
1.0205
Vitamin C 1.6355
1.5907
Phosphorous 1.3550
0.7472

Participation 0.0806
0.2722

Sample size = 1054

Table I presents the summary statistics for the Rural Income Maintenance Experiment sample. The average household contains about four people, and the average head of the household is forty-eight years old and has eight years of education. There are additional variables in the Rural Income sample which may be pertinent to either the decision to participate in the Food Stamp Program or to the attainment of adequate nutrition. If a household grows or otherwise produces its own food, then nutrition may be higher in that household than in others. Almost half the families in the sample consume food produced at home. It may also be the case that if the health of the family is poor or if the availability of health care is limited, individuals may have a greater incentive to supplement their financial resources by applying for food stamps. We have one measure of the health of the family: the number of times a child in the family was sick during the year, and several measures of the availability of health care: the distance in miles to the nearest doctor, the nearest hospital, the nearest outpatient clinic and the nearest pharmacy. We do not expect that these measures will affect the level of nutrition of the family, only the decision to participate in the Food Stamp Program. At the bottom of Table I are the average Nutrient Adequacy Ratios for the families for the ten nutrients. On average the rural families attain more than adequate levels of all nutrients, except for calcium. Finally, only a small proportion of the sample, 8 percent, actually participates in the Food Stamp Program.

Table II shows the summary statistics for those individuals from the Food Stamp Cashout Project sample. The average age of the participants is about seventy-three years; most participants live alone, and the average educational level is seven years. Only one-third of the sample is either black or male. The average income is $342.00 per month, and the average amount of food stamps received is $17.00. The average nutrient intake amounts for the elderly are also shown in Table II. On average, the levels of nutrients are not as high as those for the rural areas. The largest shortages are for calories and calcium, while the average level of the other nutrients is adequate or very nearly so.(14)

Tables III and IV present the results of the endogenous switching model for the Rural Income sample and the Food Stamp Cashout Project sample for representative nutrients.(15) The variable Participate is the selection bias correction for participating in the Food Stamp Program (the [Lambda]'s). For the rural families, there is no evidence of selection bias in the coefficients in the nutrient equations. Further, neither food stamp income nor other income has any effect on the level of nutrients consumed. The most important determinants of nutritional adequacy seem to be the number of adults and elderly in the household, which has consistently negative effects on nutrition, and the knowledge of elementary nutrition, which has a consistently positive effect.

The decision to participate in the Food Stamp Program is significantly increased by the number of children and decreased by the number of adults in the household. Older households and those with more health problems are more likely to participate. Those with more assets are less likely to participate.

For the sample of elderly people, the results are qualitatively different. The decision to participate in the Food Stamp Program has a consistently positive and often significant effect on the levels of all nutrients. Further, once the effect of participation is controlled for, the effect of an increase in income (either food stamp or other) is to reduce the level of nutrients consumed. As in the results from the rural families, the more knowledge of nutrition an individual has, the higher will be the level of every nutrient. Finally, we also include variables which measure how often an elderly person gets out of the house. Out Daily and Out Weekly are dummy variables which measure whether a person can get out every day or at least once a week. An individual who is more active or able to leave the house may find it easier to acquire and prepare food and may, therefore, be expected to have higher levels of nutrients.

TABLE II

Summary Statistics: Food Stamp Cashout Participants


Standard
 Mean
Deviation

Sex (1 = male) .3304 .4703
Race (1 = Black) .3426 .4745
Age Minus 65 8.1857 5.7815
Live Alone 0.8078 0.3940
Male-Alone 0.1760 0.3808
Education 6.9688 3.8279
Knowledge 2.1192 1.0672
Out of the House Daily 0.6368 0.4808
Out of the House Weekly 0.2599 0.4386
Height 64.5713 3.8095
Non-Food-Stamp Income 342.7583 110.1705
Actual Food Stamp Bonus 17.0218 23.0412
Potential Food Stamp Bonus 32.9127 24.4544
Distance to Food Stamp Office 6.2151 7.5853
Assets 319.8724 510.5313
New York Demonstration 0.0684 0.2524
New York Comparison 0.1102 0.3132
S. Carolina Demonstration 0.2380 0.4258
S. Carolina Comparison 0.3078 0.4615
Oregon Demonstration 0.1212 0.3264
Rural 0.2645 0.4411
Weekend 0.2444 0.4297
Calories 0.6479 0.3270
Protein 1.0461 0.6792
Calcium 0.5666 0.4140
Iron 0.8207 0.4597
Vitamin A 0.9799 1.1750
Thiamine 0.9031 0.4882
Riboflavin 0.9203 0.5249
Niacin 0.8108 0.5066
Vitamin C 1.1582 1.2788

Participation 0.5022 0.4999

Sample size = 1542

Sampling weights: Sum: 1542.0179. Sum of squares: 3697.9705. Design
effect: 2.39811.

The probability of participation is lower among those who live alone and among those with more education, assets, and income. The Distance variable measures distance from the food stamp office and has, not surprisingly, a negative effect on the decision to participate.

TABLE III

Results for Selected Nutrients of Weighted Estimation Corrected for
Selection Bias: Rural Income Maintenance Participants

 Calories Protein
 Coeff. t-stat Coeff.
t-stat

Constant 1.055 7.051 1.760
7.735
Race 0.084 2.102 0.217
3.823
Age of Head -0.002 -0.986 -0.008
-2.796
Live Alone -0.077 -1.067 -0.280
-2.871
Children -0.011 -1.031 0.015
0.895
Adults -0.036 -2.460 -0.076
-3.795
Elderly -0.078 -1.896 -0.158
-3.061
Education -0.005 -0.767 -0.003
-0.324
Knowledge 0.061 4.429 0.111
4.884
Income 0.000 1.018 0.000
0.408
Food Stamp Bonus 0.000 1.086 0.001
1.094
Home Food 0.114 3.383 0.099
2.000
Weekend -0.024 -0.592 0.055
0.858
Participate -0.057 -1.370 -0.049
-0.756

 Iron Participation

Constant 0.921 4.185 -1.504
-2.185
Race 0.049 0.917 0.130
0.694
Age of Head 0.005 1.588 0.026
2.560
Live Alone -0.010 -0.069 -0.094
-0.315
Children -0.012 -0.090 0.311
6.716
Adults -0.057 -2.795 -0.258
-3.341
Elderly -0.046 -0.909 -0.626
-2.528
Education -0.001 -0.114 0.028
0.970
Knowledge 0.081 3.722
Income -0.000 -0.983 -0.001
-1.108
Food Stamp Bonus -0.001 -0.583
N. Carolina -0.128 -1.489 -0.602
-2.712
Year = 1970 0.038 0.691 -0.702
-3.920
Home Food 0.088 1.545 -0.085
-0.575
Weekend -0.035 -0.541
Participate -0.023 -0.175
Assets 0.001
-3.558
Hospital 0.016
3.152
Conditions 0.318
5.819

V. THE PROBABILITY OF ATTAINING A NUTRITIONALLY ADEQUATE DIET

As stated above, whether a particular variable causes the level of a nutrient to increase or decrease is not the relevant measure of whether nutritional adequacy is attained. Rather, we are interested in whether the probability of attaining an adequate diet is improved by certain variables. Tables V and VI present the effects of selected independent variables on the change in the probability of achieving an adequate level of selected nutrients.(16) Table V presents the results for the sample of rural families. The effects of the continuous variables are presented at the top of the page; the column entries measure the size and direction of the change in the probability of attaining an adequate level of a nutrient when the independent variable is changed by the amount given in parentheses. This change is a rounded value approximately equal to one standard deviation. Increasing the number of people reduces the levels of all nutrients. Income has only small effects; food stamps move the probability by a few percentage points.

TABLE IV

Results for Selected Nutrients of Weighted Estimation Corrected for
Selection Bias: Food Stamp Cashout Participants

 Calories Protein
 Coeff. t-Stat Coeff.
t-Stat

Constant 0.178 1.003 0.333
1.027
Sex -0.220 4.751 -0.143
1.816
Race -0.031 1.709 -0.034
0.979
Age - 65 -0.000 0.546 0.007
2.161
Live Alone -0.133 2.687 -0.266
3.463
Male-Alone 0.050 0.972 0.092
1.027
Education 0.005 1.964 0.006
1.240
Knowledge 0.041 5.291 0.103
6.637
Out Daily 0.041 1.535 0.138
2.225
Out Weekly 0.041 1.408 0.114
1.850
Height 0.009 3.903 0.014
3.264
Income -0.000 2.395 -0.001
2.654
Food Stamp Bonus -0.001 2.292 -0.002
1.820
Weekend -0.006 0.325 -0.074
2.168
Participate 0.033 2.246 0.053
1.854

 Calcium Iron

Constant 0.152 0.632 -0.046
-0.173
Sex 0.077 1.502 0.012
0.209
Race -0.035 -1.527 -0.047
-1.860
Age - 65 -0.001 -0.753 -0.000
-0.545
Live Alone -0.072 -1.381 -0.13
-2.176
Male-Alone -0.036 -0.582 0.019
0.278
Education 0.015 4.390 0.006
1.611
Knowledge 0.028 2.826 0.050
4.494
Out Daily 0.001 0.017 0.070
1.603
Out Weekly 0.015 0.361 0.035
0.765
Height 0.006 1.989 0.014
3.896
Income -0.000 -1.215 -0.000
-2.105
Food Stamp Bonus -0.000 -1.261 -0.001
-2.128
Weekend -0.038 -1.703 -0.052
-1.906
Participate 0.054 2.885 0.036
1.634

 Riboflavin Niacin

Constant 0.435 1.489 0.155
0.600
Sex 0.057 0.881 -0.124
-1.860
Race -0.055 -1.814 0.020
0.685
Age - 65 0.000 0.014 0.003
1.652
Live Alone -0.152 -2.434 -0.196
-3.109
Male-Alone -0.079 -1.084 0.086
1.110
Education 0.015 3.776 0.009
2.356
Knowledge 0.051 4.121 0.057
4.861
Out Daily -0.002 -0.043 0.080
1.718
Out Weekly 0.006 0.132 0.074
1.489
Height 0.010 2.530 0.010
2.793
Income -0.001 -3.016 -0.000
-1.059
Food Stamp Bonus -0.002 -2.423 -0.002
-2.028
Weekend -0.042 -1.302 -0.019
-0.638
Participate 0.081 3.439 0.057
2.321

 Participation

 Coeff. t-Stat

Constant 2.056 6.233
Sex -0.246 -1.167
Race 0.113 1.373
Age - 65 -0.010 -1.804
Live Alone -0.609 -2.906
Male-Alone 0.185 0.818
Education -0.039 -3.548
Knowledge 0.024 0.785
Out Daily -0.166 -1.529
Out Weekly -0.041 -0.350
Income -0.002 -5.572
Assets -0.000 -2.964
New York Demonstration 0.034 0.220
New York Comparison -0.007 -0.055
S. Carolina Demonstration -0.309 -2.576
S. Carolina Comparison -0.095 -0.806
Oregon Demonstration 0.217 1.673
Weekend 0.077 1.011
Distance -0.008 -1.864

The effects of discrete variables are computed slightly differently. For each dummy variable, the first row gives the probability if everyone in the sample has a value of the variable equal to zero. For example, if everyone were female, the probability of attaining adequate amounts of calories equals 72.1 percent. The second row presents the probability if everyone has a value of the variable equal to one; if everyone were male, the probability would be 70.4 percent. The last row shows the effect of being male on the probability: the probability falls by 1.7 percentage points. In general the largest results are for changes in the knowledge of nutrition. For example, an increase in knowledge (measured as an increase in the number of food groups) from zero to two increases the probability of attaining an adequate level of calories from 8.8 percent to 71 percent and an additional increase to four food groups increases the probability to 91 percent.

In Table VI analogous results are presented for the elderly persons. Overall, the probability of achieving an adequate diet is relatively low. The effect of an increase in education by four years is substantially positive. The effects of food stamp and other income are substantially negative for many of the nutrients. Living alone and not leaving the house often also have large negative effects. But again, the most dramatic effect is for knowledge of nutrition, which has effects in the range of 5 to 44 percentage points. Very large changes in income would be required to duplicate these changes; about $300 per month in income would increase the probability less than 10 percentage points.

TABLE V

Change in the Probability of Attaining an Adequate Level of
Selected
Nutrients: Rural Income Maintenance Participants

 Calories Protein Iron

Overall 0.710 0.966
0.865

Continuous

Age (+10) -0.041 -0.017
0.048
Children (+1) -0.016 0.005
0.001
Adults (+1) -0.126 -0.018
-0.087
Elderly (+1) -0.278 -0.040
-0.068
Education (+4) -0.047 -0.001
-0.005
Income (+160) 0.002 0.000
-0.004
Food Stamp Bonus (+100) 0.055 0.008
-0.032

Discrete

Sex Female 0.721 0.967
0.868
 Male 0.704 0.966
0.862

 0.017 -0.001
-0.006

Race White 0.589 0.947
0.828
 Black 0.820 0.985
0.883

 0.231 0.038
0.055

Live Alone No 0.727 0.980
0.864
 Yes 0.459 0.884
0.849

 0.268 -0.016
-0.015

Knowledge 0 0.088 0.858
0.290
 2 0.710 0.966
0.864

 0.622 0.108
0.574

 2 0.710 0.966
0.864
 4 0.907 1.000
0.976

 0.193 0.034
0.112

Home Food No 0.557 0.960
0.824
 Yes 0.873 0.981
0.929

 0.316 0.021
0.105

Weekend No 0.716 0.964
0.872
 Yes 0.641 0.977
0.824

 -0.075 0.013
-0.048

Finally, Tables VII and VIII present changes in the Nutrient Adequacy Ratios for selected nutrients when there are changes in the independent variables. Overall, the Nutrient Adequacy Ratios are close to one for both groups for all nutrients. The largest increases occur when people's knowledge of nutrition increases. For the sample of elderly people, other large increases occur when individuals do not live alone or get out of the house frequently. The effects of income and food stamp income are relatively small.

VI. CONCLUSION

We have examined the effect of food stamp support and income on the nutrition of two groups of persons in the U.S., one elderly and the other rural. We find that adequate income is no guarantee of adequate nutrition; increased income, either [TABULAR DATA FOR TABLE VI OMITTED] restricted to food stamps or otherwise, is associated with reduced nutrient intake in both data sets. There has been no clear consensus in the literature as to what this effect should have been. Recent papers have found some positive effects, but most papers have found weak effects.

Our results suggest that even rudimentary knowledge of nutrition can increase nutrient intake considerably. Education in years has no effect in the elderly case, but substantial effects in the rural case, which confirms the mixed results for education found in the literature. Whenever it can be measured the effect of prior knowledge of nutrition is always very large. As Clarkson [1975, 56] wrote, "a more promising avenue [than increased income] for reducing malnutrition is to provide education on the value of improving nutrition as well as on the dietary value of alternative food sources (including specific combinations of food) and on methods of preparing foods."

TABLE VII

Change in the Nutrient Adequacy Ratio for Selected Nutrients: Rural
Income Maintenance Participants

 Calories Protein Iron

Overall 0.975 0.996
0.987

Continuous

Age (+10) -0.005 -0.003
0.006
Children (+1) -0.002 0.001
0.000
Adults (+1) -0.014 -0.003
-0.010
Elderly (+1) -0.036 -0.009
-0.008
Education (+4) -0.005 0.000
0.000
Income (+160) 0.001 0.000
-0.000
Food Stamp Bonus (+100) 0.005 0.001
-0.003

Discrete

Sex Female 0.976 0.996
0.988
 Male 0.975 0.996
0.987

 0.001 0.000
-0.001

Race White 0.960 0.992
0.982
 Black 0.884 0.999
0.989

 0.024 0.007
0.007

Live Alone No 0.978 0.998
0.987
 Yes 0.946 0.985
0.886

 0.032 -0.013
-0.001

Knowledge 0 0.879 0.982
0.902
 2 0.975 0.996
0.987

 0.096 0.014
0.085
 2 0.975 0.996
0.987
 4 0.996 1.000
0.999

 0.021 0.004
0.012

Home Food No 0.963 0.995
0.984
 Yes 0.992 0.998
0.994

 0.029 0.003
0.010

Weekend No 0.976 0.996
0.988
 Yes 0.969 0.997
0.983

 0.007 0.001
-0.005

[TABULAR DATA FOR TABLE VIII OMITTED]

1. The specific programs and the data are discussed in detail below.

2. This argument is supported by the theory and observations of Vickery [1977], Clarkson [1975] and Silberberg [1985].

3. A rigorous analysis of this issue is beyond the scope of this paper.

4. An extensive review of the literature is in Butler and Raymond [1995].

5. See O'Connor, Madden and Prindle [1976], Madden and Yoder [1972], Guthrie, Madden and Yoder [1972], Lane [1978], Johnson, Burt and Morgan [1981], Whitfield [1982], Akin et al. [1985] and Butler, Ohls, and Posner [1985].

6. See Brown and Tieman [1986], Windham, Wye and Hansen [1982], Emmons [1986], Perkin, Crandall and McCann [1988], Johnson, Smiciklas-Wright and Crouter [1992], Haines, Hungerford, Popkin, and Guilkey [1992], Murphy, Hudes and Viteri [1992], and Bianchetti et al. [1990].

7. The information on the Rural Income Maintenance Experiment is taken from Setzer et al. [1976].

8. Additional information on this project is contained in Blanchard et al. [1982]. Barber, Hilton, and Ohls [1982] describe the public use file. All the information in this section is drawn from these sources.

9. The nutrients are calories, protein, calcium, iron, Vitamin A, thiamine (Vitamin [B.sub.1]), riboflavin (Vitamin [B.sub.2]), niacin, Vitamin C, and phosphorous (for Rural Income program participants only).

10. A critical issue is how to determine what foods an individual has actually consumed. There are two methods, the twenty-four hour recall survey, where an interviewer asks a respondent to name every food she or he has eaten in the last twenty-four hours, and a food journal method, where the respondent writes down everything he or she eats for several days or weeks. There is some controversy in the literature over which method is appropriate. Pearl [1979] states that the twenty-four hour survey cannot be used, since it produces cross-section data and the only accurate measure of nutrient intake is a time series. However, we argue that we can use cross-section data in a regression, since an observation drawn at a particular time provides an accurate assessment of the average quality of the diet, so long as there is no autocorrelation in the random disturbances which affect nutrient intake. We have some evidence for our claim: we have two observations on the nutrient intake of some of the families in the Rural Income Maintenance Experiment. For those families from North Carolina, we computed the correlation coefficients of the residuals from the nutrient intake equations. Although all of the correlations were significantly different from zero (a correlation of 0.008 would be significant), none of the correlations was large; they ranged from 0.073 to 0.159.

11. These are fruits and vegetables, bread and grains, meat or high-protein substitutes, and dairy products.

12. We thank an anonymous referee for this insight.

13. The [Lambda]'s are computed differently for those who do and do not participate in the Food Stamp program. Specifically, [Lambda] = [Phi]([Gamma][z.sub.i])/[Phi][Gamma][z.sub.i]) for those who participate, and [Lambda] = -[Phi]([Gamma][z.sub.i])/[Phi](-[Gamma][z.sub.i]) for those who do not participate.

14. It should also be pointed out here, that the Recommended Dietary Allowance may overstate the amount of a nutrient actually required to maintain good health. Guthrie et al. [1972] deemed a diet adequate if only two-thirds of the Recommended Dietary Allowance is attained. In this light, the shortages of calcium and calories seem less dramatic.

15. The results for the other nutrients do not differ qualitatively from the results presented here. For a complete set of the empirical results, see Butler and Raymond [1995].

16. As stated above, these results are not qualitatively different from those for the rest of the nutrients (Butler and Raymond [1995]).

REFERENCES

Akin, John S., David K. Guilkey, Barry M. Popkin, and Karen M. Smith. "The Impact of Federal Transfer Programs on the Nutrient Intake of Elderly Individuals." Journal of Human Resources, Summer 1985, 383-404.

Barber, Grayson, Susan Hilton, and James C. Ohls. "Documentation for Public Analysis File Containing Survey Data Collected for the SSI/Elderly Food Stamp Cashout Demonstration Evaluation." Mathematica Policy Research Study. Washington D.C.: The Urban Institute Press, 1982.

Behrman, Jere R., and Anil B. Deolalikar. "Will Developing Country Nutrition Improve with Income? A Case Study for Rural South India." Journal of Political Economy, June 1987, 492-507.

Benus, J., J. Kmenta, and H. Shapiro. "The Dynamics of Household Budget Allocation to Food Expenditures." Review of Economics and Statistics, May 1976, 129-38.

Bianchetti, A., R. Rozzini, C. Carabellese, O. Zanetti, and M. Trabucchi. "Nutrient Intake, Socioeconomic Conditions, and Health Status in a Large Elderly Population." Journal of the American Geriatric Society, May 1990, 521-26.

Blanchard, Lois, J. S. Butler, Pat Doyle, Russell Jackson, James C. Ohls, and Barbara Posner. "Final Report: Food Stamp/SSI Elderly Cashout Demonstration Evaluation." Princeton, N.J.: Mathematica Policy Research, Inc., June 1982.

Brown, Judith E., and Patricia Tieman. "Effect of Income and WIC on the Dietary Intake of Pre-schoolers: Results of a Preliminary Study." Journal of the American Dietetic Association, September 1986, 1189-91.

Butler, J. S., James C. Ohls, and Barbara M. Posner. "The Effect of the Food Stamp Program on the Nutrient Intake of the Eligible Elderly." Journal of Human Resources, Summer 1985, 405-20.

Butler, J. S., and Jennie E. Raymond. "The Effect of the Food Stamp Program on Nutrient Intake." Vanderbilt University Department of Economics Working Paper No. 95-W01, 1995.

Butler, J. S., and Julie A. Schoenman. "Stigma In the Food Stamp Program: An Analysis UsIng Objective and Subjective Indicators." Working paper, Vanderbilt University, March 1986.

Clarkson, Kenneth W. Food Stamps and Nutrition. Washington, D.C.: American Enterprise Institute for Public Policy Research, 1975.

Davis, Audrey K. "Nutritional Hazards of Retirement," in Handbook of Geriatric Nutrition, edited by Jeng Hsu and Robert Davis. Park Ridge, N.J.: Noyes Publications, 1981, chap. 16.

Devaney, Barbara, and Thomas Fraker. "Cashing Out Food Stamps: Impacts on Food Expenditures and Diet Quality." Journal of Policy Analysis and Management, Summer 1986, 725-41.

Devaney, Barbara, and Robert Moffitt. "Dietary Effects of the Food Stamp Program." American Journal of Agricultural Economics, February 1991, 202-11.

Emmons, Lillian. "Food Procurement and the Nutritional Adequacy of Diets in Low-Income Families." Journal of the American Dietetic Association, December 1986, 1684-93.

Guthrie, Helen A. Introductory Nutrition. St. Louis: Times Mirror/Mosby College Publishing, 1986.

Guthrie, Helen A., J. Patrick Madden, Marion D. Yoder, and Helene Perrault Koontz. "Effects of USDA Commodity Distribution Program on Nutritive Intake." Journal of the American Dietetic Association, September 1972, 287-92.

Haines, P. S., D. W. Hungerford, B. M. Popkin, and D. K. Guilkey. "Eating Patterns and Energy and Nutrient Intakes of U.S. Women." Journal of the American Dietetic Association, June 1992, 698-704.

Heckman, James J. "The Common Structure of Statistical Models of Truncation, Sample Selection, and Limited Dependent Variables and a Simple Estimator for Such Models." Annals of Economic and Social Measurement, Fall 1979, 475-92.

Holbrook, Robert, and Frank Stafford. "The Propensity to Consume Separate Types of Income: A Generalized Permanent Income Hypothesis." Econometrica, 39(1), 1971, 1-21.

Horton, Susan, and Cathy Campbell. "Wife's Employment, Food Expenditure, and Apparent Nutrient Intake: Evidence from Canada." American Journal of Agricultural Economics, August 1991, 784-94.

Johnson, R. K., H. Smiciklas-Wright, and A. C. Crouter. "Effect of Maternal Employment on the Quality of Young Children's Diets: The CSFII Experience." Journal of the American Dietetic Association, February 1992, 213-14.

Johnson, S. R., James A. Burt, and Karen J. Morgan. "The Food Stamp Program: Participation, Food Cost, and Diet Quality for Low-Income Households." Food Technology, October 1981, 60-70.

Kohrs, Mary Bess, Dorice M. Czajka-Narins, and James W. Nordstrom. "Factors Affecting the Nutritional Status of the Elderly," in Human Nutrition: A Comprehensive Treatise, Volume 6: Nutrition, Aging, and the Elderly, edited by Hamish N. Munro and Darla E. Danford. New York: Plenum Press, 1989, 305-31.

Lane, Sylvia. "Food Distribution and Food Stamp Program Effects on Food Consumption and Nutritional 'Achievement' of Low Income Persons in Kern County, California." American Journal of Agricultural Economics, February 1978, 108-16.

Madden, J. Patrick, and Marion D. Yoder. "Program Evaluation: Food Stamps and Commodity Distribution in Rural Areas of Central Pennsylvania." Department of Agricultural Economics and Rural Sociology, The Pennsylvania State University, College of Agriculture, Agricultural Experiment Station, University Park, Pennsylvania, Bulletin 780, June 1972.

Murphy, S. P., Rose D. Hudes, and F. E. Viteri. "Demographic and Economic Factors Associated with Dietary Quality for Adults in the 1987-88 Nationwide Food Consumption Survey." Journal of the American Dietetic Association, November 1992, 1352-57.

O'Connor, J. Frank, J. Patrick Madden, and Allen M. Prindle. "Nutrition." Volume V, chapter 6, Rural Income Maintenance Experiment Final Report. Madison, Wisconsin: University of Wisconsin, Madison, 1976.

Pearl, Robert B. "Possible Alternative Methods for Data Collection on Food Consumption and Expenditures." Unpublished manuscript, University of Illinois, 1979.

Perkin, Judy, Lee A. Crandall, and Stephanie F. McCann. "Ethnicity and Dietary Intakes of Low-Income Mothers Served by a North Florida Family Practice Center." Journal of the American Dietetic Association, September 1988, 1081-86.

Pitt, Mark M., and Mark R. Rosenzweig. "Health and Nutrient Consumption Across and Within Farm Households." Review of Economics and Statistics, May 1985, 212-23.

Setzer, Florence, Lee Bawden, William Harrar, and Stuart Kerachsky. "Summary Report: The Rural Income Maintenance Experiment." Washington, D.C.: Office of Income Security Policy Research, November 1976.

Silberberg, Eugene. "Nutrition and the Demand for Tastes." Journal of Political Economy, October 1985, 881-900.

U.S. Department of Agriculture. Handbook of the Nutritional Contents of Foods. (Formerly titled Agricultural Handbook No. 8.) New York: Dover Publications, 1975.

Vickery, Clair. "The Time-Poor: A New Look at Poverty." Journal of Human Resources, Winter 1977, 27-48.

Whitfield, R. A. "A Nutritional Analysis of the Food Stamp Program." American Journal of Public Health, August 1982, 793-99.

Windham, C. T., B. W. Wyse, and R. G. Hansen. "Nutrient Density of Diets in the USDA National Food Consumption Survey 1977-78: 1. Impact of Socioeconomic Status on Dietary Density." Journal of the American Dietetic Association, January 1982, 28.

J. S. Butler, Associate Professor, Vanderbilt University and Jennie E. Raymond, Associate Professor, Auburn University. This research was supported by a Small Grant from the Institute for Research on Poverty of the University of Wisconsin. The authors accept responsibility for any errors.