QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Rose, D. Articles by Charlton, K. E. Search for Related Content PubMed PubMed Citation Articles by Rose, D. Articles by Charlton, K. E.

---

Community and International Nutrition

Quantitative Indicators from a Food Expenditure Survey Can Be Used to Target the Food Insecure in South Africa^{1 ,2}

Department of Community Health Sciences, Tulane University School of Public Health and Tropical Medicine, New Orleans, LA 70112 and ^* Nutrition and Dietetics Unit, Department of Medicine, University of Cape Town, Observatory 7925, South Africa

³To whom correspondence should be addressed. E-mail: diego{at}tulane.edu.

	ABSTRACT

TOP ABSTRACT INTRODUCTION SUBJECTS AND METHODS RESULTS DISCUSSION LITERATURE CITED

 
Recent work on assessing household food insecurity has focused mainly on experiential-based measures using qualitative survey questions. In this paper, we employed two quantitative measures to estimate prevalence rates for household food insecurity in South Africa. One measure, termed food poverty, assessed whether the amount spent by a household on food was inadequate to purchase a low cost food plan. Low energy availability assessed whether the food energy available to a household, through its purchases and home production, was less than the sum of its members’ recommended energy intakes. The 1995 Income and Expenditure Survey, a large representative survey of South African households (n = 28,704), was used for this secondary data analysis. Results showed that 43% of households were in food poverty in October 1995, and 55% had a low energy availability. These indicators allowed classification of households into four groups: food poverty only; low energy availability only; food insecure on both measures; and food secure. These groups differed on various aspects of household food consumption, suggesting that these indicators can be used to target different types of interventions to meet specific needs. Both bivariate and multivariate analyses showed that households that were food insecure on both measures were more likely to be in rural areas, have low incomes or large household sizes, and be headed by Africans or individuals of mixed ancestry. These patterns of food insecurity were corroborated by previous research on income poverty and nutritional status, suggesting that food poverty and low energy availability are useful, quantitative indicators for assessing food insecurity in South Africa.

KEY WORDS: • household food security • Income and Expenditure Survey • food spending • food availability • South Africa

	INTRODUCTION

TOP ABSTRACT INTRODUCTION SUBJECTS AND METHODS RESULTS DISCUSSION LITERATURE CITED

 
Considerable progress has been made in the past decade in defining and measuring household food insecurity. In the United States, the fielding of a nationally representative food security module as a component of the Current Population Survey in 1995 was the first round of an annual monitoring of the condition (1 ). Results indicated a household food insecurity prevalence rate of 12%, with 4% experiencing a more serious form of food insecurity with hunger.

There has also been substantial concern about household food security in middle and low income countries. The importance of household food insecurity in the etiology of malnutrition in the developing country context has been illustrated in the well-known UNICEF causal framework (2 ). Although food insecurity and hunger are often referred to as hidden problems in the United States, there is nothing obscure about the situation in developing countries. For example, in 1995, >30% of children in developing countries were malnourished using a weight-for-age indicator (3 ).

The U.S. food security measure is based on a number of qualitative interview questions that elicit respondents’ personal experiences regarding food security and hunger. Experiential-based measures have also been used in low income countries (4 ,5 ). Some authors have argued that combining quantitative and qualitative approaches is a useful strategy for policy research (6 ) or for identification of the food insecure (7 ). However, less attention has been paid in the recent literature on household food security to the development of quantitative indicators. Data sets from previously fielded living standards surveys exist in many developing countries. If quantitative indicators of food insecurity could be developed using such data, this would be advantageous to government agencies and nongovernmental organizations, which often have scarce resources to meet their information needs.

One quantitative indicator of household food insecurity, termed food poverty, measures whether the amount of money spent by a household on food is enough to purchase a nutritionally adequate basic diet. There are various methods for applying this indicator to population groups (8 ,9 ), including one developed previously by the authors (10 ). In this paper, we report on a second quantitative indicator of food insecurity, termed low energy availability, which assesses whether the energy available to a household from its food purchases and home production of food is less than the sum of its members’ recommended energy intakes. We show how these two indicators can be used in combination to target different types of interventions to groups of households with different types of food security problems. The 1995 Income and Expenditure Survey (IES)⁴ from South Africa, which is at the core of our secondary data analysis, also allows us to describe the socioeconomic characteristics and food consumption behaviors of these groups.

	SUBJECTS AND METHODS

TOP ABSTRACT INTRODUCTION SUBJECTS AND METHODS RESULTS DISCUSSION LITERATURE CITED

 
Study sample.

The food security measures developed here were based on the 1995 IES, which was conducted by Statistics South Africa in October of 1995 to determine expenditure patterns of South African households and to form the "basket" of consumer goods and services used in the calculation of the Consumer Price Index (11 ). Statistics South Africa removed all household-specific identifiers before making the IES data publicly available. This practice assured the confidentiality of survey respondents. The study of existing records from publicly available data sources in which subjects cannot be identified is considered exempt from human subjects review (12 ).

The Income and Expenditure Survey has been conducted every 5 y in South Africa since 1985. The 1995 survey was the first in the post-Apartheid era. Unlike previous versions, it covered all areas of the country including metropolitan, urban and rural areas. The sample, stratified by race, province and urbanization, consisted of 30,000 households, of which 29,595 cooperated (11 ). The 1991 population census was used as a frame for drawing the sample, and included estimates of the size of the population in the formerly independent states (Transkei, Bophuthatswana, Venda and Ciskei). Of the cooperating households, 28,704 had usable food expenditure data and formed the basis of the work conducted here. Additional details regarding sampling procedures (13 ) and exclusions (14 ) have been described previously.

Food poverty indicator.

A household was defined to be in food poverty when the amount of money it spent on food was inadequate to purchase a basic nutritionally adequate diet. To capture this concept, we created a simple variable, the ratio of household food spending to the cost of a basic food plan. Households in food poverty, i.e., those spending less than their basic food plan, had values <1 for this variable. The numerator of this ratio is the sum of a household’s reported food expenditures plus the estimated monetary value of the food that it consumed from home food production. The denominator is the cost of a nutritionally adequate food plan for a particular household. These food plans, discussed below, were developed for 9 different individual types based on age and gender. Thus, the cost of each household’s food plan was actually a sum of the costs of the individual food plans for all the members of that household.

Empirical data on household food spending, i.e., the numerator in the above ratio, was obtained from the IES. It contained reported data on monthly spending, in Rands, on 124 foods and food groups that were obtained from face-to-face interviews with the household head. The value of gifts in the form of food was also reported by the respondent and was included with data on monthly spending. The 124 foods encompassed a full range of what could be bought in South Africa. In some cases, these were relatively disaggregated foods. For example, five different types of cereal flours were itemized, i.e., cake flour, bread flour, mealie-meal flour, sorghum-meal flour and other meal and flour. In other instances, the foods resembled food groups. For example, different types of fresh meats were aggregated into five groups: beef and veal; poultry; pork; mutton, lamb, and goat; and other meats. Complete coverage of potential food purchases was provided by inclusion of several "other food" categories in each of the main product breakdowns (cereals, meats, fish, fats and oils, dairy, vegetables, fruits, sugars, sweets, beverages, condiments, and ready-to-eat and mixed dishes). Information on the reported quantities consumed of home-grown foods and livestock was also collected in another part of the survey, which included an additional 22 items, such as grains, milk, eggs, fruits, vegetables and meats. As with purchased foods, "other" categories were included to provide full coverage of food production possibilities. Consumption of home-produced foods was converted to a monetary value using median sales prices from the IES database.

Data for the denominator in the above ratio, i.e., representing the cost of a basic food plan, came from the University of Port Elizabeth’s Household Subsistence Level (HSL) series. It is an ongoing, biannual market survey of the monthly costs of food, clothing, and other household necessities in 24 urban centers throughout South Africa’s nine provinces (15 ). The food items chosen for pricing in each of these centers were based on the 1993 food plans (or food ration scales, as they are known in South Africa) developed by the previous South African Department of National Health and Population Development. These plans provided the minimum quantities of a selection of food items that would meet the nutrient requirements for nine different age-gender groups and were developed using nutrient recommendations from the United States (16 ). We calculated average food plan costs for each of the nine age-gender groups within each of South Africa’s nine provinces. These were weighted averages for each age-gender group in each province, based on the relative population in each of the specific locations surveyed within a province. Additional details regarding the HSL (15 ) and their use in creating a food poverty variable (10 ) have been published previously.

Low energy availability indicator.

A household was defined to have a low energy availability when the energy available in its household food supplies was less than the sum of its members’ recommended energy intakes. A simple ratio was created to capture this concept. The numerator is a sum of the energy available in each household’s reported food purchases plus the energy consumed from food produced at home. The denominator is a sum of the recommended energy intakes for each individual in a household, multiplied by 30 to convert it to the same monthly time frame as the numerator. Households that scored <1 on this ratio were defined as having a low energy availability.

The IES was used to provide information on household food purchases and home production. For each household, the amount of money, in Rands, reportedly spent on each of the 124 foods was divided by the price of that food to determine the monthly quantity available to the household of each food. Data on the prices of foods were obtained primarily from the Statistics South Africa series on retail prices (17 ). This series contained prices on >300 individual food items, far greater detail than was found in the HSL series (which we used above for the food poverty indicator because it provided information on the costs of entire food plans).

One limitation of the price data from Statistics South Africa was that prices were published as national averages. To account for spatial variation in relative prices across South Africa, food-specific provincial price indices were employed that made use of contemporaneous price information from the HSL series. Using the two data sets in concert allowed us to combine the strengths of both. As mentioned previously, the HSL reported food plan costs for 24 locations throughout South Africa. In addition to the aggregate food plan costs, prices were also reported for 23 specific foods in these locations. We used the latter to develop 23 provincial food price indices (average price of each food in each province divided by national average price for that food), one for each of the foods for which price data were available from HSL. For each province and for each food, the national food prices from Statistics South Africa were then multiplied by the appropriate provincial food price index to give food prices that more closely reflected actual food prices in each of the provinces. Additional details concerning the development and use of these prices to obtain available food quantities have been described previously (14 ).

The South African Food Composition Database Version 1.2 (18 ), developed by the Medical Research Council (MRC) of South Africa, was used to determine the energy content of foods that were reported as purchased or consumed from home production in the IES. The MRC database includes nutrient information for >1400 food items, whereas the IES collected information on 124 purchased foods and 22 home-produced foods. Thus, food items from the MRC database were selected that matched the IES foods. On occasions in which the IES foods could be represented by more than one MRC food, simple averages were taken for the energy from the appropriate MRC foods.

The nutrient data in the MRC database were reported per 100 g of edible portion, whereas IES data represented foods in quantities as purchased. Thus, values for raw foods from the MRC tables were used whenever possible. When data for cooked foods were used (e.g., boiled rice), a conversion factor was applied to the IES food quantity data to adjust the raw weight to that of a cooked weight. These conversion factors were obtained from the nutrient database of the U.S. Department of Agriculture (19 ). To adjust the weight of foods as purchased to the appropriate weight of an edible portion, the refuse for each food item was also subtracted. Common refuse factors were also obtained from the USDA’s nutrient database (19 ).

Information on household composition from the IES was combined with energy recommendations to form the denominator of the low energy availability indicator. In planning and assessment at the group level, South African nutritionists customarily use dietary recommendations developed in the United States. Following in that tradition, we employed the recommended energy intakes listed in the 1989 Recommended Dietary Allowances (16 ). These recommendations are listed for different age-gender groups and assume a light-to-moderate activity level. Although there is substantial variation in activity levels across South Africa, use of these standards provides a benchmark with which to evaluate household food supplies. It should be noted that even with more detailed information about activity levels, we cannot make assessments regarding the adequacy of food supplies for particular households or individuals within households, because of, among other things, the probabilistic nature of nutrient requirements. Thus we use the term "low energy availability" to indicate simply that energy supplies are less than these recommendations, without making inferences regarding adequacy.

Statistical analysis.

The 1995 IES sample and accompanying statistical weights were based on the 1991 South African census, the most recently conducted census at the time of the 1995 survey. To preserve representativity, the weights were adjusted to account for the exclusions due to unusable food expenditure data. This adjustment preserved the weight of each household in a province-race-urbanization cell relative to other households in that cell as well as the influence in overall estimates of each province-race-urbanization group relative to other groups (14 ). All analyses described below were performed using these adjusted weights with the WesVar statistical package (20 ). This software uses replication techniques to estimate variances from survey data that come from complex sample designs. WesVar’s generalized jackknife replication method was used in all analyses performed here because it allows for sample designs in which two or more primary sampling units are selected from each stratum.

The two food security indicators were combined, which yielded four possible outcomes: the household was food secure; it was in food poverty only; it had a low energy availability only; or it had both food poverty and a low energy availability, which we refer to as food insecurity. ² statistics were used to test for associations between socioeconomic characteristics and the prevalence of these different food security conditions (Table 2) . One-way ANOVA was used to determine whether the classification system developed with these food security indicators provided groups whose food expenditure patterns were significantly different from one another (Table 3) . Multiple comparison post-testing was performed using Bonferroni’s test with an overall Type I error rate of < 0.05

Adjusted odds ratios for all of the independent variables in the multinomial logit model were estimated simultaneously for each of the three conditions: food poverty only, low energy availability only and food insecurity. Independent variables in the model were all dichotomous and described various socioeconomic characteristics of the household. One variable was used to indicate when a household lived in an urban area. Three dichotomous variables were used to describe the race of the household head. These indicated whether the head was African, of mixed ancestry or Indian; Caucasians served as the reference group. Categorization according to "race" continues to be used in South Africa to monitor the social progress of groups that were discriminated against during the Apartheid years. People of "mixed ancestry" have also been referred to as "Coloured" in South Africa. Two other variables described the household head. One indicated when the household head was female. Elderly status of the household head was represented by a variable indicating whether he or she was 60 y old. This latter variable was included in our analysis because previous work in South Africa has shown that households headed by the elderly are at increased risk of food poverty (21 ). There were two dichotomous household size variables indicating either a small household size (1–2 people) or a large household size (7 people), with households of 3–6 members as the reference group. There were also two dichotomous income variables indicating whether a household’s income was in the lowest or highest 20% of the income distribution, with households in the middle 60% of the income distribution as the reference group. A final variable indicated whether the household produced some of the food that it consumed.

	RESULTS

TOP ABSTRACT INTRODUCTION SUBJECTS AND METHODS RESULTS DISCUSSION LITERATURE CITED

 
In October of 1995, 42.6% of South African households were in food poverty, whereas 54.6% demonstrated low energy availability. A two-by-two analysis (Table 1 ) is useful for understanding the overlap of households with both conditions. There was reasonably good correspondence between the two measures, i.e., 91% of households in food poverty also had an energy availability that was low, and 71% of households that had a low energy availability were in food poverty. However, the two measures encompassed different sorts of information; thus, they did not yield identical results in classification.

The prevalences of these four food security conditions varied by socioeconomic characteristic (Table 2 ). Significant associations (P < 0.05) between demographic variables and food security prevalences were found for all variables reported in this table. Over 50% of rural households were food insecure on both measures, whereas only 30% were food secure. The pattern was almost reversed for urban households, i.e., 26.5 and 50.5% were classified as food insecure and food secure, respectively. Food insecurity was highest among households headed by Africans, followed by those of mixed ancestry, Indians and Caucasians. Food insecurity rates were higher among households headed by females and the elderly, and among households that consumed food from their own production. Food insecurity rates increased with increasing household size and decreased with increasing income. The patterns of food security were predictably the opposite of what was seen for food insecurity for all of these groups. The patterns of the FP and LEA groups also differed from each other. Households in food poverty only (but with adequate available energy) were more likely to be rural, African, headed by females and in the low end of the income distribution. In contrast, the LEA group was more likely to be urban, non-African, headed by males and in the middle of the income distribution.

Food expenditure behaviors also differed across these groups. ANOVA showed that the main effect of food security classification was significant (P < 0.05) for all food consumption and income variables studied (Table 3 ). Most pair-wise comparisons also showed significant differences (P < 0.05) in means. Not surprisingly, food secure households reported spending more on food (906 Rands/mo) and purchased a greater number of different foods (40 out of the possible 124 discussed earlier) than did food insecure households, at 332 Rands/mo spent on 23 foods. Interestingly, the FP group reported spending roughly half the amount of the LEA group. These patterns were similar when household size was considered, i.e., when food expenditures were expressed on a per capita basis. The food secure reported spending more money on food, but because they had higher incomes, their food expenditure share was lower than other groups. On average, they spent 30.8% of their income on food. Although the FP group had the highest reported food expenditure share at 49.4%, the LEA group spent the lowest percentage of their budget on food.

The makeup of the food budget also differed across these groups. The FP group reported spending 44% of their food budget on grains and cereals and only 14% on meats, poultry, and fish, whereas, at the other extreme, the LEA group reported spending 16% on grains and cereals and 31% on meats poultry and fish. For many items, cereals, meats, dairy, fruits, sugars, other foods, and food away from home, the spending of the LEA group was closer to that of the food secure than to that of the food insecure.

Further understanding of the determinants of the different food insecurity conditions comes from the multinomial logistic regression (Table 4 ). Odds ratios were considered significant when the 95% confidence interval around them did not include 1. Controlling for other characteristics, the odds of being food insecure for African-headed households was 6.2 times the odds for the reference group of Caucasian households. Households headed by those of mixed ancestry were also more likely to be food insecure. Income and household size continued to be important predictors of food insecurity in the multivariate analysis. Those in the bottom 5th of the household income distribution had odds of being food insecure that were 13 times the odds for those in the middle of the income distribution. Households with 7 persons were also more likely to be food insecure than the reference group (households with 3–6 members). Households headed by women were less likely to be food insecure than male-headed households.

Small households or those with high incomes had lower odds of having a low energy availability than reference group households. Households headed by elderly individuals or females were also less likely to be in the LEA group, whereas urban households were more likely to be in the LEA group. Unlike the findings on the odds of being in the food insecure or FP groups, those in the lowest income quintile were not at increased risk of being in the LEA group.

For the most part, results from the multivariate analysis were similar in direction to those obtained with the bivariate analysis. One exception was for households headed by females. The multivariate results indicated they had lower odds of being food insecure than households headed by males. Households headed by females had lower incomes than households headed by males, which probably accounted for the higher food insecurity rates among this group in the bivariate results. For large households, the odds of being in the LEA group were greater than for the reference households (those with 3–6 persons). At first glance, this may seem contrary to the bivariate results, in which the percentage of large households in the LEA group was less than the percentage of reference households in the LEA group. However, the unadjusted odds of being in the LEA group for large households (10.5/11.7 = 0.90, from Table 2 ) was, in fact, greater than it was for reference households (19.3/38.2 = 0.51); thus, the two analytic methods were in agreement on this outcome. Bivariate results showed that households headed by Indians had higher odds of being in the LEA group than those headed by Caucasians. Indian households were typically more urban and of larger size than Caucasian households. Controlling for these factors may explain why the multivariate results showed Indian households with a lower odds of being in the LEA group. Households with home production had lower odds of being in either the food poverty or food insecure groups. This also contrasted with the bivariate results.

	DISCUSSION

TOP ABSTRACT INTRODUCTION SUBJECTS AND METHODS RESULTS DISCUSSION LITERATURE CITED

 
Two types of quantitative measures of household food insecurity were developed in this paper and applied to South African survey data on reported household food expenditures and estimated consumption from own production. The food poverty indicator, an economic measure of food insecurity, assessed how much the household spent relative to the cost of a basic nutritionally adequate food plan. The second indicator, a nutritional measure of food insecurity, assessed the energy availability in household food supplies. The two indicators were used in tandem to identify different types of food security problems.

Food insecure households, i.e., those with low food spending and low available energy, accounted for 39% of the South African population in 1995. This rate is consistent with results from poverty studies in that country. Woolard and Leibbrandt (22 ) found that for 6 of 7 different poverty measures, the percentage of individuals below the poverty level in 1993 ranged from 40 to 57% (26% was the rate for the other measure). The food insecurity rate of 39% reported here is a household measure. This translates to a food insecurity rate of 51%, when expressed on an individual basis because food insecure households have larger household sizes. This rate is well within the range of their poverty estimates. Although food insecurity and poverty are not the same, it is not uncommon for the overall rates to be relatively close. For example, the 1995 national survey on food security in the United States found a food insecurity rate of 12%, relatively close to the 15% poverty rate for the same year.

The socioeconomic correlates of food insecurity reported here are also consistent with previous research. Leibbrandt and Woolard (23 ) found the highest rates of poverty among Africans, followed by households headed by people of mixed ancestry, Indians and Caucasians, respectively. Households headed by females and those living in rural areas were more likely to be poor in that study. All of these are similar to the results reported here on food insecurity.

Provincial stunting rates in South African children aged 1 to 9 y from the 1999 National Food Consumption Survey (24 ) corroborate provincial results on food insecurity found here. A ranking of the provinces from the highest to the lowest on stunting rates is identical for the first six provinces to a ranking on food insecurity (not shown). The last three of the nine provinces in South Africa, i.e., those with the lowest stunting rates, were also the three with the lowest food insecurity rates, albeit in a slightly different ordering. The 1999 survey also employed a qualitative, experiential-based food security instrument, analogous to those developed in the United States. On the basis of this instrument, provincial rates of household food insecurity ranged from 48 to 91% (5 ). These rates were much higher than those based on the two indicators used here, which ranged from 20 to 54%.

As with any indicator based on survey data, there are limitations to the indicators we present here. Both indicators rely on data from the 1995 Income and Expenditure Survey. In principle, the IES instrument offered complete coverage of all food purchases and home-grown foods. The listing of 124 foods was particularly disaggregated for major staples, such as cereals, and included several "other food" categories in each of the main product breakdowns. But, as with most dietary and food security instruments, a weakness of the IES may be that it depended on a respondent’s recall. Surveys based on food expenditure records, such as the U.S. Consumer Expenditure Survey, do exist in high income countries, but are infrequent in low and middle income countries. However, analytic experience with the U.S. Nationwide Food Consumption Surveys has shown that respondent-estimated total weekly food cost is a better proxy for long-run food expenditures than an item-by-item valuation of foods reported in a 1 wk food record (25 ). Another comparison of a recall approach with a day-by-day recording of food expenditures showed quite a close correspondence between the two techniques (26 ). The IES is not directly comparable to these studies because it used a hybrid approach in which recall information was obtained, but on an item-by-item basis. Further research in consumer economics is clearly warranted to understand fully the issues of precision and accuracy in the recall of food expenditures.

A second limitation concerns the use of food price data for the food poverty indicator. For this indicator, we used the HSL series because it provided information on the cost of a basic, nutritionally acceptable food plan in many locations throughout the country. However, these were all urban centers. Rural prices for items listed in the basic food plan could have been higher than these urban prices due to increased transportation costs and perhaps decreased competition. Rural residents also might have had higher energy needs than urban residents, which is another reason why the costs of the fixed quantity food plans might have been too low in rural areas. If the cost of a basic food plan in rural areas should have been higher than the estimate we used, we would expect our estimates of rural food poverty to be too low. But rural households have opportunities for economizing on meeting their food needs, including substitution of locally produced foods, village exchange mechanisms and collection of wild foods. In principle, the Income and Expenditure Survey registered the first two of these three categories of food items because there was a separate section on home-grown foods and the value of gifts was included in the food expenditure categories. The collection of wild foods was absent from the IES, although it is not a major feature of the South African food economy. To the extent that these alternative sources were missing or inadequately reported in the IES data, we expect rural households would have done better than we estimated. This would have dampened the potential underestimate of food poverty due to higher priced rural foods, discussed above. Clearly, the estimates of rural food poverty are high, even if we must consider them only approximate.

A related concern is the use of food price data in the low energy availability indicator. For this indicator, we used national price data from Statistics South Africa because the detailed pricing of individual food items in this series allowed us to approximate quantities of foods available to a household given information on only their reported expenditures on these foods. We approximated provincial variation in food prices with information from the HSL series. This was probably an adequate adjustment at the provincial level because many economists consider spatial price variation within a country to be relatively small compared with year-to-year variations in prices. We are on more tenuous footing with respect to the urban/rural dichotomy; the issues outlined in the preceding paragraph are all relevant to this indicator as well. Statistics South Africa has emphasized the need to collect data on rural prices (27 ). Such an effort would improve the accuracy of these indicators in the future.

Another limitation of the available energy indicator is that it did not incorporate information on energy expenditures of individuals in the household. Rather, the normative reference used in this indicator was derived only from information regarding the age and gender of household members and their corresponding reference energy standards. As is standard practice in South Africa, we used U.S. standards, which assume a light-to-moderate activity level. These might have been too low for certain groups of individuals, for example, those from agricultural households involved in strenuous activities. Using our indicator, the energy available in the food supplies of these households might have appeared to be adequate, when in fact it was too low. As it is, we found that rural households were more likely than urban households to be low on the household available energy measure, 60 vs. 49%, respectively. Thus, the true gap between rural and urban households may have been even wider because energy expenditure was likely higher in the rural population. If information were available, subsequent research on this topic could make improvements in the household energy availability indicator by varying the recommended energy levels based on occupations of household members.

Despite their limitations, the indicators described here can provide insights into the types of households suffering from different types of problems. We have shown that our prevalence estimate of the food insecure, i.e., those that scored low on both indicators, is consistent with other national studies on poverty and undernutrition in South Africa. But use of the two indicators enabled us to classify households into other groups, which have different food consumption and food security profiles. The FP households, those with food poverty only, might in some sense be considered "successful shoppers" because they did not spend very much on food but had adequate household energy supplies. The LEA group, in essence, is a group of households that "shopped for other attributes," spending their money on taste, quality, convenience and other attributes in food in addition to energy. Certainly these descriptions are simplistic, but they sketch a broad picture that is supported by both the demographic characteristics and food consumption behavior described in Tables 2 3 –4 . The "successful shopper" households were more likely to be rural and to be headed by Africans or the elderly. They had lower incomes and spent a much greater share of their budget on cereals. Those that shopped for other attributes in food had habits more consistent with urban patterns of consumption, i.e., a lower percentage of income spent on food, lower food budget shares on cereals and higher shares on meats and dairy products. They were more likely to be headed by males, nonelderly, or those of mixed ancestry and had higher incomes than the other food insecure groups.

The classification of households afforded by these indicators allows for tailoring policy responses to specific needs. Clearly the food insecure, i.e., those that are low on both indicators, must enhance their purchasing power, and they should be the priority for the targeting of resources. A number of different interventions such as job creation programs, income transfers and food assistance can be (and are in South Africa) directed toward this objective. It is difficult to argue for income or food assistance transfers to the LEA group, however. On average, their reported per capita food spending was >2.5 times that of food insecure households and close to 2 times the spending levels of a group that managed to purchase enough energy, the FP group. Nutrition and consumer education would be a better alternative for this group, for example, in lowering the cost of the energy that they purchase. The FP group, the so-called successful shoppers, was a very small group, only 4% overall. This is a group for which further study is warranted, perhaps in the same vein of research as that done with positive deviants (28 ). In particular, it would be worth finding out what aspects of their food consumption behavior, if any, can be transferred to other groups or whether, in fact, their apparently adequate available energy belies an inadequacy in some other aspect of their diet.

In sum, we have developed and tested two quantitative indicators of household food consumption behavior, one based on the amount of money spent on food, the other based on the energy available in those food supplies. When used together, these indicators provide insights into different types of food insecurity in South Africa. The information can be used to geographically target the worst-off households. The measures also allow for targeting of different types of resources to households with different types of problems. Most of the data required for development of these indicators comes from household expenditure surveys, as well as food price and food composition databases. These sources are not without their limitations. However, the commonplace existence of these data sources, combined with the reasonable nature of the results found here, suggests that these food security assessment techniques may have applicability in a wide range of developing countries.

	ACKNOWLEDGMENTS

 
We would like to thank the University of Cape Town’s Southern Africa Labor Development Research Unit for providing access to the 1995 Income and Expenditure Data. We appreciate the assistance of Petro Wolmorans, Nelia Steyn and Lesley Bourne of the South African Medical Research Council for their help in using MRC’s food composition database. We also appreciate the assistance of David Stoker and Pieter Bossert from Statistics South Africa on sample design issues.

	FOOTNOTES

 
¹ Presented in part at Experimental Biology 2001, March 31–April 4, 2001, Orlando, FL [Rose, D. & Charlton, K. (2001) Objective measures of household food security in South Africa. FASEB J. 15: A253 (abs.)].

² Funding for D.R. to work on this project came in part from a Visiting Scientist Grant from the Medical Research Council of South Africa.

⁴ Abbreviations used: FP, those households in food poverty only; HSL, Household Subsistence Level series on market prices; IES, 1995 South African Income and Expenditure Survey; LEA, those households with a low energy availability only; MRC, Medical Research Council of South Africa.

Manuscript received 21 January 2002. Initial review completed 19 February 2002. Revision accepted 16 August 2002.

	LITERATURE CITED

TOP ABSTRACT INTRODUCTION SUBJECTS AND METHODS RESULTS DISCUSSION LITERATURE CITED

 

1. Andrews, M., Bickel, G. & Carlson, S. (1998) Household food security in the United States in 1995: results from the food security measurement project. Fam. Econ. Nutr. Rev 11:17-28.

2. UNICEF (1990) Strategy for Improved Nutrition of Women and Children in Developing Countries 1990 UNICEF New York, NY. .

3. Smith, L.C. & Haddad, L. (2000) Explaining Child Malnutrition in Developing Countries: A Cross-Country Analysis. Research Report no. 111 2000 International Food Policy Research Institute Washington, DC. .

4. Wolfe, W. S. & Frongillo, E. A. (2001) Building household food security measurement tools from the ground up. Food Nutr. Bull. 22:5-12.

5. Maunder, E., Gericke, G., Labadarios, D., Steyn, N., MacIntyre, U., Huskisson, J., Voster, H. H., Swart, R., Dannhauser, A. & Nesamvuni, A. E. (2000) The food procurement and household food security of children aged 1–9 years in South Africa: the National Food Consumption Survey, 1999. S. Afr. J. Clin. Nutr. 13:99.

6. Maxwell, D. (1998) Can Qualitative and Quantitative Methods Serve Complementary Purposes for Policy Research? Evidence from Accra. FCND Discussion Paper no. 40 1998 International Food Policy Research Institute Washington, DC. .

7. Chung, K., Haddad, L., Ramakrishna, J. & Riely, F. (1997) Identifying the Food Insecure: The Application of Mixed-Method Approaches in India 1997 International Food Policy Research Institute Washington, DC. .

8. Greer, J. & Thorbecke, E. (1986) A methodology for measuring food poverty applied to Kenya. J. Dev. Econ. 24:59-74.

9. Ravallion, M. (1998) Poverty Lines in Theory and Practice. LSMS Working Paper no. 133 1998 World Bank Washington, DC. .

10. Rose, D. & Charlton, K. E. (2002) Prevalence of household food poverty in South Africa: results from a large nationally representative survey. Public Health Nutr 5:383-390.[Medline]

11. Central Statistical Service (1997) Income and Expenditure of Households 1995. Statistical Release P0111 1997 Central Statistical Service Pretoria, South Africa. .

12. Office of Human Subjects Research, National Institutes of Health (1995) Guidelines for the Conduct of Research Involving Human Subjects at the National Institutes of Health 1995 OHSR, National Institutes of Health Bethesda, MD. .

13. Central Statistical Service (1996) Living in South Africa—Selected Findings of the 1995 October Household Survey 1996 Central Statistical Service Pretoria, South Africa .

14. Rose, D., Bourne, L. & Bradshaw, D. (2002) Food and Nutrient Availability in South African Households: Development of a Nationally Representative Database 2002 Medical Research Council Tygerberg, South Africa. .

15. Potgieter, J. F. (1995) The Household Subsistence Level in the Major Urban Centers of the Republic of South Africa. Fact Paper no. 100 1995 Institute for Planning Research, University of Port Elizabeth Port Elizabeth, South Africa. .

16. Food and Nutrition Board, National Research Council (1989) Recommended Dietary Allowances 10th ed. 1989 National Academy Press Washington, DC. .

17. Central Statistical Service (1996) Retail Prices 1995. Statistical Release P0141.3 1996 Central Statistical Service Pretoria, South Africa. .

18. Medical Research Council (1999) South African Food Composition Database Version 1.2 1999 Medical Research Council Tygerberg, South Africa. .

19. Agricultural Research Service, U.S. Department of Agriculture (1999) USDA Nutrient Database for Standard Reference, Release 13 1999Nutrient Data Laboratory Home Page, http://www.nal.usda.gov/fnic/foodcomp..

20. Westat, Inc. (2001) A User’s Guide to WesVar, Version 4.1 2001 Westat Rockville, MD. .

21. Charlton, K. & Rose, D. (2001) Nutrition among older adults in South Africa: the situation at the beginning of the millennium. J Nutr 131:2424S-2428S.[Abstract/Free Full Text]

22. Woolard, I. & Leibbrandt, M. (1999) Measuring Poverty in South Africa. DPRU Working Paper no. 99/33 1999 Development Policy Research Unit, University of Cape Town Cape Town, South Africa. .

23. Leibbrandt, M. & Woolard, I. (1999) Household Incomes, Poverty and Inequality in a Multivariate Framework. DPRU Working Paper no. 99/31 1999 Development Policy Research Unit, University of Cape Town Cape Town, South Africa. .

24. Labadarios, D., Maunder, E., Steyn, N., MacIntyre, U. & Nel, H. (2000) National Food Consumption Survey in Children Aged 1–9 Years: South Africa 1999. Report prepared for the Food Fortification Task Group 2000 University of Stellenbosch Stellenbosch, South Africa. .

25. Basiotis, P. P. (1992) Determining the best available variable to use as a proxy for an unobservable, economic variable: the case of food expenditures at home. Am. J. Agric. Econ. 74:1311.

26. Oliveira, V. & Rose, D. (1996) Food Expenditure Estimates from the 1995 CPS Food Security Supplement. How Do They Compare with the Consumer Expenditure Survey? ERS Staff Paper no. 9617. 1996 U.S. Department of Agriculture, Economic Research Service Washington, DC. .

27. Hirschowitz, R. (1997) Earning and Spending in South Africa: Selected Findings of the 1995 Income and Expenditure Survey 1997 Central Statistics Pretoria, South Africa. .

28. Shekar, M., Habicht, J. P. & Latham, M. C. (1992) Use of positive-negative deviant analysis to improve programme targeting and services: example from the Tamil Nadu Integrated Nutrition Project. Int. J. Epidemiol. 21:707-713.[Abstract/Free Full Text]

This Article Abstract Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Rose, D. Articles by Charlton, K. E. Search for Related Content PubMed PubMed Citation Articles by Rose, D. Articles by Charlton, K. E.