Contents - Previous - Next

This is the old United Nations University website. Visit the new site at http://unu.edu

The demand for higher-status food and nutrition in rural India: the experience of Matar Taluka.
II. Regression Analyses

C. H. Shah
Gujarat Institute of Area Planning, Ahmedabad, India

This article is a continuation of one that appeared in the previous issue of the Food and Nutrition Bulletin (vol. 8, no. 2). This segment presents the results of regression analyses to determine the influence of variables including production profiles, housing conditions, and durable asset scores on food purchase patterns. Dr. Shah identifies two components of food costs: one to obtain sufficient nutrients (nutrient costs), and the other to satisfy preferences (preference costs). In examining total expenditures on foods from both status and basic groups, he finds that households deficient in calories are more willing than non-deficient households to pay for higher-preference foods that are not as cost-effective in terms of nutrients. There is a need for studies among other populations to determine the extent to which Dr. Shah's findings concerning the purchase of status foods are a generalized phenomenon.

* * *

Regression analyses were done for (a) expenditure on food, (b) calorie function, and (c) proportions of food expenditure for the different food groups and the relative preference values for these groups. The Appendix contains definitions of abbreviations used in the regression analysis tables.

Table 22 gives the results for the regression analyses on food expenditure. We did analyses for all of the households together as well as separately for calorie-deficient and calorie-non-deficient households.

We repeated the exercise with the log of food expenditure as an alternative dependent variable, using stepwise regressions for the D and the ND households to assess the relative importance of different independent variables.

Two regressions (1 and 2) in table 22, with the dependent variables of food expenditure per capita in rupees, the log of food expenditure per capita, and two alternative specifications of total expenditure per capita, along with other micro-variables and macro-variables, gave nearly the same values for R²: 0.9306 and 0.9357. The superiority of either of the regressions and of either of the forms of specifications of per capita total expenditure was, therefore, difficult to establish. Moreover, we found that the same set of variables had statistically significant coefficients for two regressions, with two exceptions. The dependency ratio had a significant coefficient for the quadratic form only and the dummy for non-deficient households had a significant coefficient for the log-log-inverse regression only.

Both the quadratic and the log-log-inverse regression analyses gave similar results: the elasticity of expenditure on food increased with respect to total expenditure per capita. This result differed from that predicted by Engel's Law, according to which the elasticity of expenditure on food would tend to decline as the income level rose, so that for higher incomes a smaller proportion of the total income would be spent on food. The discrepancy may perhaps be attributable to the very low initial levels of income.

Educational variables both at the village and household levels had significant coefficients, but with opposite signs—positive for the village-level variable and negative for the household-level variable. The elasticity in both cases was low. The opposite signs of the coefficients implied that households in villages with higher literacy levels tended on the whole to have higher expenditures on food, but, for any given village, households with a larger percentage of literates and better educated members had relatively lower per capita expenditures for food.

Two other variables had statistically significant coefficients: the amount of current borrowings and the percentage of food supplies obtained from home-produced food. Although its value was low, the negative coefficient for loans suggested that those who borrowed committed a portion of their total income to repaying the loan and paying interest, so that they had to spend a somewhat smaller amount on food. Those who produced their own food tended to spend more, at comparable prices, on food. If a price differential between food purchased and procured from a household's own produce was allowed for, in all probability the coefficient might turn out to be statistically non-significant.

TABLE 22. Food functions

The variables at the village level that influenced the amount of expenditure on food, when overall expenditure was held constant, were: current income level, literacy, permanent income (durable assets), loans, and home supplies.

The regressions for the D and the ND households were in the log-log-inverse form. Values of R² for both were high: 0.87 for the D households and 0.92 for ND households. Variables with statistically significant coefficients common to both groups were total expenditure per capita, the durable asset score, and the amount of home-grown supplies. The elasticity of expenditure on food with respect to total expenditure per capita would be lower for the D households, rising to approach 1.01, and that for the ND households would be relatively higher, rising to approach 1.21. Thus, those who had better nutritional status would tend to spend more from additional income and viceversa. Both the D and the ND households would tend to increase expenditures on food with increasing supplies obtained from their own production, but in this regard the ND households had greater elasticity.

The durable assets score had negative, significant, and nearly comparable elasticity for both the D and the ND groups. Because the D group had an edge over the ND group, the former would have a relatively slightly smaller amount devoted to food from additional total expenditure. Thus, if at all, the D group of households would spend less, not more, on food out of additional income compared to the ND group.

More interesting is the comparison of variables that had relevance to one group and not to the other. The size of household, production conditions (at village level (V₁)), literacy, food habit score, and dependency ratio influence expenditure on food for the D group of households. The first two had a direct relationship, and the others an inverse relationship with expenditures on food for this group. In the case of the ND group, housing conditions (V₂) and borrowing had negative influences on food expenditures, while retail outlets (V₆) and education (V₇) had positive influences. The elasticities of food expenditures in all these cases were low.

Step regressions revealed a slightly different picture. Their R² values were 0.87 and 0.92 respectively for the D and ND groups respectively. For the D group of households, none of the village variables entered the final step. But in the case of the ND group, all the variables that had significant coefficients when all variables were considered entered the final step regression.

The conclusion that emerged was that the D group of households was influenced mostly by household-level variables. In the case of the ND group, the specific influence was that of borrowing, which showed a mixed pattern for the D and ND households. The D group showed the negative influence of food habit, dependency ratio, and the extent of literacy. Thus, the behavioural patterns of the D and ND groups regarding expenditures on food in response to stimuli of mainly economic factors seemed to differ.

TABLE 23. Preference component: demand function (quadratic). Dependent variable: expenditure on status food less nutritional worth

a. Figures in parentheses refer to standard errors of coefficients.
* Significant at 5 per cent level.
** Significant at 1 per cent level.

STATUS FOODS AND BASIC FOODS

We have already examined the allocation of expenses on basic foods and different categoreis of status foods for D and ND groups of households. A distinct pattern of preference among D group households emerged: D households seemed to prefer the "taste" or "quality" of food, whereas the ND households emphasized the nutritional efficiency of their food expenditures (table 23).

Two alternative forms of functions—quadratic and log-log-inverse—were used for the total expenditure variable. Other variables were entered linearly and were measured in their original units. We ran regressions for (a) D households, (b) ND households, and (c) all households, and fitted separate functions for expenditures on (a) high-status foods, (b) medium-status foods, (c) low-status foods, and (d) basic foods. We also fitted quadratic functions for expenditures on all status foods together, and repeated regressions for a preference component, i.e. expenditure less nutritional worth of food (tables 24-26).

MARGINAL PROPENSITIES TO CONSUME PREFERRED FOODS

If a household has an additional rupee to spend, how is it allocated between nutritional and preference components?

The results of regression equations are shown in the form of marginal propensities in table 27.

Of every additional rupee increase in total expenditure, nearly half would be spent on food. There were no significant differences in overall allocations for expenditures on food among low-income, medium-income and high-income expenditure groups.

There were differences observed in allocations between (a) all status foods and (b) basic foods. Status foods claimed about 81 per cent of additional expenditures on food. We have excluded tea, coffee, and spices which claimed about 10 per cent of the total expenditures on food. The poor devoted relatively more of their incomes to basic foods, that is, 30 per cent compared to 15 per cent for the highest expenditure group or the rich. On the whole, as well as for each of the three expenditure groups, basic foods claimed a low proportion of additional to total food expenditures. The preference shifted from low- to high-status food as total expenditure increased. Medium-status foods seemed to claim a constant percentage of the additional expenditures on food.

NUTRITIONAL EFFICIENCY AND PREFERENCE

Table 28 shows the percentage status of nutrition worth and "preference or taste" component.

When total expenditure increased by one rupee, about 48 per cent of the increase went to satisfying taste. Even the poor (40 per cent of the total households) spent over two-thirds of this additional rupee on satisfying taste. As was expected, the taste component was distinctly higher for the rich (the upper 30 per cent of households). The shares of different preference or status foods in total and "preference" components are given in table 29.

TABLE 24. Preference component: demand function (semi-log inverse). Dependent variable: expenditure on preference (Stat. F) (repees)

a. Figures in parentheses refer to standard errors of coefficients.
* Significant at 5 per cent level.
**Significant at 1 per cent level.

Demand elasticities for deficient and non-deficient households show the basic differences in the behavioural patterns of the two groups. The elasticities given in table 30 are totals. We allowed variations in levels of other explanatory variables in the regressions as we varied total expenditure levels for the poor, middle-level, and upper-level households.

Two distinct features of the response patterns for the D and ND groups can be seen in table 30. First, for the D group, demand elasticities were nearly constant, only mildly increasing from lower to higher expenditure classes. In contrast, for the ND group, demand elasticities increased sharply from very low levels for the low-expenditure class. Consequently, among the poor, the D households had relatively much higher demand elasticities. For the middle expenditure class, the D households stir) had higher demand elasticities relative to those observed for the ND households. For the upper-expenditure class the situation was reversed: among the rich, the demand elasticities for the ND households were higher. These differences in elasticities obtained for the D and ND households for all categories of food and for preference components in food categories of different status.

The behaviour pattern of the two groups of households differed on the inter-food-group elasticities. For the D group of households we consistently found the highest elasticity for expenditure on low-status (preference) foods, followed by high- and medium-status foods. The demand elasticities for the basic foods were the lowest. The pattern of demand elasticities for the preference component was different. The low-preference food group had the lowest elasticities for the preference component in it, while the highest elasticities were obtained for the preference component in the medium-preference foods. For high- and medium-status foods, the demand elasticities for the preference component were higher than those for the respective status foods themselves. The reverse was true for the low-status foods. This finding suggested that demand for the preference component was dominant in the high- and medium-status foods. Both the poor and rich households in the D group seemed to have the same preference pattern for status foods; in other words, there was a high preference component in these foods. Their food preference pattern was adopted at low levels of total expenditure per capita and maintained over expenditure levels.

The response behaviour of the ND group was distinctly different. The ND group households did not have constant elasticities for the poor, middle, and rich classes: the elasticities increased for the preference foods and the preference component and declined for the basic foods. Among the ND households who were poor, expenditures on basic foods were predominant. Among the ND households in the upper expenditure class high-status foods were in a dominant position because they had the highest demand elasticities.

The ND group of households seemed to emphasize nutrition at low levels of per capita expenditure even though their energy consumption was not low (i.e. below the recommended daily allowance). The ND households in the upper expenditure class seemed to make a priority of taste over nutrition. The demand elasticities component for the poor ND households was lower than the demand elasticities for expenditures on different groups of status foods; for the rich the relative magnitudes of elasticities favoured high-status foods.

Continue

Contents - Previous - Next