Household income, women’s income share and food calorie intake in South Western Nigeria

Food Policy 29 (2004) 507–530 www.elsevier.com/locate/foodpol

Household income, womenÕs income share and food calorie intake in South Western Nigeria Adebayo B. Aromolaran

*

Economic Growth Center, Yale University, No. 27, Hill House Avenue, P.O. Box 208269, New Haven, CT 06520-8269, USA Federal University of Agriculture, PMB 2240 Abeokuta, Ogun State, Nigeria

Abstract Current debates on the nature of intra-household preferences and resource control raises the question of how intra-household redistribution of income from men and women compares with increasing total household income as strategies for increasing calorie intake among low income households. I estimate calorie-income and calorie-womenÕs income share elasticities for a random sample of low income households from the rural areas of south western Nigeria. I find that calorie-income elasticity is positive and small, 0.194, but four times as large as calorie-womenÕs income share elasticity, suggesting that calorie intake is affected more by total disposable income than by its distribution pattern. Second, I find a small and negative effect of womenÕs income share on per caput calorie intake,
0.04, which is a rejection of the hypothesis that increases in womenÕs share of income will increase calorie intake and suggests that intra-household income re-distribution from men to women may not be an effective policy for increasing food calorie intake among households in the study area.
2004 Elsevier Ltd. All rights reserved. Keywords: Nigeria; WomenÕs income share; Household per capita expenditure and calorie intake

* Present address: Economic Growth Center, Yale University, No. 27, Hill House Avenue, P.O. Box 208269, New Haven, CT 06520-8269, USA. Tel.: +1 203 436 4827; fax: +1 203 432 3898. E-mail addresses: [email protected], [email protected] (A.B. Aromolaran).

0306-9192/$ - see front matter
2004 Elsevier Ltd. All rights reserved. doi:10.1016/j.foodpol.2004.07.002

508

A.B. Aromolaran / Food Policy 29 (2004) 507–530

Introduction A wide range of empirical literature has provided evidence that the level of percapita calorie intake has a strong positive but non-linear relationship with household income, after controlling for household and demographic variables (Boius et al., 1992; Subramanian and Deaton, 1996; Grimard, 1996). 1 Prior to 1987, calorie-income elasticity for low-income populations throughout the developing world was estimated to be between 0.4 and 0.8 (Boius and Haddad, 1992). Thus, income increases for the poor as a food policy strategy have received strong justification in that it is expected to reduce malnutrition (Alderman, 1986). However, Behrman et al. (1987) analyzed ICRISAT data for India, and found calorie-income elasticity estimates that were not significantly different from zero. They concluded that the linkage between income and nutrient consumption is weak and that nutrient improvements should not be expected with income gains in low income communities. This result was reinforced by Boius et al. (1992) who estimated calorie intake-total expenditure elasticity ranging between 0.08 and 0.14 with four different estimation techniques using a sample of Philippine farm households. Boius and Haddad (1992), argue that several studies after Behrman et al. (1987) reported calorie-income elasticity estimates which are in most cases lower than 0.2. This revisionist school attributed the previous high estimates of calorie-income elasticity to two major sources. The first source is the wide use of calories estimated from food expenditure data rather than direct collection of data on food intake quantities, due to the dearth of food quantity information in household surveys. The second source of upward bias is believed to be the endogeneity of household income in the calorie intake – income model. The use of calorie intake quantities from food expenditure data creates two kinds of upward bias in the estimate of calorie-income elasticity. The first is non-classical measurement error bias 2 and the second is aggregation 1 The article by Boius et al. (1992) presented results of 30 investigations into calorie-income elasticity between 1979 and 1991. The range of calorie-income elasticity estimates for those who used calories from food expenditures was 0.22–1.18, while estimates from studies that used calories from 24 h recall of quantity intake range between 0.01 and 0.37. Subramanian and Deaton (1996) estimated calorie-total expenditure elasticity of between 0.3 and 0.5 for households in rural Maharashtra in India. Grimard (1996) reported the calorie-expenditure elasticity for urban Pakistan to range between 0.51 and 0.25 from low to high-income households and 0.62 to 0.35 for the rural sector. 2 The non-classical measurement error bias can arise in two ways. First, when expenditure is used as a proxy for income (as in this paper) and calorie intake quantity data are derived from food expenditure data, the measurement error in both the left and right hand side of the equation are correlated. In other words, when calorie intake quantity is calculated as the ratio of food expenditure to the price, part of the measurement error in the expenditure variable (when it proxies for income) on the right hand side is carried over to the calorie quantity measure on the left hand side of the equation. This is called nonclassical measurement error and results in upward bias of the income elasticity of calorie intake. Second, because expenditure data does not distinguish between food consumed at home and those given as gifts to the less privileged, calorie quantity intake for the rich is over estimated and that of the poor is underestimated. This results in an upward bias in calorie-income elasticity. According to Boius (1994), failure to take accurate account of food transfers from the rich to the poor is a source of upward bias for calorie-income elasticity generated from food expenditure survey.

A.B. Aromolaran / Food Policy 29 (2004) 507–530

509

bias. 3 This study addresses these problems by using data collected directly on actual food intake quantities to estimate calorie-income elasticity. Upward bias in calorie-income elasticity estimate which results from endogeneity of household income is also of two kinds namely: simultaneity bias 4 and omitted variable bias. 5 In this study, I adopt the instrumental variable technique (IV) to address this problem of endogeneity of income. On the other hand, the conventional school of thought which support the traditional view that calorie-income elasticities are sizeable at least among low income households argue that recent low estimates of calorie-income elasticity in households could arise from two sources. First is the frequent use of current income as a measure of wealth rather than permanent income (Behrman and Deolalikar, 1990), 6 while the second is measurement error in income and expenditure. A major problem associated with data collected in developing countries is the difficulty in obtaining accurate data on income and expenditure. 7 Both the use of current income and measurement error in the data would bias the ordinary least square regression estimate of calorie-income elasticity downwards. 8 Per-capita expenditure is therefore used in place of current income in this study since it is a better proxy for permanent income. Furthermore, the use of per capita expenditure as proxy for income reduces measurement error in income since it is easier to get information on expenditures than on income in developing countries. To further reduce the magnitude of error, I used income, expenditure and daily calorie intake values which are averages over 12 fortnightly visits to each sample household.

3 Aggregation bias describes a situation in which consumers substitute more expensive and low concentration sources of calories in the diet for less expensive and high calorie concentration foods as income increases. Consequently, when food calorie intake is estimated from food expenditure survey data with no quantity information, this substitution effect is lost in aggregation of the two types of foods and results in the overestimation of calorie-income elasticity. 4 This is due to the reverse causation between calorie intake and income and is based on the nutritionproductivity hypothesis. Strauss (1986), using Sierra Leone data found a highly significant effect of calorie intake on labor productivity, and took this to be an evidence in support of the reverse causation hypothesis. 5 This bias arises because labor income is a choice variable in the household decision model that allows for labor supply/leisure choices. If both labor supply and consumption decisions are driven by same unobserved factor such as Ôtaste for workÕ, then some proportion of the estimated effect of income on calorie intake may be the result of spurious correlations between calorie intake and income rather than the evidence of a causative effect from income to calorie intake. 6 The reasons for this are twofold: First, current income is a very noisy measure of the wealth flow into a household and this enlarges the value in the denominator of the regression coefficient estimator. Second, the covariance between current income and food consumption is believed to be lower than that between permanent income and food consumption (permanent income hypothesis). Thus if current income is used, the numerator of the regression coefficient estimator is understated. The interaction of these two effects would lead to a downward bias in the estimate of the regression coefficient of income in the food calorie intake equation. 7 Two major reasons for this are: the disproportionately large informal sector with little or no formal income records, and the general view that personal income is a private good and is not meant for public consumption. 8 This bias is called the classical measurement error bias or attenuation bias.

510

A.B. Aromolaran / Food Policy 29 (2004) 507–530

The second point of investigation in this study is the effect of women share of income on calorie intake in low income households. Studies that investigate the impact of changes in household resource control pattern on calorie intake are not common. No such study is available for Nigeria. Empirical investigation into the relationship between women income share and other outcome variables such as food expenditure share, fertility, child survival, young childrenÕs weight for height, and household expenditure are more common (Thomas, 1990, 1992, 1993). Hoddinott and Haddad (1995), using data from Cote De ÔIvoire found a positive but small marginal effect of womenÕs income share on household food budget share. A doubling of the proportion of household cash income received by wives would lead to a meagre 1.9% rise in budget share of food eaten within the household. Hopkins et al. (1994) found that in Niger that changes in female annual income, while controlling for male income impacted positively on household food expenditures. These results, they claim, hold for both earned and non-labor income. Thomas (1990, 1992, 1993) found that in Brazil, the identity of the household member controlling income affects nutrient intake, fertility, child survival, and young childrenÕs weight for height, and household expenditure. On the whole, the observed impact of womenÕs income share on household expenditure patterns is generally agreed to be a reflection of gender differentiated preferences. However empirical literature on Africa concerning this issue is still very limited and is principally because of the dearth of disaggregated household data on income and consumption. The analysis in this paper in based on household consumption, income, and expenditure data disaggregated by gender and thus gives us a rare opportunity to examine the effect of changes in the pattern of resource control on per caput calorie intake. Not many studies have examined the effect of changes in women income share on the quantity of calorie intake by household members in Africa. Most of the previous studies that examined the influence of women income on food consumption estimated women income share elasticity for food expenditure or food expenditure shares, rather than calorie intake which is the focus in this paper. This paper hopes to contribute to the growing literature on determinants of calorie intake at the household level by investigating the response of calorie intake of individuals in the household to increases in both per capita expenditure and the share of household income controlled by women. The model estimated in this paper also allows for differential income responses for individuals, taking account of the effect of age and sex. 9 The major findings of this study are as follows. First, calorie-income elasticity is estimated by IV as 0.194 and this suggests that calorie intake does not get a substantial share of marginal increases in household income. Thus, increasing household disposable income may not be a very effective strategy for bringing about increased food energy intake among low income populations in Nigeria. The instrumental variable (IV) estimate of calorie intake-womenÕs income share elasticity is
0.04. We show that this negative marginal response is neither a consequence of reallocation of income towards more expensive and low calorie food 9 Not many existing studies have used individual level data to examine calorie-income response relationships as used in this paper (Behrman and Deolalikar, 1990).

A.B. Aromolaran / Food Policy 29 (2004) 507–530

511

sources nor an evidence of positive transaction cost resulting from imperfect substitution between male and female income in food consumption. This estimate is a rejection of the hypothesis that per caput calorie intake is positively related to increasing women share of household income and suggests that wealth redistribution from men to women would not increase per caput food energy intake in low income households in rural south western Nigeria. Furthermore, the estimate of calorie-income elasticity is about four times the size of the estimate of calorie intake-womenÕs income share elasticity. This suggests that female-income targeting may not be as effective in raising calorie intake levels of household members as male-income targeting. In addition, the results show that the OLS estimates of per capita expenditure and women share of income coefficient are smaller than the 2SLS estimates of the same coefficient. This downward bias of OLS estimate of elasticity coefficient is an indication that classical measurement error bias may be a more important source of bias in this investigation than the other potential sources of bias discussed earlier. Finally, it is observed that the OLS and 2SLS estimates of women share of income are both negative, showing some consistency in the sign attached to the coefficient of per capita expenditure and womenÕs income variables irrespective of type of estimator (i.e. OLS or 2SLS). Importance of food calories intake Economic analysis of calorie consumption by households derives from the important role calories play in the definition of important welfare concepts such as health and productivity. 10 According to Maxwell and Frankenberger (1992), enough food is mostly defined with emphasis on calorie and on requirements for an active, healthy life rather than simple survival. 11 Food calorie intake has been found to have a strong empirical linkage with both human health and productivity. The human body needs energy to maintain normal body function (basic metabolic rate), engage in required minimal activity related to good health and hygiene (standard minimum requirement), and carry out productive activities to sustain the supply of energy and other required nutrients to the body. In addition, food calorie intake is needed for growth among children and can also affect the assimilation of micronutrients, since the body may fail to assimilate other nutrients if there is food energy deficiency. The level of calorie intake (both stock and flow) by an individual should therefore be adequate to sustain these functions over his expected lifetime. When this life time calorie consumption pattern falls short of a minimum threshold, the individual is at a health risk. Secondly whenever there is a persistent short fall in the flow of calorie intake relative to the amount required for optimal productive activity, the inflow of other nutrient intakes is likely to be affected 10 See Stiglitz (1976) for a detailed discussion of the efficiency wage hypothesis which provides the theoretical framework for understanding the link between productivity and calorie intake. 11 Food energy requirements are often used as proxy for all nutritional requirements, even though adequacy in calories may occur simultaneously with serious deficiencies in other nutrients.

512

A.B. Aromolaran / Food Policy 29 (2004) 507–530

since the resources required to acquire these nutrients is obtained from productive work. This situation is especially true in populations where the major income-earning asset is human labor efforts. That is, populations made up of poor households where non-earned income forms an insignificant component of full income. In such populations, increased calorie intake may imply increased productivity, increased income and thus increased nutrition. 12 Increased nutrition is associated with sustained increments in productivity and thus sustained access to food energy intake. Theoretical framework The analysis in this paper is approached from the point of view of a cooperative bargaining household model. There are two general classes of household models namely: income pooling and bargaining models. The income pooling models assume that household demand (or expenditure share) is not affected by the identity of the individual that earns the income. The effective constraint on the household welfare function is the pooled household income. On the other hand, the bargaining model throws away the income pooling assumption of the income pooling models and allows for the explicit effect of distributional factors on demand or expenditure share. Income pooling models are of two types – unitary models and collective models of the household. The unitary model (which is a restricted form of income pooling models) assumes that the preferences of household members are uniform or that the preference of just one household member (the household head or dictator) is imposed on all other members. The unitary household thus maximizes a welfare function whose only component is the utility function of the dictator or household head. On the other hand, the collective model is a more general formulation of the income pooling hypothesis. Unlike in the unitary model, it allows for differences in preferences between actors within the household. In the set of collective models, household welfare function is defined as I

Uh ¼ i¼1

X

Ui U i ;

i ¼ 1; . . . ; I

ð3:1Þ

and Ui = K where Uh is the household welfare function, Ui is individual iÕs utility function in a household with I individuals. Ui is the welfare or Pareto weight attached to the utility function of each individual i. By characterization of this model, Ui is fixed and does not change with changes in factors that affect resource control power (also called distributional factors) within the household such as individual income, assets, and schooling. Thus, changes in distributional factors do not affect the relative expenditure share on goods in a collective model. It is assumed that Ui is set from marriage and does not change throughout the life time of the household. So the collective model results in a demand system derived from a Pareto optimal allocation with a fixed Ui vector. There is no movement along the utility possibility frontier of 12 Using household level data from Sierra Leone, Strauss (1986) found highly significant effect of calorie intake on labor productivity.

A.B. Aromolaran / Food Policy 29 (2004) 507–530

513

the household, since only one point on the contract curve is relevant. 13 The predictions of the collective model is that income changes only affect household demand directly through the Slutsky income effect and not through the power sharing factor in the household welfare function. The unitary model is a special case of the collective model where Ui = [1,0]. The second broad class of household models is the non-income pooling or bargaining model. This is the model that is assumed for the analysis carried out in this paper. The bargaining model jettisons the income pooling assumption of the collective model and allows for the explicit effect of distributional factors on household demand or expenditure. The model specifically assumes that Ui 6¼ [1, 0] and that Ui 6¼ K. Thus, the whole range of utility possibility frontier is the set of feasible Pareto optimal allocations for the decision making environment characterized by bargaining. In this model, changes in power sharing or distributional factors are expected to lead to changes in Ui and the changes in Ui are in turn expected to result in changing demand pattern or expenditure shares. Thus, the bargaining model predicts that income changes affect demand both directly (through the Slutsky income effect) and indirectly (through the power sharing factor, Ui). Assume that a household in this study is made up of a man (m), a woman (f), and others members who are non-income earners (c); individuals in the household have differentiated preferences; and household income is not pooled. Suppose that the each individual in the household derive utility from two composite good: calorie/energy producing good, C, and non-calorie producing goods, Q. Calorie itself cannot be purchased but its intake depends on the amount of food item Xj consumed. The amount of food item, Xj, consumed in turn depends on its price, Pj, and a number of tastes factors such as characteristics of the individual (ci) and household level characteristics (ch). We assume that the pareto/welfare weights of the man, Um, and the woman, Uf, sum to unity, implying that other members of the household, (c), who have no bargaining power have pareto weights Uc = 0. Also the household income, Yh, is the sum of the individual incomes of the man, Ym, and the woman, Yf. These individualized incomes are called power sharing factors. Given a particular level of household income, higher levels of Yf would imply higher bargaining power for the woman or higher Uf. Thus Uf is a function of the distributional/power sharing factor Yf/Yh. The household solves the maximization problem stated in expressions (3.2)–(3.7): Maximize U h ¼ Um U m ðC; QÞ þ Uf U f ðC; QÞ:

ð3:2Þ

Subject to:

13

Y h ¼ pj X j þ Q;

ð3:3Þ

Y h ¼ Y m þ Y f;

ð3:4Þ

C ¼ CðX j ; ci ; ch Þ;

ð3:5Þ

The contract curve is a locus of Pareto optimal allocations for the household.

514

A.B. Aromolaran / Food Policy 29 (2004) 507–530

Ui 6¼ K;

where K is a constant and i ¼ ðm; fÞ;

Uf ¼ Uf ðY f =Y h Þ

and

Um ¼ ð1
Uf Þ:

ð3:6Þ ð3:7Þ

From this constrained maximization problem, we derive an optimal demand function for calorie intake as a function of prices, household income, a power sharing or distributional factor, individual and household level characteristics. Formally, C ¼ CðX j ðpÞ; Y h ; Uf ðY f =Y h Þ; ci ; ch Þ:

ð3:8Þ

Data Data collection procedure A combination of cluster and systematic random sampling procedure was used to select 40 households from each of 12 rural/semi-urban communities in 3 of the 6 states that make up south-western Nigeria. 14 Thus, a total of 160 households per state and 480 households in all were selected. However, the analysis used information from only 472 households, amounting to 2573 individuals who are at least 2 years of age (See Table 1). 15 The needed information was collected through the use of interview schedules/questionnaires that were personally administered by the field assistants and field supervisors. Food price data were obtained through community market surveys. 16 Food consumption information was collected on individual basis from the households. Quantities of daily food intake were collected for each member of the household, using a 48-h recall method. 17 Each household was visited at least once in two weeks. Data were collected over a period of 6 months (October 1999–March 2000). Thus daily food intake quantities for each individual in the household were collected two times every month for a total of twelve times in 6 months. The analysis reported here was based on per caput daily food consumption averaged over the 6 months of data collection. This is designed to reduce measurement error in food consumption and smooth day-to-day fluctuations in food intake. These quantities were then converted into kilogram units. The daily calorie intake level of the individual Ci, was computed through the formula: 14

See Aromolaran (2001) for detailed description. Children below 2 years were not considered in the analysis because many of them were still being breast-fed. 16 Food prices only vary across the 12 local government areas. 17 Individuals were asked about the amount of food they consumed in the last 24 h, and then in the preceding 24 h. There was no direct weighing of food quantities. The method used to obtain the measure of food quantities was indirect. A standard size of each major food item was prepared and weighed at the research office. These physical measures were taken as the unit of measurement on the field. Each survey personnel carried the physical measures with them and used them to assist the individuals in assessing the quantities of food items taken in the past 48 h. For example, if one respondent says he consumed twice the field unit for a particular food item, then his intake of that item in kilograms will be the weight of the standardized field unit multiplied by two. 15

Table 1 Summary of sample design Selected rural local government areas

Name of selected local community

OGUN

Egbado North Obafemi Owode Remo North Ijebu East

Aiyetoro Owode Isara Ogbere

40 40 40 40

207 194 211 202

40 39 39 40

191 175 171 187

ONDO

Irele Ondo East Akure North Akoko North west

Irele Bolorunduro ItaIju/Ogbolu Oke-Agbe

40 40 40 40

348 254 229 236

38 36 40 40

196 206 228 230

OYO

Ibarapa North Itesiwaju Surulere Saki East

Ayete Otu Iresa Ago Amadu

40 40 40 40

278 322 241 201

40 40 40 40

250 309 226 204

480

2923

472

2573

South Western Nigeria

No. of households sampled

No. of individual sampled

No. of households retained for analysis

No. of individual retained for analysis

A.B. Aromolaran / Food Policy 29 (2004) 507–530

State

Source: Aromolaran (2001).

515

516

Table 2 Identification of instruments, over-identification test and first stage regressions Mean (SD)

Estimate of predictive power of identifying instruments

Full first stage regressions

Log of per-capita expenditure

Women share of income

124,533 (269,392)

Log of per caput daily calorie intake 8.52e
09 (0.31)

4.50e
07 (9.80)

7.56e
06 (7.33)

Log of per capita expenditure 1.27e
07 (4.80)

0.419 (0.575)


0.0371 (
1.75)


0.441 (
1.72)

0.868 (1.18)


0.224 (
11.24)


5.72e
06 (
6.56) 1.390 (1.34)

0.0736 (0.180)

0.0550 (
1.05)

1.362 (10.1)

13.83 (5.21)

0.312 (5.1)


0.799 (
0.32)

0.304 (0.342)


0.0358 (
1.59)


0.0286 (
0.064)

39.20 (22.0)

0.0230 (0.71)

39.31 (23.6)

Included

Not included

Not included

Not included

Not included

Included Included

Not included Not included

Not included Not included

Not included Included

Not included Included

5.946 (23.5) 0.000 0.637 2360

7.145 (239) 0.000 0.11.34 2360

16.30 (23.7) 0..000 0.253 2360

7.999 (20.3) 0.000 0.709 2360


104.7 (
6.39) 0.000 0.500 2360

Dependent variable

Value of household farm business asset Wife/husband ratio of years of schooling Women share of household farmland value Women share of household business asset Log of per-capita expenditure Women share of income All other explanatory variables in the original equation Constant Prob > F Adjusted R2 No. of observation

Women share of income

Figures in parenthesis are t-values. F-test result in the last 2 columns is the p-value resulting from the joint significance test on the four identifying instruments while controlling for the other included variables. The R2 estimates in columns 4 and 5 are partial R2 and they show the predictive power of the identifying instruments in the expenditure and womenÕs share of income equations.

A.B. Aromolaran / Food Policy 29 (2004) 507–530

Overidentification test

A.B. Aromolaran / Food Policy 29 (2004) 507–530

Ci ¼

m X

Aij Bj ;

517

ð4:1Þ

j¼1

where Aij is the weight in grams of the average daily intake of food commodity j by individual i. Bj is the standard measure of calorie found in each type of food commodity Aj. A total of 46 food items that were considered common food items in the area constituted our basket of food. 18 (i.e. J = 46). Income from different occupational sources was obtained on a fortnightly basis for a period of 6 months. No distinction was made between labor and non-labor income in the computation of the income variable used in this analysis 19 and all income data were collected from each income-earning members of the household. Expenditure information by item was also collected every two weeks from each income-earning member of the household. Socio-economic characteristics of households The average household size in the sample consists of 7 persons. 12 percent of household heads and 8 percent of senior wives have some tertiary education, while 44 percent of household heads and 48 percent of senior wives have no formal education. The average husband in the study area has 5.1 years of schooling while the average wife has 4.3 years of schooling. Only 24 percent of the household heads and 9 percent of senior wives are wage earners. Thus, most household heads and senior wives are non-wage earners, have attended school for less than 6 years (Table 3). Table 3 further shows that womenÕs share of household income is about 31 percent and womenÕs share of total household food expenditure is 32 percent. 20 We also observe that the food share in total womenÕs consumption expenditure is 32 percent, indicating that women spend more of their income on non-food consumption items. WomenÕs share of own-farm produce consumed at home is as low as 17 percent,

18 The food items include Yam, Pounded Yam, Asaro (Mashed Yam Meal), Amala Isu (Dried Yam Dough) Eba, Fufu (Wet Cassava Flour Dough), Amala Lafun (dried cassava flour dough), Boiled Beans, Moinmoin (boiled bean dough) Akara (fried bean dough), Boiled Rice, Ogi (Liquid Maize Pap), Eko (Solid Maize Pap), Boiled Maize, Potatoes, Butter, Milk, Tea/coffee, Sugar, Beverages (e.g. Bournvita, Milo) Plantain/Dodo, Bread, Biscuit/snacks, Vegetables, Pomo (Cow Skin), Beef, Pork, Sheep/Goat Meat, Chicken, Eggs, Fish, Orange, Pawpaw, Banana, Vegetable, Oil, Pepper/Tomato, Palm Oil, Gari, Cocoyam, Melon, Okro, Groundnut, Cray Fsh, Snail, Games (Bush Meat), Elubo Isu (yam flour). 19 Given the possibility of reverse causality between calorie intake and labor income, the use of nonlabor income instead of the sum of both labor and non-labor income on the right hand side would have been a good idea. However, this could not be done in this case because of non-availability of separate useable data on non-labor income. 20 Incomes earned were recorded separately for men and women. Expenditures made by men and women were also recorded separately. Both income and expenditure measures included farm produce consumed at home. To avoid double counting, only expenditures made from own-income were recorded for an individual. For example a woman may expend 100 units of money on consumption during the month. If 80 percent of it were transfers from her husband, then her own expenditure is just 20 units of money.

518

A.B. Aromolaran / Food Policy 29 (2004) 507–530

Table 3 Description of model variables Variable

Mean

SD

Individual daily calorie intake (kcal) Unit cost of calorie intake (Naira/1000 kcal) Per capita consumption expenditure (N/month) Natural log of Individual daily calorie intake Natural log of Individual daily food expenditure Natural log calorie price Natural log of per capita consumption expenditure WomenÕs share of household income (ratio) WomenÕs share of total business asset (%) WomenÕs share of farm size (%) WomenÕs share of land value (%) Ratio of senior wifeÕs to husbandÕs education WomenÕs share of food expenditure (%) WomenÕs food share in total women expenditure (%) WomenÕs share of home consumed farm produce. WomenÕs share of value of crops on farm. Household size (number of persons) Age of Individual Years of education of household head Years of education of senior wife Cash purchase share in total food expenditure (%) Number of wives Number of female aged in household (60+) Number of male aged in household Number of female adults in household (19–59) Number of male adults in households (19–59) Number of female adolescents in household (11–18 years) Number of male adolescents in household (11–18 years) Yam price (N/kg) Cassava flour price(N/kg) Yam flour price (N/kg) Cowpeas price (N/kg) Rice price (N/kg) Maize price (N/kg) Gari price (N/kg) Palm oil price (N/kg) Beef price (N/kg) Fish price (N/kg)

2204 28.57 2135 7.591 3.998
3.593 7.291 0.305 30.42 14.4 7.03 0.42 32.1 32.0 17.0 11.11 7.11 26.7 5.08 4.31 57.4 1.3 0.17 0.35 1.59 1.26 0.69 0.81 13.0 27.7 60.34 48.5 56.07 30.10 14.56 102.07 249.3 152.6

969 7.69 1894 0.486 0.509 0.280 0.901 0.274 34.12 29.4 18.0 0.58 30.1 24.6 30.5 26.8 3.26 19.4 5.43 5.02 32.7 0.81 .45 0.55 1.0 1.23 1.0 092 7.46 20.0 34.7 9.3 4.25 8.71 3.97 17.11 138.1 115.8

Source: Field survey, August 1999–April 2000.

showing that men supply most of the farm produce consumed at home. WomenÕs share of total farm size is only 14 percent. 21 Furthermore, while womenÕs share of 21 Not all plots are individually owned, since we have both individual and family farms. Yet all plots could be easily assigned to either the husband or the wife. All family farms are assigned to the household head (who are in 88% of the time males). This is because the men fully control the incomes and expenditure from these farms. Plots assigned to women are those that belong solely to the women; plots upon which they exercise a dominating control over income and expenditure patterns. So this variable can be alternatively called womenÕs ownership share of total acreage farmed by household.

A.B. Aromolaran / Food Policy 29 (2004) 507–530

519

total farmland value is just 7 percent, womenÕs share of business assets is about 30 percent, implying a lower involvement of women in farming compared with nonfarm businesses. Average calorie intake per capita is 2204 kcal, which is below the 2500 kcal recommended by FAO as standard minimum daily requirement. This shortfall is aggravated by the fact that poor families generated virtually all their incomes through direct labor efforts. The results also show that 57.4 percent of total value of food consumed at home is purchased from the market.

Empirical model The major hypotheses to be tested by the empirical model specified here are that:  Increases in household income would increase per-caput calorie intake in low income rural households in south western Nigeria.  Increases in womenÕs income share conditional on total household income would increase per-caput calorie intake in rural south western Nigeria. The structural form equation derived from the individual preference model adopted as framework for this study is represented as C ¼ a0 þ a1 X þ a2 W þ a3 I þ a4 H þ a5 P þ e;

ð5:1Þ

where C is natural log of individual or per caput daily calorie intake (kcal), X is log of per-capita consumption expenditure (Proxy for income), W is women share of household gross income (range 0–1), 22 I is the vector of individual level variables (age, sex, education, etc.), H is the vector of household level variables (household composition variables), P is the vector of prices per kilogram in Naira of 7 food items that make up 85% of the sources of calorie intake in the study area, e is the disturbance or error term. The two major parameters of interest in this study are a1 and a2. The former is the expenditure/income elasticity of per caput calorie intake while the later represent the marginal response of calorie intake to increases in the share of household income that is under the control of women. Based on evidence from empirical literature, we expect a positive a1 with values ranging between 0.0 and 0.4. A positive a2 would imply that a redistribution household income from men to women would result in increased per caput daily calorie intake in the household. A negative a2 would imply that the redistribution of income from men to women (conditional on household income) would reduce per caput daily calorie intake and would suggest that more income in the hands of women relative to men may not be a good strategy for increasing per caput daily calorie intake in the household.

22 The women share of income is calculated as income earned by women in the household divided by the sum of incomes earned by men and women in the household.

520

A.B. Aromolaran / Food Policy 29 (2004) 507–530

Econometric estimation issues Estimating Eq. (5.1) by the ordinary least square (OLS) regression procedure would imply an assumption that all the right hand side variables in the model are truly exogenous. 23 This may not be the case in this model with income and expenditure variables on the right hand side. WomenÕs share of income is endogenous because the income variable used in this model is basically labor income, making its value an outcome of labor supply choices. There is also the problem of simultaneity which is due to the possibility of reverse causality between calorie intake and income. This relationship is the central theme of the efficiency wage hypothesis (see Stiglitz, 1976). It proposes that higher income earning opportunities are open to those who are better nourished. However, Subramanian and Deaton (1996) ignored the possibility of reverse causation between calorie intake and income. They showed that in rural Maharashtra, south western India, it is implausible that income is constrained by nutrition because the cost of calories necessary for a dayÕs activities is less than 5% of the daily wage. In the study area, the cost of calories necessary to satisfy the FAO minimum daily standard of 2500 kcal is as much as 91 percent of the per-capita expenditure per day. If we follow the reasoning of Subramanian and Deaton (1996), it is plausible to think that income may be substantially constrained by nutrition in rural south western Nigeria. 24 Furthermore, per-capita expenditure is endogenous since the decision to expend is taken side by side with the decision to consume calories or purchase food items. This implies a correlation between X and e, and introduces omitted variable bias. These potential problems of endogeneity of income, simultaneity between calorie intake and income, and omitted variable bias would result in an upward bias of OLS estimate of a1 and a2 due to the violation of a number of underlying assumptions of OLS estimator. Another potential source of bias in the OLS estimates of per caput expenditure and women income share coefficients in this investigation is the classical measurement error bias which is also referred to as attenuation bias. The presence of measurement error in the information collected on income and expenditure would result in a downward bias of the OLS estimate. Given the usual difficulty in getting accurate information on income and expenditure of individuals and households in developing countries, measurement error bias may be a very important source of bias in this study. Attempts made in this study to reduce the effect of these potential sources of bias on the estimate of per capita expenditure and womenÕs income share elasticity can be categorized into two. First are measures taken at the data collection stage, and second are measures taken at the estimation stage. Figures on food intake, income and expenditure used in the analysis are averages of data collected through multiple visits 23 That is if l, m and e are the error terms in per capita expenditure, women income share and per caput calorie intake equations, the following conditions must hold for per capita expenditure and women share of income to be truly exogenous in the calorie intake model: E(a˚l) = 0; E(a˚m) = 0; where E implies mathematical expectation. 24 The cost of 2500 kcal in the study area is 70.00 Naira and the average daily per capita expenditure is 77.00 Naira (This is about $1.00 at the time of data collection in 1999).

A.B. Aromolaran / Food Policy 29 (2004) 507–530

521

over a period of 6 months. This is expected to reduce classical measurement error since we use the mean of the distribution of quantities/values for each respondent over a period of time. Second, data on calorie intake was obtained directly from food quantity and not indirectly from food expenditure data. This reduces the potential upward bias from non-classical measurement errors. 25 Third, values of time varying and time invariant assets of the household and individuals were collected at the beginning and end of the survey. The average values were used in the analysis. The instrumental variable two stage least square (2SLS) estimation procedure is used to address the problems relating to measurement error bias, simultaneity and endogeneity of income/expenditure. With this procedure, expression (6.1) is estimated in place of expression (5.1). C ¼ b0 þ b1 X  þ b2 W  þ b3 I þ b4 H þ b5 P þ e ;

ð6:1Þ

where X* and W* are predicted values of the two endogenous explanatory variables and e* is an error term that is uncorrelated with X* and W*. To obtain X*W* and e*, expressions (6.2) and (6.3), called first stage regression equations are estimated. X ¼ d0 þ d1 I þ d2 H þ d3 P þ d4 Z þ l;

ð6:2Þ

W s ¼ U0 þ U1 I þ U2 H þ U3 P þ U4 Z þ m;

ð6:3Þ

where I, H, and P are as defined in expression (5.1) and Z is the 1 · K vector of identifying instruments. Both l and m are assumed to be well behaved (i.e. independently and identically distributed, i.i.d.) with mean zero and constant variance, and X* = X
l, while W* = W
m. For the set of instruments used in this study to be valid, d4 and U4 would have to be identified. That is the following conditions must hold for each equation. EðZX Þ 6¼ 0;

EðZW Þ 6¼ 0;

EðCZ=X ; W Þ ¼ 0;

ð6:4Þ ð6:5Þ

where E represents mathematical expectations. Thus for an instrument to be valid it must be strongly correlated with per capita expenditure, X, or womenÕs income share, W (expression (6.4)) and should not be significantly correlated with per caput daily calorie intake when conditioned on per capital expenditure and women income share (expression (6.5)). That is, after controlling for X and W, the effect of the Z-vector of instruments on per caput calorie intake should be close to zero.

25 This non-classical measurement error is created if calorie intake quantity is computed from food expenditure data. In such cases, any measurement error in the expenditure on the right hand side is carried on to the calorie intake value on the left hand side. As a result the estimated coefficient of per-capita expenditure would potentially have an upward bias.

522

A.B. Aromolaran / Food Policy 29 (2004) 507–530

Four such variables were found and used as instruments for per capita expenditure and women income share in this study. 26 These are total value of farm business asset, 27 ratio of the years of schooling of senior wife to the male household head, women share of household farmland value and the share of household business asset controlled by women. 28 Theoretically, the value of farm business assets owned by the household is expected to be a fair measure of household income earning capacity. 29 Holding the value of household farm assets constant, it is expected that increases in women share of farm land asset or women share of household business asset would increase percapita consumption expenditure. That is we expect per capita consumption expenditure to increase as women in the household become wealthier relative to men in terms of farm and non-farm business asset ownership. WomenÕs income share is expected to be predicted fairly well by women share of farm land asset, women share of household business asset, and the ratio of womenÕs years of schooling to that of the household head. That is, as women get control over more farm and non-farm business assets relative to men in the household, we expect their share of household in26 The first task in the estimation was to find strong predictors for the endogenous variables on the right hand side of the specified calorie intake. Four significant predictors were selected after for both log per capita income and womenÕs share of income after running a series regressions. These are ratio of wifeÕs years of schooling to husbandÕs years of schooling, total farm business asset, women share of total value of household farm land, womenÕs share of business asset. Over-Identification test for the calorie intake model was then carried out on these 4 variables. This test was carried out to select those predictors of per capita expenditure and womenÕs share of income that could serve as true identifiers of these right hand side endogenous variables in the calorie intake equation. This was done by regressing log per caput food calorie intake in kilocalories on the 4 variables earlier listed as identifiers. A variable is selected as a potential identifier if the estimated coefficient is not significant at 5% level of significance. This resulted in all 4 variables showing up as true model identifiers for the calorie intake model. 27 The total value of farm business assets is the sum of the value of all farm assets owned by the household. This includes land, farm implements, farm buildings, farm machinery. This estimate does not include farm supplies like fertilizer, seeds chemicals etc. The womenÕs share of farmland value is measured as the ratio of value of farm land under the full control of women to total household farmland value. Total household farm land was multiplied by the current market price of farm land in the locality to obtain total value of household farmland. 28 Hopkins et al. (1994) instrumented annual male income with the following variables: land area operated by males, the value of male livestock assets, a dummy variable for male literacy, the number of wives, the dependency ratio, a variable for single or multiple conjugal units within the household, the age of the household head, quadratics of several variables and interaction terms. The annual female income was instrumented using the land area operated by female, the value of female livestock assets, a dummy variable for female literacy, the number of infants in the household, the number of caretakers in the household (girls between the ages of ten and fifteen), a market village dummy, an ethnicity dummy, quadratics of several variables, and interaction terms. The instruments that were used to predict the log of per capita expenditure in Hoddinott and Haddad (1995) and Haddad and Hoddinott (1994) include: the amount of land owned by the household, the logarithm of the per capital value of holdings of consumer durable, the number of rooms per capita in the dwelling, the per capita floor area of the dwelling, and dummy variables equaling one; if the walls of the dwelling are cement, stone or brick; if the floor of the dwelling is cement, stone or brick; if the dwelling is owned by the household and is located in an urban cluster; if the household grows any major internationally traded cash crop. 29 The data shows that farm business asset constitutes 77 percent of business assets owned by the average household.

A.B. Aromolaran / Food Policy 29 (2004) 507–530

523

come to rise. Also, a decline in the schooling gap between household men and women is expected to lead to an increase in the share of household income under the control of women. An important condition for the identification of d4 and U4 in this study is that the amount of farm business assets owned by the households, women share of farm land asset, the womenÕs share of household business asset, and women/men ratio of years of schooling would only affect calorie intake indirectly through their (direct) effect on per capita expenditure and womenÕs share of income. If this additional condition is satisfied, then these four variables would be acceptable as true identifying instruments for per capita income and womenÕs share of income in the calorie intake model. An over-identification test was carried out to find out whether these four variables are true identifiers, To do this, I estimate expression (6.6) C ¼ a0 þ a1 X þ a2 W þ a3 I þ a4 H þ a5 P þ a6 Z þ e;

ð6:6Þ

where all variables are as defined earlier and Z contains four elements. For any given element of the Z-vector to be a true identifier, the estimate of the corresponding element of a6-vector must be statistically insignificant. That is, after controlling for the two right hand side choice variables, the effect of a true identifying instrumental variable on calorie intake should be statistically not different from zero. In other words, it must be the case that the instrumental variable has no effect on per-caput calorie intake that is independent of its effect on per capita expenditure and women share of income.

Results First stage regression results and over-identification test From the results of the first stage regressions shown in Table 2, 30 we observe that total value of farm business assets and womenÕs share of household farm asset value are significant and positive predictors of per capita expenditure. That is, the wealthier the household is in terms of farm business asset, the higher the per capita consumption expenditure. Also, after controlling for the value of household farm assets, increases in womenÕs share of farm land asset would increase per-capita consumption expenditure. Women share of household business assets turns out to be the main identifying variable for women share of income. That is, after controlling for other model variables, the higher the proportion of household business assets that belong to women, the higher the share of the household income flow that is under the control of women. 30

Columns 4 and 5 in Table 2 show the results of the F-test for joint significance of identifying variables, while columns 6 and 7 present the results of first stage regressions which includes all other explanatory variables in the original model specification in expression (5.1).

524

A.B. Aromolaran / Food Policy 29 (2004) 507–530

The third column of Table 2 presents the result of the over-identification test. It shows that all the four identifying instruments are not significant predictors of per caput calorie intake once we control for per capita expenditure and womenÕs income share in the calorie intake equation. This implies that all four instruments are true identifiers of independent variations in per-capita expenditure and womenÕs income share in the calorie-income model. That is, they exert indirect effect on calorie intake through their direct effect on household income and womenÕs share of income. The table also shows a fairly reasonable predictive power for the full set of instruments. The R2 for the per capita expenditure and womenÕs share of income equations are 71.0 and 50.0 percent, respectively. The strength of the joint predictive power of the set of identifying instruments alone is demonstrated by the fairly reasonable R2 values of 11.0 percent for per capita expenditure and 25.0 percent for womenÕs share of income. F-test results in the last two columns of Table 2 show p-values of 0.00 which implies a rejection of the hypothesis that the four identifying instruments are not jointly significant in both the womenÕs income share and per capita expenditure equations.

Income and women’s share of income elasticity of calorie intake We observe from Table 4 that the ordinary least square (OLS) estimates of per capita expenditure and womenÕs share of income coefficients are significant at 5 percent level 31 and smaller than the two-stage least square (2SLS) estimates of the same coefficients. This may be an indication that classical measurement error bias is a more important source of bias in this investigation relative to other potential sources of bias discussed earlier. 32 Table 4 also shows that the OLS and 2SLS estimates of women share of income are both negative, while both estimates for per capita expenditure are positive. Thus we find consistency in the sign attached to the estimated coefficient of per capita

31 The standard errors of the estimated coefficients were corrected for clustering within households by using the robust command in STATA software package. The reason is that the simple estimates of standard errors become incorrect when we have multiple observations, which are not independent within a data. In the data set used for this analysis most of the income and expenditure variables, as well as household head and senior wife characteristics fall into this category of variables. All individuals that belong to the same household have the same values for these variables. This is referred to as clustering within households. 32 The Hausman test for exogeneity of endogenous variables was carried out for the 2 versions of each model. The test was aimed at finding out whether there is sufficient difference between the coefficients of the instrumental variable two stage least square regression (2SLS) and the standard ordinary least square regression (OLS), to indicate that OLS would be inconsistent for our model. The results showed that we do not have sufficient statistical grounds to reject the null hypothesis that ‘‘differences between coefficients of the OLS and the instrumental variables regression estimates are not systematic’’. The implication is that OLS is also a consistent coefficient estimator for the estimated equations. However the estimated coefficients of the endogenous variables (women share of household income, per capita income, and per capita expenditure were consistently larger in the 2SLS compared with the direct OLS).

A.B. Aromolaran / Food Policy 29 (2004) 507–530

525

Table 4 Results of regression of log per caput daily calorie intake on per capita expenditure and womenÕs share of household income Estimation method

Individual calorie/day (OLS)

Individual calorie/day (IV-2SLS)

Log per pita consumption expenditure Women share of household income (ratio) Age of individual Quadratics of individualÕs age Sex of individual No. of aged females in household (60+) No. of aged males in household No. of adult females in household (19–59) No. of adult males in households (19–59) No. of adolescent females in household (11–18 years) No. of adolescent males in household (11–18 Years) Household head years schooling Senior wife years schooling Household head major occupation (non-farming = 1, farming = 0) Senior wife major occupation (non-farming = 1, farming = 0) Yam price Cassava flour price Yam flour price Cowpeas price Rice price Maize price Gari price Palm oil price Beef price Fish price Constant No. of observations

0.0408 (3.45)
0.0500 (2.22) 0.0276 (21.73)
0.00026 (14.22) 0.0383 (3.22)
0.0339 (2.70)
0.663 (5.48)
0.0497 (6.99)
0.0156 (2.87) 0.0053 (0.84) 0.0218 (3.22)
0.0002 (0.01)
0.0010 (0.59)
0.0006 (1.04)

0.194 (2.48)
0.1301 (2.57) 0.0260 (18.14)
0.000242 (12.37) 0.0413 (3.30)
0.0322 (2.25)
0.0732 (5.88)
0.0429 (5.06)
0.0133 (2.25) 0.00780 (1.21) 0.0302 (3.77)
0.00286 (1.44)
0.00132 (0.72)
0.00132 (1.07)


0.0012 (1.47)


0.0027 (1.50)


0.0487 (13.65) 0.0098 (3.051)
0.0049 (3.45)
0.0148 (3.52) 0.0540 (7.95)
0.0143 (4.88)
0.0601 (24.11) 0.0138 (11.65)
0.0031 (5.89) 0.0086 (3.89) 5.882 (23.35) 2360


0.0642 (7.41) 0.0187 (3.17)
0.00643 (3.67)
0.0459 (2.78) 0.00916 (4.78)
0.0179 (5.39)
0.0503 (8.72) 0.0187 (6.40)
0.00656 (3.43) 0.00111 (4.03) 4.573 (6.90) 2360

Figures in parenthesis are t-values.

expenditure and womenÕs income variables irrespective of type of estimator (i.e. OLS or 2SLS). 33 The first column of Table 5 reports the income/expenditure and womenÕs income share elasticity of calorie intake. The response of per caput calorie intake to changes in per-capita consumption expenditure is estimated as 0.194. That is an increase of 100% in per-capita expenditure will only increase calorie intake by 33 We also observe from Table 4 that individual daily calorie intake is dependent on age of individual with a quadratic form of relationship. Older individuals have higher intakes. Calorie intake increases by 2.6 percent from age 2 to 3 and increases at a decreasing rate on till the age of 54.5 years when a decline sets in. We also observe that years of schooling of both household male head and senior wife were not significant factors as far as variations in individual calorie intake in the study area are concerned. Finally, males generally had about 4 percent more calorie intakes than females.

526

A.B. Aromolaran / Food Policy 29 (2004) 507–530

Table 5 Estimated income elasticity coefficients for calorie intake quantity, calorie price, market purchased share of food expenditure, and food expenditure Dependent variable

Log per caput calorie intake Log calorie price Market purchased share of food expenditure Log per caput food expenditure

Coefficient of per-capita expenditure

Coefficient of womenÕs share of income

Estimated coefficient

Income/ expenditure elasticity

Estimated coefficient

Elasticity

0.194 (2.48) 0.1406 (3.26) 0.0269 (0.61)

0.194 0.1406 0.00368


0.13 (
2.53)
0.005 (
0.17) 0.3890 (10.24)


0.0403
0.00155 0.210

0.3143 (4.61)

0.3143


0.13 (
2.58)


0.0403

The complete set of controls in Table 4 is included in each equation. Figures in parenthesis are t-values.

19.4%. Theoretically, it is expected that income 34 increases would enable individuals in low income households to increase their food calorie intake. This in turn is expected to improve nutrition status, health and productivity of household members. The observed low calorie intake elasticity suggests that calorie intake does not get a substantial share of marginal increases in household income. This result is in line with the conclusion of Boius et al. (1992) that most recent studies have reported calorie-income elasticity which are less than 0.2 in contrast to conventional wisdom that calorie-income elasticity for low income populations in the developing world ranges between 0.4 and 0.8. Thus, increasing household disposable income may not be a very effective strategy for bringing about increased food energy intake among low income households in south western Nigeria. The instrumental variable (IV) estimate of the elasticity of womenÕs share of income in the calorie equation as shown in the top row of Table 5 is
0.04. That is, a doubling of the share of household income earned by women from the current average of 31 percent to 62 percent will result in a 4 percent decline in per caput calorie intake of the household from the current average of 2204 to 2116 kcal. Even though the negative effect is small, the result is a clear rejection of the hypothesis that per caput calorie intake responds positively to increasing womenÕs share of household income and suggests that income redistribution from men to women would not increase per caput food energy intake in the population under study. This result does not support the general thinking that intra-household resource reallocation from men to women would increase food consumption and food energy intake. Rather it would imply that food calorie intake by household members is enhanced with more income in the hands of men relative to women. 34 Income is proxied here by consumption expenditure. Consumption expenditure is widely used in place of income because of a number of reasons. One is that it is subject to less errors of measurement and second is that it is a better approximation of permanent income, if we assume that households smooth consumption over their life time.

A.B. Aromolaran / Food Policy 29 (2004) 507–530

527

However, it can be argued that the observed non-positive response of per caput calorie intake to changes in womenÕs share of household income may be an evidence of female preference for more expensive foods with less energy content. To check this, I estimate the effect of womenÕs share of income on food calorie price. 35 The elasticity estimate of
0.0055 is presented in Table 5 and shows that increases in womenÕs share of income has no effect on the unit cost of calorie consumed, suggesting that women do not seem to reallocate towards more expensive calorie sources as their income share increase. Another plausible explanation for the negative sign of the womenÕs income share coefficient is the issue of transaction costs. Since the estimate of the effect of womenÕs share of income conditions on household income, increasing womenÕs share of income would, by definition, be associated with decreasing share for men. Even if household resources are not reallocated to more expensive foods with less calorie content, calorie intake could still decline if it is more expensive to get a unit of calorie with womenÕs income than with menÕs income. Consequently, the household will expend more to consume the same quantity and quality of food if income is redistributed from men to women. The notion of transaction cost here refers to the difference between the price of a food bundle obtained from the husbandÕs largely agricultural income and the cost of the same bundle of food obtained from the womanÕs largely non-farm business income. This situation can arise if food consumption from menÕs income is mainly from his farm and food consumption from female income is mainly from the market. 36 To check if it is the case in the study area that food consumption from womenÕs income comes more from the market relative to food consumption from menÕs income, I estimate an equation whose dependent variable is the share of purchased food in total food expenditure and the right hand side variables are the same as the calorie intake equation in expression (6.1). A positive and significant coefficient of womenÕs share of income in this equation would suggest that households purchase more of their food from the market as womenÕs income share increases. This however does not guarantee that the same quality and quantity of food is obtained at higher cost in the market. To check this, I estimate an equation with log per caput food expenditure as the dependent variable and the same right hand side variables as the three previous equations. A significant and positive womenÕs share of income coefficient would suggest that intra-household income redistribution to women would make the households spend more per capita on food. A positive sign for womenÕs share of income in both equations would suggest that the negative sign on the calorie intake equation coefficient may be due to transaction 35 Food calorie price here is a proxy for how expensive the calorie being taken is. A higher value of food calorie price/cost implies higher quality calorie source. It is computed as the ratio of per caput food expenditure to per caput food calorie intake. i.e. calorie price (Naira/kcal) = per caput food expenditure (Naira)/per caput calorie intake (kcal) A significant and positive coefficient of womenÕs share of income would imply that women actually reallocate towards more expensive calorie sources which may have less calorie content. 36 It is assumed here that the same quality and quantity of food purchased in the market would cost more than if obtained from own farm due to positive marketing margins/markups.

528

A.B. Aromolaran / Food Policy 29 (2004) 507–530

costs. Otherwise, we would not have enough evidence to infer transaction cost as the cause of the negative sign of the womenÕs income share coefficient in the calorie intake equation. Columns 4 and 5 in Table 5 show the elasticity estimates from these two equations for womenÕs share of income. The elasticity estimate for share of food purchased from market is positive while that for the log of per caput food expenditure is negative, implying that transaction cost would not be a an appropriate explanation for the negative sign on the womenÕs share of income coefficient in the calorie intake equation. 37 Thus, the negative sign on the womenÕs share of income coefficient is more likely to be an indication that food calorie intake would respond negatively to a reallocation of household income from men to women, rather than a consequence of a reallocation of income towards more expensive and lower calorie foods or an evidence of positive transaction cost for substituting female for male income imperfectly in household food consumption. Thus, more income in the hands of women relative to men would not be the best strategy for increasing calorie intake of household members in the study area. Finally, if we compare the estimated elasticity coefficients in terms of absolute value, we find that the effect of per-capita expenditure on per caput food calorie intake is about four times the size of the effect of womenÕs share of income. These results suggest that female-income targeting may not be as effective in raising per caput calorie intake levels of household as male-income targeting.

Conclusion This study investigates how per caput calorie intake in low income households of south western Nigeria respond to changes in total household income and womenÕs share of household income. I utilize primary data collected with multiple visits over a period of 6 months from 2573 individuals in 480 randomly selected households. The study addresses three major questions. First, ‘‘does a change in womenÕs income share significantly affect per caput calorie intake?’’ If it does, then what is the direction of the effect, and how does the magnitude of the effect compare with that of total household income. The findings of the study support the following conclusions. First, increases in womenÕs share of household income are likely to result in marginal declines in food calorie intake by individual household members. This suggests that wealth redistribution from men to women would not increase per caput food energy intake in low income households in rural south western Nigeria, as hypothesized in this study. 37 Elasticity for womenÕs share of income in the calorie intake, calorie price and food expenditure equations are calculated as the product of the estimated coefficient and the mean value of womenÕs share of income in the sample (0.31). Income Elasticity for the share of food purchased from market is calculated as the ratio of the estimated coefficient to the mean value of log per capita expenditure, the womenÕs share of income elasticity is computed as the product of the estimated coefficient and the mean value of womenÕs share of income, divided by the mean value of market purchased share of food expenditure.

A.B. Aromolaran / Food Policy 29 (2004) 507–530

529

Second, I find that increases in total income of household has about four times as much effect on per caput calorie intake as increases in womenÕs share of income and this implies that targeting household income may be a more effective strategy for increasing calorie intake levels than targeting women income share.

Acknowledgement I thank the African Economic Research Consortium, Nairobi Kenya and the Rockefeller Foundation Grant for Postdoctoral Research on the Economics of the Family in Low Income countries, Yale University, for providing the financial support for this study. The study benefited from comments and suggestions by John Strauss, Germano Mwabu, Eric Thorbecke, David Sahn, T. Paul Schultz, Garth Frazer, Christopher Udry, Robert Evenson, participants at the bi-annual research workshops of the African Economic Research Consortium, participants at the Development Lunch Workshop of the Economic Growth Center, Yale University and two anonymous referees. Femi Enilolobo, Michael Adewale and Ayo Ibrahim provided excellent research assistance. I am responsible for any errors that remain.

References Alderman, H., 1986. The Effect of Food Price and Income Changes on the Acquisition of Food by Lowincome Households. International Food Policy Research Institute, Washington, DC. Aromolaran, A.B., 2001. Research Project on Household Food Security, Poverty Alleviation and Women Focused Development Policies in Nigeria: Final Research Report, African Economic Research Consortium (AERC), Nairobi Kenya, July 2001. Behrman, J.R., Deolalikar, A.B., 1987. Will developing country nutrition improve with income? A case study for rural South India. Journal of Political Economy 95 (3), 108–138. Behrman, J.R., Deolalikar, A.B., 1990. The intra-household demand for nutrients in rural south india: individual estimates, fixed effects, and permanent income. Journal of Human Resources 25 (4), 665– 696. Boius, H., 1994. The effect of income on demand for food in poor countries: are our food consumption databases giving us reliable estimates?. Journal of Development Economics 44, 196–226. Boius, H., Haddad, L.J., 1992. Are estimates of calorie-income elasticities too high?: a recalibration of the plausible range. Journal of Development Economics 39 (2), 333–364. Boius, H.E, Haddad, L.J., Kennedy, E., 1992. Does it matter how we survey demand for food? Evidence from Kenya and Philippines. Food Policy 17 (5). Grimard, F., 1996. Does the poorÕs consumption of calorie respond to changes in income?: evidence from Pakistan. The Pakistan Development Review 35 (3), 257–283. Haddad, L., Hoddinott, J., 1994. WomenÕs income and boy–girl anthropometric status in the cote dÕIvoire. World Development 22 (4), 543–553. Hoddinott, J., Haddad, L., 1995. Does female income share influence household expenditures? evidence from cote dÕIvoire. Oxford Bulletin of Economics and Statistics 57 (1). Hopkins, J., Levin, C., Haddad, L., 1994. WomenÕs income and household expenditure patterns: gender or flow? Evidence from Niger. American Journal of Agricultural Economic 76 (5), 1220–1225. Maxwell, S., Frankenberger, T., 1992. Household Food Security: Concepts, Indicators, Measurements. A Technical Review. UNICEF and IFAD, New York and Rome.

530

A.B. Aromolaran / Food Policy 29 (2004) 507–530

Stiglitz, J.E., 1976. The efficiency wage hypothesis, surplus labor, and distribution of income in LDCs. Oxford Economic Papers 28 (2), 185–207. Strauss, J., 1986. Does better nutrition raise productivity?. Journal of Political Economy 94 (2), 297–320. Subramanian, S., Deaton, A., 1996. The demand for food and calories. Journal of Political Economy 104 (1), 133–162. Thomas, D., 1990. Intra-household resource allocation: an inferential approach. Journal of Human Resources 25 (4), 635–664. Thomas, D., 1992. The distribution of income and expenditure within the household. In: Paper presented at the IFPRI-World Bank Conference on Intra-Household Resource Allocation: Policy Issues and Research Methods. February. International Food Policy Research Institute, Washington, DC, Processed. Thomas, D., 1993. Incomes, Expenditure and Health Outcomes: Evidence on Intra-household Resource Allocation. Mimeo IFPRI, Washington.