Paper Review 02: “Does female income share influence household expenditures? Evidence from C?te d'Ivoire”
John Hoddinott and Lawrence Haddad (1995)
The common preference model of household resource allocation postulates that all income is pooled and a dictator (often female or male head) determines the allocation. This is based on the assumption that household members collectively maximize some household level welfare function and income is spent as such the marginal rate of substitution between a pair is symmetric to any other pair. This is called the unitary model. Many influential studies have rejected its existence, one of which is this paper written by John Hoddinott and Lawrence Haddad.
The title of the paper draws immediate attention to the rejection of income pooling in an African setting of C?te d’Ivore. The findings move into the literature of collective models where each member has their own utility function and decisions are a result of bargaining powers of members.
Due to the heterogeneity of preferences nested within households it is important that policy implications such as public work schemes and transfer programs are targeted strategically, in particular towards women. Additionally, policies must target the issues within the households than generalizing to households only.
In Zambia, diets would have improved due to the complementary benefits of consuming maize and beans. Hence, women were asked to intercrop beans with maize, the latter being a male-controlled crop. Women were discouraged as they would lose ownership of beans by cropping on lands normally allocated to beans. By contrast, Dankelman and Davidson (1988) found that the project to enhance soybean production supplementing protein diet was successful as it allowed for the collective nature of households. Women in Togo were targeted through exchange visits and workshops arranged in their homes. Soybeans were not seen as cash crop changing women’s status in households rather legumes for sauces. The crops remained in the control of women and in some cases were even allocated small plots of land for cultivation.
Research is mounting that household expenditure patterns are influenced by the portion of income accruing to women, while holding all income constant. These findings tend to hold for both earned and unearned income of women. These influences include nutrient-intakes, fertility, child survival and child’s height for weight. It was seen that rise in woman’s unearned income increased child survival by twenty times in comparison to a man’s unearned income in Brazil (Thomas, 1990).
Studies have also corroborated that relative to women, men are seen to spend a larger portion of their income on alcohol, tobacco, entertainment and other consumer goods. Women, however, are likely to spend on food, healthcare, clothing, education and other general household items. The spending pattern of the latter is consistent with specialized roles of wives/mothers and ideally is socially desirable.
The reconciling of different preferences under collective models relate to cooperative and non-cooperative solutions. Cooperative solutions result in decisions where one cannot be made better off without worsening the other resembling a Nash-bargaining game. Non-cooperative solutions exist where members do not engage in binding or enforceable contracts with each other. A husband’s budget is separate from his wife’s and they respond to changes in each other’s allocation of budget solely according to their needs. The spouse makes decisions according to his/her own sphere and will change actions voluntarily. This sole phenomenon forms the basis of this paper’s research.
Hoddinott and Haddad addressed this very non-cooperative bargaining model of household expenditures with the then unpublished work by Ulph (1988). Empirical investigation complemented by consistency checks to assess robustness makes this paper highly referenced in literature when discussing the policy implications in gender-specific control of income and its spending patterns in households.
If we were to review the objectives of the authors where firstly, use of the Ulph model that rejects income pooling, secondly, use of the Working-Leser Expenditure (WLE) function that helps to create the relationship to understand the spending patterns, thirdly, using Two Stage Least Squares (2SLS) estimation to account for endogeneity in the variables log of per capita expenditures and women’s shares of cash income and fourthly, including alternative functional form specifications, reduced form estimates and results from single sex adult households that work to reduce inconsistency.
The paper begins with a general introduction of gender-specific incomes rejecting the unitary model and questioning the income pooling assumption. The authors, thus indicate the need for targeting, particularly towards women in policy-making. They further justify their research by referencing several studies that found similar results but econometric tests remained limited in an African setting.
They account for this limitation which arises when dealing with household surveys in developing countries. As such the Cote d’Ivore Living Standards Survey (CILSS) does not contain information on gender-specific incomes. A star aspect of the paper is the use of Ulph model. While these surveys provide aggregated income of households the Ulph model assumes income is not pooled within the household. Rather the share of household expenditures allocated to particular goods is a function of the intra-household distribution of income.
It is worth noting some of the features of the model such as the use of a Cobb-Douglas utility function for male and female showing that if the share of income is skewed towards a member, the member’s choice of expenditure will dominate and sustained as Nash-equilibrium. Moreover, as income moves from zero to one of a member the set of choices on expenditure for that member will also increase. In the disagreement of preferences of a member, the theory argues that both the members make certain ‘strategic purchases’ in order to pre-commit the household to a minimum expenditure of his or her preferences. Such inequalities are more visible among rural households in developing countries where suppose, individuals are able to enforce their preferences often through their ‘perceived contribution’ and gender-specified roles. Example, household-head may often be referred to a male who is termed as the bread-earner while females are seen to take up roles of care for the household members, allowing the former more spending power over his set of choices.
There exists a unique expenditure decision for both males and females with which it is possible to derive a demand/reaction function. Extending on the WLE function by adding Ulph’s ‘share of women’s income is regressed on share of expenditures on goods. This variable is further identified in the model as the portion of income accrued to wives[2] of the male head. Other regressors include logarithm of household characteristics, proportions of different demographic groups and dummy variables for household location. The WLE function serves its purpose in estimating elasticities of the goods relative to the independent variables.
Two variables deserve attention: and budget share of goods. Budget share of goods include food, fuel, children’s clothing, other cash expenditures, meals consumed outside the home, adult clothing, alcohol, cigarettes, jewellery and entertainment. These clearly indicate the gender-specific preference among households. Before gender segregating income, total household cash income includes proceeding from sales of crops, net sales from livestock, wage employment, entrepreneurial activities, remittances and other sources.
Hoddinott and Haddad are able to showcase their sophistication in the calculation of through the use cluster-based households. They used ethnographic references of earlier studies particularly from the late 80s to match gender control of cash from crops owned by members. They were proactive in mentioning the shortfalls of such an approach. It may happen that both grow the same crop but separately, or cash accrues to the husband but he pays some of it to his wife for her assistance in the fields or when there is ethnic-mingling it is difficult to identify crop allocations.
The authors continue with assumptions of gender-specific control of income from a broad range of agriculture crops segregated further across clusters. An example, In Bete clusters: male control half of the maize and rice crops and vice versa for females. With this approach, 20.1% (2.3% comes from wage employment, 9.1% from own business activities and 8.8% from agriculture) of total household cash income accrues to spouses of the male head.
The regression was estimated using 2SLS on a set of multiple instruments.[3] It was seen that share of cash income significantly affected some of the key budget shares of goods. Notably both significantly positive on food but negative effect on children’s clothing, meals eaten out, adult clothing, alcohol and cigarettes. This adds to the literature on household expenditures patterns in Sub-Saharan Africa.
The authors showcased the magnitude of these effects under four scenarios: no cash earned, cash earned equalling the mean, cash earned is doubled and all cash accrues to wives. Effects were the largest when incomes were doubled with spending on alcohol and cigarettes falling by 25.0% and 15.3%, respectively.
The authors accounted for non-linearity in Engel Curves by including women’s education and a quadratic term for log of per capita expenditures. The results were robust and similar to those estimated firstly. However, tobit[4] estimates for women share of income on cigarettes was not statistically significant.
These results are further consistent with perceived gender ideologies where women have traits of “good mothering” relating to spending on food while men (stronger among male-headed households) have a right to spending money solely for own satisfaction. Agricultural contribution such as working on farms by men provide direct evidences of their contribution while it must also be acknowledged that women also engage in energy-intensive work within the household. Examples include, engaging in household chores during pregnancy, walking in search of fresh water, serving as domestic helpers in urban households. These may often be unaccounted for.
One of the nicest feature of this paper is that the authors provide limitations immediately after proposing a judgement to justify their claims. To an extent they have not able to are now discussed.
While the authors were quick to conclude targeting of women for policy interventions, many studies would have used this finding as a strategy to reducing poverty. Appleton (1991) finds for Cote D’Ivoire that predicted probability of illness is more in boys than girls. And while we can assume that mothers are aware of this, with rises in their income they may provide more nutrients to boys to safeguard their health.
This puts Hoddinott and Haddad’s paper up for improvement. While increasing will increase spending on food it could be dissected to see whether there is inequality among children for reasons outside altruism of parents.
The effect of on children’s clothing is justified by the claim that clothing items such as shirts and shoes required for school is the responsibility of the father. However, it contrasts with the gender “realms” and “ideologies” established initially. Phipps and Burton (1998) do similar calculations to find positive and significant expenditure for children’s clothing when controlling for mother’s income. Similarly, Lancaster, Maitra and Ray (2005) use 3SLS to account for endogeneity in IH balance of power, where again controlling for women’s power increases spending on children’s education.[5]
It may happen that as rises mothers may begin to look for cheaper alternatives to increase spending on essentials. It maybe that mothers are more aware of market prices and are able to bargain cheaply for such items. Notably, parameter for fabric remained significant and positive which may further mean that mothers are buying the fabric to be sewn into uniforms/clothes at home which is again cheaper than buying from markets. In fact, can it mean that as wives’ income rises they may want to pamper themselves by purchasing fabric?
The strength of the IV will ensure reduction of endogeneity in a model. The instruments used in this paper include all demographic variables, location dummies and those specific to the per capita expenditures and women’s share of cash income. Of particular interest is the instrument wives’ shares of asset holdings. It can be argued that control and accumulation of women’s assets maybe endogenous to marriage. Validity of this as an indicator of bargaining power maybe conditional upon the receipt of such assets before marriage. When were replaced by its instruments, RFE[6] showed consistent relationship between food and alcohol but not for tobacco, meals eaten outside, children and adult clothing.
The CILSS provides information on both healthcare and fertility which are important variables of consideration in literature. In the study of Ghana, medical expenses were negatively related to women’s bargaining power in both rural and urban areas. These medical expenses comprised of curative care which can result in healthier households (Doss, 2006).
A closer look at the survey shows ‘food’ comprises of 18[7] items. These were collected from markets in each cluster where the household survey is located. This can allow us to identify indirectly the necessary food items that impact health positively. While higher income resulted in better health among children and adult, the magnitude of the effects was higher for women (Thomas, Lavy and Strauss,1996).
The study employs a restricted model of single-gender households[8]. Results were consistent with the hypothesis with an exception of increased spending on jewellery in an all-female adult household. The spending on children and adult clothing was not however supported.
A closer look at the CILSS allowed for criticism in the variable ‘household size and composition’. The CILSS shows a decline in household size from 8.31 to 6.32 between 1980 to 1987. This requires attention as calculations of per capita expenditures take variations of household size. Similarly, the proportion of males aged 15-24 fell from 23.8% in 1985 to 18.4% in 1988 in Abidjan (Coulombe and Demery, 1993). These issues reduce representativeness and may also underestimate male’s expenditure patterns in the household.
The panel nature[9] of the CILSS produces selectivity bias in the sample. Suppose these males are likely to migrate hence are less likely to be retained in following years but they are considered variables of interest in cross-sectional analysis.
This paper serves its purpose in rejecting the unitary model by employing sophisticated econometric techniques however it lacks in certain areas of its significance, mainly by drawing conclusion about a relatively small portion of expenditure on alcohol and cigarettes. Intra-household inequality is as multidimensional as poverty and inequality is certainly not confined to this.
References are available, upon request.
[2] Includes widows where the male head is deceased.
[3] Instruments included all demographic variables, the location dummy variables and instruments specific to per capita expenditures and women’s share of cash income.
[4] Households may report zero consumption for certain goods. Zero shares are accounted by an unobservable latent variable. It might be that one person may not purchase it as frequently as others. Hence, the insignificant results.
[5] Spending on education may also include spending on children’s clothing, such as school uniform.
[6] Reduced Form Estimates
[7] 1. Beef (with bone) 10. Palm oil 2. Fresh fish 11. Local maize (grain) 3. Imported rice (other than Uncle Ben's) 12. Local millet (grain) 4. Local rice, husked 13. Cassava (raw) 5. Dry onion 14. Yams (early) 6. Lettuce 15. Plantains 7. Salt (large grain) 16. Oil palm nuts (seed) 8. Canned tomato paste (70 gms) 17. Shelled peanuts 9. Peanut butter 18. Eggs (each)
[8] This includes all female/male adult household, assuming children do not influence household preferences.
[9] The survey had 1600 households each year. Half of the households were revisited the following year. Thus producing four cross-sectional data sets and three overlapping panel of 800 households.