Gender, headship and intrahousehold resource allocation

Gender, headship and intrahousehold resource allocation

World Development, Vol. 22, No. 10, pp. 1535-1547, 1994 Copyright 0 1994 Elsevier Science Ltd Pergamon Printed in Great Britain. All rights reserve...

1MB Sizes 0 Downloads 80 Views

World Development, Vol. 22, No. 10, pp. 1535-1547, 1994

Copyright 0 1994 Elsevier Science Ltd


Printed in Great Britain. All rights reserved 0305-750x/94 $7.00 + 0.00


Gender, Headship and Intrahousehold Resource Allocation SUDHANSHU


University of the West Indies, Mona Kingston, Jamaica Summary. - Household head as reported in household surveys may not always identify the main economic provider and decision maker within the household. Research based on this definition of headship may lead to misleading conclusions on the link between gender and proverty. Treating female headed households as a homogenous group can also be inappropriate. Results from Jamaica show that it is not female headedness ner se that is associated with child welfare, but the absence of a potential female decision maker within the household.

1. INTRODUCTION The increasing incidence of female-headed households (FHHs) in developing countries has stimulated policy research on the link between the gender of household head, poverty, and the status of children. Since women in developing countries often face substantial constraints in their access to income-earning opportunities (Folbre, 1991), there is a concern that households with female heads are particularly vulnerable, and that children who reside in FHHs may suffer from lower educational and health outcomes which can limit their future choices as adults. The common practice, however, of classifying households by the gender of the survey-reported head may not accurately identify the main economic provider or decision maker of the household, and can result in misleading conclusions on the links among gender, poverty, and the welfare of children. The concept of headship was originally introduced in household surveys to avoid double counting, without any reference to economic support or authority. Potential biases in reported headship will be prevalent in developing countries where households consist of extended families and where social and cultural norms tend to identify the oldest male member of the household as head. There is evidence from Latin America and West Africa that a significant number of reported maleheaded households (MHHs) are actually maintained by women. Rosenhouse (1989) compares reported heads in a Peruvian household survey with “working heads” - the individual who contributed the most work effort on behalf of the household. Using this

alternative definition of head she finds the number of FHHs increases from 17 to 29% in the sample, and the prevalence of FHHs in the lowest income quintile increases from 20 to 34%. Similarly Lloyd and Brandon (1993) find that in Ghana, a woman was the primary source of economic support in a significant number of reported male-headed households. The simple male vs. female-headed household dichotomy used in the literature can also mask important differences within the two groups, with implications for intrahousehold resource allocation. When female headship occurs from several different types of processes, the behavior and economic status of these resultant FHHs can also be quite different. In a recent paper in World Development, Kennedy and Peters (1992) show that migrant FHHs in Malawi actually have higher per capita expenditures than any other type of household, while in both Kenya and Malawi, the poorest households are defucto FHHs. While it is true that most FHHs are distinguished by the absence of a regular adult male, international comparisons of FHHs must also consider regional mating and kinship patterns, as well as the role of women in the economy. Thus in West Africa, where women are an important part of shifting agriculture and where matrilineal descent patterns exist, and the Caribbean, where women play an active role in the


*I would like to thank the Planning Institute of Jamaica for kind permission to use the data, and Sue Horton and two anonymous referees for comments on an earlier draft. This paper is part of my PhD dissertation at the University of Toronto. Responsibility for all errors is mine. Final revision accepted: May 4, 1994.



economy and have considerable autonomy, a significant number of households report a female head despite the presence of a coresident male partner. Since headship is often used as an indicator of authority within the household, another research issue is whether unpartnered female-headed households behave differently from partnered ones. This study contributes to the research on FHHs in three ways. Using household data from Jamaica, the paper provides evidence on the accuracy of the reported headship variable in identifying the main economic provider (and perhaps decision maker) in the household. Following Rosenhouse (1989), households are classified by gender of reported and “working” head to see how these alternative definitions of headship influence the poverty profile. I then compare whether different classifications of headship significantly change the ranking of children’s outcomes between male and female-headed households. In particular, I test the hypothesis that when the household’s main source of economic support comes from a woman, the welfare of children increases. Finally, I investigate the validity of treating female-headed households as a homogenous group. Since a quarter of all FHHs in the sample contain a resident male spouse, I separate these households and test whether the presence of a male spouse influences the intrahousehold allocation of resources. This study will be of interest to researchers and policy makers for two reasons. First, it provides evidence on the usefulness of reported headship as a way of classifying households and identifying vulnerable groups for the targeting of social welfare policies. Second, it examines the link between family structure and the welfare of children, and in particular, establishes a positive association between women’s labor force participation and certain child outcomes.

2. DATA AND METHODOLOGY The data for this study come from the 1989 Jamaican Survey of Living Conditions (SLC), a multipurpose household survey based on the Living Standards Measurement Surveys of the World Bank, and is described in detail in Grosh (1991). The SLC was matched with the national Labour Force Survey (LFS) taken one month earlier, in order to obtain data on individual labor market behavior. The match was successful for 93.4% of reported household heads, leaving a sample of 2,982 households.’ The first step is to ascertain if the reported head in the survey actually identifies a source of household economic support. As can be seen from Table 1, in a full quarter of the cases the survey head was not even working. For FHHs only 62% of reported heads were working at the time of the survey. Table 2 shows that 82% of the time the reported head is the oldest

Table 1. Laborforce status of reported household heads by gender

Status Working Unemployed Not in labor Force Disabled




86.4 4.5 8.3

61.5 11.1 26.4

16.2 7.2 15.8




Table 2. Relation to head of oldest household member by gender of head Relation Head Spouse Parent Other relative




84.5 10.2 2.9 1.8

77.5 15.0 2.5 3.9

81.5 12.3 2.8 2.7

member of the household. The figure is lower for FHHs, only 78%.? In households where the head has a partner present (46% of all households - Table 3), the reported head is male 78% of the time. These figures demonstrate the problems with headship described in Section 1: (a) the reported head is not necessarily the main economic provider of the household, but often simply the oldest resident member, and (b) male and female-headed households tend to represent different things -a male head frequently implies an intact couple, while a female head more frequently represents a single or unpartnered woman. I therefore examine the effect of redefining household heads according to the amount of time spent working in the labor market. Ideally we would like to pick the member that contributes the most to the maintenance of the household; unfortunately income data in the LFS are very unreliable with over 60% of the data missing. Moreover, income earned may not necessarily be income contributed to the household. For example, there is evidence that a lesser share of men’s income is devoted to other household members while women’s income tends to be family income (Kumar, 1979; Dwyer and Bruce, 1988). On the other hand, hours worked in the market may be viewed as an indicator of commitment on behalf of the household, especially if women have to work longer hours to earn the same amount of cash income because of labor market discrimination.3 The Jamaican LFS gives crude information on time spent in the workplace. Hours worked in the Table 3. Households where head has a partner present MHH








reference week are coded into one of seven eight-hour blocks; months worked in the last year are given in three-month periods. Because of these data limitations “working heads” were chosen according to the following criteria. The individual who worked the most market hours in the reference week4 was designated head. When two or more individuals worked the same number of hours, the one who worked the most months in the previous year was designated head. When months worked was also the same, the oldest person was designated head. In multiple earner households (41% of the sample), age was used to determine the head in 42% of the cases (a total of 517 households, or 17% of the sample). In these households the head is thus the oldest full-time working member.J In order to capture the potential heterogeneity within male and female-headed households, I introduce a second classification of headship. The category of working head is broken down by whether the head has a partner present or not. There is evidence that women contribute more of their income to the household (Dwyer and Bruce, 1988), and income under the control of women is more strongly associated with greater household expenditures on human capital (Thomas, 1990). The objective is to see whether a single female-headed household allocates resources differently than households where the female head has a partner. The presence of a spouse may change the internal dynamics and bargaining power within the household and lead to a different set of outcomes for household members.

3. REPORTED VS. WORKING HEADS Using the new definition of head, 76% of reported heads also turned up as working heads (Table 4, Row 1). Table 5 shows that female heads are more likely to be reclassified than male heads; 27% of self-reported FHHs switched gender of head, while only 11% of MHHs became female headed using the working head definition. As a result the number of FHHs in the sample dropped from 41 to 37% under the new classification of heads.” Who are these alternative heads? Table 4 breaks down the relationship of the working head to the Table 4. Relation of working head to reported head


ALLOCATION Table 5. Gender of working head by gender of reported head

Working head



Male Female

88.7 11.3

27.2 72.8

head. In reported FHHs, 14% of the working heads are children (usually a son) and 12% are spouses of the reported head. For MHHs though, the alternative head is most likely to be a spouse. Only 5% of the working heads turn out to be offspring of the reported head in MHHs. An interesting aspect of the results is that in 12% of reported female-headed households, the (male) spouse actually works more hours in the labor market (Table 4, Row 2). Some reasons why the female partner may be reported as the head despite earning less income is if she has more education or is older than her spouse, or if she owns the house the family resides in. Focusing on these 148 households, in 78% of the cases the (reported) female head does indeed have the same level of eduation or higher than her spouse. In the cases where the reported female head has less education than her spouse, she is older 23% of the time. Of the remaining 25 “unexplained” households, 13 owned the house they live in, but we cannot tell whether the house belongs to the head or her spouse. Tables 6 and 7 compare adjusted expenditure levels and demographic composition for households under the two definitions of headship.’ Differences in household composition and adjusted expenditures are noticeably reduced when we compare working male and female heads instead of reported heads. Reported FHHs achieve a consumption level that is only 88% of their male counterparts’; working FHHs achieve a consumption level that is 97% of their male counterparts’. In addition, household size as measured by adult equivalents is actually smaller for working FHHs compared to working MHHs. Table 7 shows that reported FHHs are overrepresented in the poorest quintiles; using our alternative definition of headship, FHHs are now actually slightly underrepresented in the poorest quintiles. In fact the proportion of households with a working female head in the richest quintile is exactly the same as the proportion found in the whole sample. reported

by gender of reported head Working head Head SpollSe Child Parent Other relative Other nonrelative




(a) Multiple earners

82.2 9.1 4.7 0.3 3.3 0.6

65.7 12.3 13.5 0.2 7.4 1.0

75.5 10.4 8.3 0.2 5.0 0.7

It was noted earlier that 41% of the sample had more than one earner. The concept of a single decision maker or unified household may not be relevant in the presence of so many multiple-earner households. In Peru, Rosenhouse (1989) reports that half of the


WORLD DEVELOPMENT Table 6. Socioeconomic variable means by gender of reported and working head Reported heads

Schooling (yrs) Age Household size Adult equivalents 18+ 12-17 6-11 O-5 No. of workers % rural Adjusted exp. Enrollment (%)$ Labor force part Illness’ Observations

Working heads



7.93 48.19

7.8 1 48.96

8.04 44.50

8.06 45.68*

3.95 2.88 2.34 0.55 0.57 0.49 1.66

4.38’ 3.03: 2.35 0.67’ 0.72’ 0.65’ 1.44

4.12 2.99 2.42 0.57 0.59 0.54 1.67

4.15 2.87* 2.21’ 0.66’ 0.70; 0.58 1.39’

0.60 10770

0.52’ 9445’

0.60 10346

0.50t 10011

69.62 19.33 4.85

68.81 19.63 7.90’

67.60 21.16 5.94

71.57 17.22* 6.77







* p < .10for t-test for difference in means between MHHs and FHHs in each category i p < .05 : Enrollment and labor force participation of 14-17-year olds * Measured by whether the child had diarrhoea in the reference week (O-5-year olds) Table 7. Distribution of households by adult equivalent espenditure quintile


Reported heads MHH FHH

Working heads MHH FHH

Poorest 2 3 4 Richest

53.9 56.1 61.3 59.5 63.0

46.1 43.9 38.7 40.5 37.0

64.4 65.5 62.0 62.0 63.4

35.6 34.5 38.0 38.0 36.6






households have more than one earner, and multipleearner households tend to be poorer than single earner ones. Not only does the prevalence of multiple-earner households decrease with income in her sample, but FHHs tend to have more workers than MHHs.

Table 8 explores the relationships among multipleearner households, poverty and headship for Jamaica.

Consistent with the Peruvian study, the adjusted expenditure level of multiple-earner households is the lowest, and increases as the number of earners decreases. Indeed, the mean number of workers per household is 1.84 in the poorest quintile, and decreases steadily to a mean of 1.27 in the richest quintile (results not shown). The prevalence of FHHs is lower, however, in multiple earner households using either definition of headship, although more so for the working head definition. The latter result is consistent with Lloyd and Brandon (1993), who reported that in Ghana, when the main worker of the household is a woman, she is much more likely to be the only worker in the household. One implication of this finding is that female working heads are more likely to represent the concept implied by the term

Table 8. Socioeconomic I,ariable means by number of laborforceparticipants in the household Single earner Adjusted exp. Size Adult equivalents Reported FHHs Working FHHs Observations

111412 3.04 2.14 44.3 40.2 1750

Two earners

Multiple earners

9144 4.80 3.35

7342 7.42 5.54

36.5 32.1

37.6 30.1





“head of household” as it is used by researchers and policy makers: the main decision maker and source of financial support for the household.

(b) Children’s welfare I analyze three children’s outcomes by gender of reported and working head to see if they are sensitive to the definition of headship used. The motivation is to provide a better description of the relationship between family structure and children’s welfare. Since working heads are more likely to fulfil the conceptual idea of headship, the outcomes of interest may be more likely to reflect the characteristics of the working rather than reported head. The children’s outcomes analyzed are: school enrollment8 and labor force participation rates for 14-17-year olds, and the incidence of diarrhoea among preschoolers (children under five years old). I include both school enrollment and labor force participation because not all teenagers out of school are necessarily in the labor force. Twenty-eight percent of 14-17-year olds are not enrolled in school; of these, only 57% are in the labor force. Table 6 provides percentage means of the three outcomes by gender of reported and working head. As we move to a characterization of headship that more accurately identifies the most (economically) important person in the household, the link between a female head and improved welfare of children increases strongly. In each of the three cases, children of working female headed households are better off than children in reported FHHs. In fact, these children have higher school enrollment and lower labor force participation rates than children in any other household type. The incidence of diarrhoea among preschoolers is higher in FHHs in generaP; however, the difference between male and female-headed households becomes insignificant when working heads are compared.



Since the children’s outcomes are dichotomous, a probit function is employed to isolate the effect of headship.‘O In these equations, parental characteristics are entered separately since the child’s parents are often not the household head. For example, 85% of O-5 year olds lived with their mothers, but only 20% of these mothers were also head of the household. Father’s education was omitted due to missing values. The adjusted expenditure equation is estimated using ordinary least squares (OLS), and an F-test allowed me to exclude a set of dummy variables representing the union status of the head from the equation. In all regressions, headship is denoted by a dummy variable equalling one if the head is female. The full regression results are presented in columns (1) and (2) of Tables Al-A4 in the appendix; Table 9 summarizes the relationship between headship and the four outcomes. After controlling for demographic composition and region of residence, FHHs still have significantly lower adjusted expenditure levels than MHHs. The difference between working heads, however, is only 5.5%, compared to 9.7% for reported heads (see Table Al). Despite being poorer, reported FHHs have slightly lower teenage labor force participation rates and higher enrollment rates than reported MHHs, but worse health outcomes for preschoolers. The interesting result in Table 9 is that, consistent with the means in Table 6, the relationship between a female head and improvement in child welfare increases when we employ the working head classification.” This suggests that any disadvantage caused by the reduction in time spent with children is more than offset by the gains from participating in the labor market.‘* The full results in the appendix also highlight the importance of other household characteristics such as income, region of residence and demographic composition in determining the children’s outcomes. One exception is the illness equation (Table A4), which is poorly measured in general, and shows a weak link Table 9. Summary of coefjicient estimates offemale head on selected household outcomes*




Poverty in Jamaica is highly concentrated in the rural area, hence the higher adjusted expenditure level of working FHHs may simply reflect the tendency of these households to locate in urban areas. Similarly, children’s outcomes will in part depend on the ability of the household to provide additional caregivers and thus release other members to work. This may be especially important in working FHHs; the means in Table 6 show significantly more older children available in FHHs to perform child care and other domestic chores. I use multivariate analysis to isolate the relationship between headship, other household characteristics, and children’s outcomes.

Outcome Adjusted expenditure Labor force participation+ Enrollment Illness

Headship criterion Reported working -9.7% 4.8% 0.6% 1.9%

-5.5% -1.6% 3.6% -0.9%

* Effects are evaluated from the regression estimates presented in the appendix. For the children’s outcomes, the figures are interpreted as the probability of a positive occurrence if the household was to go from a male to a female head. These were derived by evaluating the probit function at the means of the other variables * Labor force participation and school enrollment of household members 14-17 years old; incidence of illness for O-5 year olds



between illness and income. Behrman and Wolfe (1984) among others, also find that the link between household income and child health status is empirically quite weak. The lack of significance of maternal education in any of the regression equations can be explained by the common practice of child fostering (sending children to live away from their parents) in the Caribbean, which results in a weak link between parental characteristics and children’s outcomes (see Desai, 1993, for a discussion), although maternal education can also represent genetic and early childhood factorsI Although it is tempting to attribute causality to the relationships presented, the regressions maintain headship and labor force participation as exogenous to the outcomes of interest, which may not necessarily be valid (see Browning and Meghir, 1991; Handa, 1993). For example, if more productive and motivated women participate in the labor market, then these women may also have higher children’s outcomes, and the relationship between working FHHs and children’s welfare will simply reflect this fact, instead of increased authority or bargaining power within the household.‘4 4. FAMILY STRUCTURE In this section I break down working male and female-headed households according to whether the

head has a partner residing in the household. The issue here is whether it is appropriate to treat FHHs as a homogenous group, or whether the simple malefemale classification might actually mask differences within these groups. I also consider the implications of this alternative classification for resource allocation decisions. Table 10 presents socioeconomic means for households divided by household structure. Not surprisingly, partnered households contain more people and have more workers than unpartnered ones. In addition, both partnered male and female-headed households have significantly smaller adjusted expenditures. Higher expenditure levels, however, may come at the expense of lower savings rates (for which I have no data), or simply reflect economies of scale in consumption.‘5 There are also some interesting within-group differences. Focusing on MHHs first, unpartnered MHHs have relatively fewer minors than partnered MHHs the size difference between these household falls considerably when their adult equivalent size is compared. Consistent with this observation is their different dependency burdens: each worker in a partnered MHH supports 1.74 people, while in unpartnered MHHs the figure is only 1.42 people. In FHHs the story is slightly different. The size of unpartnered FHHs, whether measured by the number of people or in adult equivalents, is consistently 75%

Table 10. Socioeconomic variable means byfamily structure and gender of working head FHHs

MHHs Partner Schooling (yrs) Age Household size Adult equivalents 18+ 12-17 6-11 O-5 No. of workers Nonworker/worker % rural Adjusted exp. Enrollment (%)$ Labor force part

Illness$ Observations

No Partner


7.97 46.21

8.13 42.511

8.05 44.79

5.12 3.54 2.76 0.75 0.83 0.78 1.87 1.74

2.95: 2.35t 2.02t 0.36t 0.30t 0.27t 1.44t 1.42t

5.01 3.53 2.78 0.76 0.82 0.66 1.91 1.62

0.61 9155 70.28 16.95 4.46 1018

0.60 11734t 61.471 30.66* 11.31 874

No Partner 8.06 45.97 3.8712.65: 2.02t 0.62t

0.66; 0.55*

1.22: 2.17:

0.52 8659

0.49 10450*

74.83 14.18 5.93

70.28 18.39 7.10


* p < .lO for t-test for difference between MHHs and FHH’s in each category tp<.05 $ Enrollment and labor force participation of 14-17-year olds 5 Measured by whether the child had diarrhoea in the reference week (@5-year olds)




of the size of partnered FHHs. The implication is that household demographic composition is very similar within this group.16 But unpartnered FHHs, have a much higher dependency burden than partnered FHHs-each worker in the former supports 2.17 people, compared to 1.62 people supported in the latter. The difference may simply be. the existence of a working spouse in partnered FHHs, although lower teenage enrollment rates and higher teenage labor force participation rates may also contribute to the difference. We thus turn our attention to intrahousehold outcomes in the next section.



household characteristics, I follow the procedure in section 3 and use multivariate analysis to clarify the relationship between children’s welfare and the presence of a resident partner in working male and femaleheaded households. I estimate separate regressions for male and female-headed households, and include a dummy variable to indicate whether the head has a partner residing in the household. Full regression results are provided in columns (3) and (4) in Tables Al-A4 of the appendix, while Table 11 summarizes the coefficient estimates of resident partner for each

Table 11. Summary of roe&em

estimates of presence of partner on selected household outcome

(a) Child welfare by family structure Section 3 showed evidence that households where the “working” head is female allocate resources in a manner that favors children. How does the presence or absence of a partner alter the distribution of resources within these households? Table 10 divides the three outcomes by family structure. In general, partnered households show higher teenage enrollment rates, lower teenage labor force participation rates and better preschooler health status than unpartnered households. Moreover, with the exception of illness, partnered FHHs have better children’s outcomes than any other type of household. The biggest within-group differences appear among MHHs, where unpartnered households display significantly different (and worse) child outcomes than partnered ones. Within FHHs, children from unpartnered households also show consistently lower outcomes, but not significantly so. Hence the lower dependency burden in partnered FHHs (compared to unpartnered FHHs) mentioned above is apparently attributable to the presence of a working spouse, rather than teenagers being pulled out of school (or housework) to enter the labor market.”

(b) Multivariate analysis One interesting observation from Table 10 is that after controlling for family structure, children’s outcomes appear to be inversely related to adjusted household expenditures. Unpartnered MHHs have the highest adjusted expenditures but the worst outcomes, while partnered FHHs have the lowest adjusted expenditure level and the highest outcomes. The addition of a spouse to a household can raise potential household disposal income without a major effect on expenditures because of savings, durable goods investment9 and economies of scale in consumption, Of course the presence of a spouse can influence children in other noneconomic ways, by providing guidance, discipline, and a positive role model. TO control for the influence of income and other




Adjusted expenditure Labour force participation Enrollment Illness

0.3% -11.2% 7.4% -4.1%

-1.0% -2.3% 2.5% 1.6%

* For the children’s outcomes, the figures give the probability of a positive occurrence associated with a partner (of the head) residing in the household. See notes to Table 9 for further details.

outcome. The results in Table 11 indicate that while the relationship between resident partner and adjusted expenditure is small, substantial difference in children’s outcomes exist within MHHs. The presence of a resident spouse in working MHHs is associated with a reduction in the probability of teenage labor force participation and child illness by 11 and 4%, respectively, and an increase in the probability of school enrollment by 7%; the coefficients in the former two regressions are statistically significant (see appendix). Within FHHs on the other hand, the differences are much smaller, and none are statistically significant. Finally, in a multivariate context, the level of household expenditure is positively related to child welfare, with the exception (once again) of child health. Taken together, the implication of these results is that even after controlling for income, there is an important link between family structure and child welfare. Indeed one conclusion that can be drawn here is that it is the absence of a potential female decision maker, rather than a female head, that is negatively correlated with child welfare.

5. CONCLUSION This paper addresses some issues surrounding the concept of headship, the classification of male and female-headed households, and the intrahousehold allocation of resources. The self-reported definition of



headship in the Jamaican SLC identifies the person who works the most market hours in the household (and hence likely to provide economic support for the household) in only three-quarters of all households. Interestingly, reported FHHs are reclassified more often than MHHs, in spite of the popular view that for social and cultural reasons, surveys in developing countries underreport the number of FHHs because many MHHs are actually maintained and supported by females. In 12% of FHHs, a female was declared head despite the presence of a working male spouse. These results are consistent with the social and economic status of women in Jamaica. There is a strong tradition of FHHs, the female labor force participation rate is one of the highest in the world (Edwards and Roberts, forthcoming), and women play an active and important role in the maintenance of their families. The welfare of children is sensitive to the definition of headship. When households are grouped by the gender of the “working” head, children in FHHs display better short-term health status and school enrollment rates, and are less likely to participate in the labor force. These results are consistent with a number of different underlying causal mechanisms: female labor force participation may represent bargaining power, access to resources, or unobserved productivity, all of which result in better child outcomes. The data and analysis presented here are not sufficient to distinguish between these different mechanisms, and they represent important topics for future research on household behavior and child welfare in developing countries. The simple male-female dichotomy also hides important differences in household income, demographic composition, and intrahousehold resource allocation within these two groups. Partner-headed

households tend to be poorer, but have better children’s outcomes. The most striking within-group differences occur in MHHs. Despite a significantly higher level of household expenditure, children in unpartnered MHHs have significantly lower school enrollment and health status, and higher labor force participation rates than partnered MHHs. In fact, children in unpartnered MHHs are at a higher degree of risk than children in any other household type. In contrast, differences within FHHs are small and not significant, and both types of FHHs have similar demographic compositions. The analysis of family structure leads to the important conclusion that it is not the absence of a female head per se, but the absence of any potential female authority within the household that is negatively associated with child welfare. As a whole, the findings of this study illustrate the need to interpret the survey reported designation of household head with care. Its relevance as a way to identify the most important person in the household will depend on local factors such as the status of women and their role in the market economy. The results also suggest that research for social welfare policy must carefully distinguish between different types of female and male-headed households; although the link between women heads and child welfare is positive, headship per se may not be the relevant concept, but rather the presence of a potential female authority figure in the household. Care must be taken, however, in generalizing these results to households in other parts of the world. Given the unique position of women in Jamaican society, it may be the case that they have substantial bargaining power within the household regardless of their headship status.

NOTES 1. The percentage of female-headed households is very slightly lower in the matched sample (41.0% compared to 42.5% in the SLC), and the excluded households exhibit the same demographic characteristics as the included ones.

sification and conclusions did not change significantly when months worked was used as the primary criterion. One reason hours worked was retained as the first criterion was because slightly more detail was provided in this question.

2. The difference between male and female-headed households, and female-headed households and the full sample, are each significant at the 5% level.

5. To check the reliability of my definition, I compared months worked in the previous year for the 723 households that changed heads. In only 18 of these households (2.5%) was it the case that the original reported head had actually worked more months in the year than the new “working” head. In only eight cases was there more than a three-month discrepancy.

3. Scott (1991) reports a very large unexplained portion of the difference between male and female earnings in Jamaica. The extent of discrimination may, however, be overstated if characteristics identified with men such as bursts of strengths are valued more than manual dexterity and stamina. which are more identified with women (Pitt, Rosenzweig and Hassan, 1990). 4. In cases where the number of hours worked in the previous week were unusual (due to sickness, holiday, etc.) only months worked in the last year were compared. The clas-

6. Eleven percent of the households reported no working members. Half of these households consisted of retired heads over the age of 65. The reported head was retained as the working head in these housetolds. 7. Total expenditure is adjusted for household composition using the following scales: O-5: 0.2; 6-l 1: 0.3; 12-17:



0.5; 18 and over: 1.0. 8. Because primary and middle Jamaica is universal I concentrate children.

school enrollment in on secondary school

9. This variable is particularly sensitive to respondent bias. For example, male respondents in MHHs may not know whether the child had diarrhoea in the previous week. 10. These equations may be interpreted as reduced form demand equations for children’s health and education, derived from a new household economics model. See Chenery and Srinivasan (1988) chapters 13 and 14. 11. Rosenzweig and Schultz (1983) make a similar point for a sample of US women. 12. The effect of headship however is generally insignificant in a statistical sense - see full results in the appendix. 13. Early childhood effects of maternal education would diminish if fostering occurred at a young age. Of the preschoolers that did not live with their mothers (15% of the sample), 25% were two years old or younger, and the other



75% were evenly split between three, four and five-year olds. The rate of fostering is much higher for 14-17-year olds in general (33%), but within this group there is not much variation by age. For example, of those teenagers not living with their mother, half were 14 or 15 years old and the other half 16 or 17 years old. 14. This example referees.

was provided

to me by one of the

15. For example, Behrman and Wolfe economies of scale within households.



16. In other words, the proportion of adults to children is similar among these households (although partnered FHHs have more of each group). 17. This is consistent with the earlier observation that when the main worker of the household is female, there is more likely to be only one earner. 18. The measure of expenditure used here includes imputed rent from housing, but does not include the flow of services from cars and time-saving household appliances.

REFERENCES Barros, R., and Louise Fox, “Female headed households, poverty, and the welfare of children in urban Brazil,” Mimeo (New Haven: Yale University Economic Growth Center, 1990). Behrman, J. and B. Wolfe, “More evidence on nutrition demand: Income seems overrated and women’s schooling underemphasized,” Journal of Development Economics, Vol. 14 (1984). pp. 105-128. Browning, M., and C. Meghir, “The effects of male and female labour supply on commodity demands,” Econometrica, Vol. 59, No. 4 (1991). pp. 925-95 1. Chenery, H., and T. N. Srinivasan (Eds.), Handbook of Development Economics Vol. 1 (New York: North Holland, 1988). Desai, S., “Children at risk: The role of family structure in Latin America and West Africa,” Population and Development Review,Vol. 18, No. 2 (1993). Dwyer, D., and Judith Bruce (Eds.), A Home Divided: Women & Income in the Third World (Palo Alto: Stanford University Press, 1988). Edwards A. C., and J. Roberts, “Structural adjustment and women in the labor force: Evidence from Latin American countries,” in S. Horton, R. Kanbur & D. Mazumdar (Eds.), Labor Markets in an Era of Adjustment (Washington DC.: World Bank, 1994). Folbre, N., “Mothers on their own: Policy issues for developing countries,” Mimeo (New York Population Council, 1991). Grosh, M. E., “The household survey as a tool for policy change: Lessons from the Jamaican Survey of Living Conditions,” Living Standards Measurement Study No. 80 (Washington DC: World Bank, 1991). Handa, S., “Family structure, female headship, and children’s welfare in Jamaica,” PhD dissertation (Toronto:

University of Toronto, 1993). Horton, S., and B. Miller, “The effect of gender of household head on food expenditure: Evidence from low income households in Jamaica,” Mimeo (Toronto: University of Toronto, Dept. of Economics, 1987). Kennedy, E., “Effects of household structure on women’s and children’s nutritional status,” Mimeo (Washington DC: International Food Policy Research Institute, 1992). Kennedy, E., and P. Peters, “Household food security and child nutrition: The interaction of income and gender of household head,” World Development, Vol. 20, No. 8 (1992). pp. 1077-1085. Kossoudji, S., and Eva Mueller, “The economic and demographic status of female headed households in rural Botswana,” Economic Development & Cultural Change, Vol. 31, No. 4 (1983), pp. 831-859. Kumar, S., “Impact of subsidized rice on food consumption and nutrition in Kerala,” Research Report No. 5 (Washington DC.: International Food Policy Research Institute, 1979). Lloyd, C. and A. Brandon, “Women’s role in maintaining households: Family welfare and sexual inequality in Ghana,” Population Studies, Vol. 47 (1993). pp. 115-131. Louat, F., M. E. Grosh and J. van der Gaag “Welfare implications of female-headship in Jamaican households,” Living Standards Measurement Study No. 96 (Washington DC: World Bank, 1991). _ Massiah, Jovcelin (Ed.), “Special issue on Women in the Caribbean,” Social and Economic Studies, Vol. 35, Nos. 2 and 3 (1986). Mohammed, Patricia, and C. Shepherd (Eds.), Gender in Caribbean Development (Kingston, Jamaica: Institute for Social and Economic Research, 1991). Pitt, M., M. Rosenzweig, & M. N. Hassan, “Productivity,



health and inequality in the intrahousehold distribution of food in low income countries,” American Economic Review, Vol. 80, No. 5 (1990), pp. 1139-l 156. Rosenhouse, S., “Identifying the poor: Is headship a useful concept?” Living Standard Measurement Study No. 58 (Washington DC: World Bank, 1989). Rosenzweig, M., and T. P. Schultz, “Consumer demand and household production: The relationship between fertility and mortality,” American Economic Review, Vol. 73,

No. 2 (1983). pp. 38-42. Scott, K., “Female labor force participation and earnings: The case of Jamaica,” in G. Psacharopoulos & Z. Tzannatos (Eds.), Women’s Employment andPay in Lutin America (Washington DC: World Bank, 1991). Thomas, D., “Intrahousehold resource allocation: An inferential approach,” Journal of Human Resources, Vol. 25, No. 4 (1990), pp. 635-664.

APPENDIX Table Al. Regression coefficients on log of adjusted expenditures (OLS)* Full sample Reported head Working head Variable

Working heads MHHs FHHs





-0.003’ (0.001) 0.054 (0.004) -0.097’ (0.023)

-0.000 (0.001) 0.059: (0.005) -0.055* (0.024)

0.001 (0.001) 0.061’ (0.006)

xUlO3’ (0.001) 0.054’ (0.007)

0.003 (0.032)

-0.010 (0.043)

Head+ Age (ye=) Schooling (years) Gender (female = 1) Partner (yes = 1) Composition 18+ 12-17 6-11 O-5 Household Rural area (yes = 1) Constant

R-squared Prob>F Nobss

-0.102’ (0.009) -0.069* (0.013) -0.055f (0.013) -0.073’

-0.124* (0.009) -0.065’ (0.013) -0.049f (0.013) -0.068’

4)).126’ (0.012) -0.047’ (0.017) 4.061 (0.012) -0.079’

-0.120’ (0.014) -0.096’ (0.020) -0.036* (0.020) -0.046’

-0.441’ (0.023) 9.318* (0.066)

-0.442* (0.024) 9.168* (0.069)

-0.4351 (0.03 1) 9.079f (0.094)

-0.457’ (0.037) 9.250’ (0.118)

0.312 0.000 2982

0.300 0.000 2982

0.290 0.000 1860

0.329 0.000 1074

* Since the dependent variable is in logs, the coefficients give the percentage effect on the dependent variable of a unit change in the independent variable. Standard errors in parentheses. +Column (1) uses characteristics of the reported head, while columns (2t(4) use the characteristics of the working heads *p < .05 5 Some working heads had missing values for education


Table A2. Regression





Full sample Working head Reported head

Hea& Age (years) Schooling (years) Gender (female = 1) Partner (yes = 1) Child Mother’s sch.

(years) Gender (male = 1) Age (years) Composition 18+ 12-17 6-11 O-5 Household Log adjusted exp. Region (1 = rural)

Pseudo R* p value (x’) Observations’ * Sample is all households


ALLOCATION of 1617-year

olds (probit)*

Working heads MHHs FHHs





0.003 (0.004) -0.020 (0.023) -0.037 (0.099)

-O.Olor (0.004) &042i (0.022) -0.079 (0.105)

-0.006 (0.005) -0.052r (0.028)

-0.004 (0.006) 0.056 (0.041)

4.47 1’ (0.157)

-0.129 (0.194)

0.027 (0.021) 0.368’ (0.099) 0.721’

0.029 (0.020) 0.344’ (0.100) 0.715’

0.018 (0.027) 0.246’ (0.131) 0.740’

0.028 (0.035) 0.483’ (0.164) 0.698’

(0.053) -0.058’ (0.035) 0.104’ (0.050) -0.010 (0.047) 0.084’

(0.054) -0.061 (0.035) 0.088’ (0.050) -0.016 (0.047) 0.068

(0.069) -0.091’ (0.047) 0.113’ (0.062) 0.090 (0.065) 0.056

(0.089) -0.031’ (0.057) 0.028 (0.088) -0.088 (0.077) 0.140

(0.05 1) -0.335’ (0.085) 0.271’ (0.112) -10.067’ (1.145)

(0.050) -0.326’ (0.086) 0.304’ (0.113) -9.267’ (1.141)

(0.064) -0.256’ (0.109) 0.364’ (0.154) -10.457’ (1.528)

(0.087) -0.412’ (0.146) 0.225 (0.179) -8.895’ (1.802)

0.256 0.000 1172

0.260 0.000 1148

that had at least one child 14-17-years

0.295 0.000 673

0.240 0.000 475

old. Standard errors in parentheses

* Column (1) uses the reported head definition, while (2)-(4) use the working head definition ‘p.c.05 ‘PC.10 MSeveral working heads had missing values for education, hence the discrepancy addition, some teenagers had missing values for labor force participation

in the sample sizes. In


WORLD DEVELOPMENT Table A3. Regression coeffrcienrs for school enrollment 14-I 7-year olds (probit)* Full sample Reported head Working head

Age (years) Schooling (years) Gender (female = 1) Partner (yes = 1) Child Mother’s sch. (years) Gender (male = 1) Age (years) Composition 18+ 12-17 6-11 o-5 Household Log adjusted exp. Region

(1 = rural)


Pseudo R? p value (x2) Observations”

Working heads MHHs FHHs





-0.001 (0.003) 0.030 (0.019) 0.019 (0.082)

0.002 (0.003) 0.034’ (0.019) 0.105 (0.085)

0.003 (0.005) 0.044’ (0.023)

-0.006 (0.005) -0.001 (0.034)

0.211 (0.133)

0.078 (0.148)

0.003 (0.017) -O.189o (0.08 1) -0.399’ (0.037)

0.002 (0.016) -0.198’ (0.08 1) Al.406’ (0.038)

-0.003 (0.021) -0.222” (0.107) 4.415’ (0.049)

0.020 (0.028) -0.174 (0.128) -0.386’ (0.061)

0.060~ (0.029) -0.017 (0.040) 0.011 (0.038) 4.098’ (0.043)

0.063 (0.029) ~.004 (0.040) 0.018 (0.038) 4.093 (0.043)

0.068’ (0.038) -0.045 (0.050) 0.024 (0.053) -0.055 (0.055)

0.067 (0.047) 0.083 (0.072) 0.024 (0.060) -0.169t (0.072)

0.332’ (0.069) 4.059 (0.089) 3.720’ (0.873)

0.344” (0.070) -0.064 (0.089) ‘3.468; (0.870)

0.3436 (0.088) -0.135 (0.120) 3.693* (1.158)

0.328’ (0.119) a.013 (0.143) 3.825’ (1.383)

0.117 0.000 1219

0.121 0.000 1195

* Sample is all households that had at least one child l&17-years

0.141 0.000 699

0.112 0.000 496

old. Standard errors in parentheses

. Column (1) uses the reported head definition, while (2)-(4) use the working head definition :p< 10. ‘PC.05 MSeveral working heads had missing values for education, hence the discrepancy addition, some teenagers had missing values for labor force participation

in the sample sizes. In





Table A4. Regression coefficients for illness O-S-year olds (probit)* Full sample Reported head Working head Variable

Working heads FHHs MHHs





-0.002 (0.005) 0.007 (0.030) 0.177 (0.131)

-0.008 (0.005) 0.010 (0.030) -0.084 (0.141)

-0.004 (0.007) -0.005 (0.039)

-0.004 (0.008) 4)).064 (0.045)

-0.358* (0.201)

0.138 (0.269)

Head Age (years) Schooling (years) Gender (female = 1) Partner (yes = 1) Child Mother’s sch. (years) Age (months) Gender (1 = male) Father present (1 = yes) Composition 18+ 12-17 611 &5 Household Log adjusted Rural area

Pseudo R’ p value (y,‘) Nabs”

-0.032 (0.029) -0.013’ (0.004) a.117 (0.121) -0.330’ (0.147)

-0.036 (0.029) -0.013’ (0.004) -0.111 (0.121) -o.464s (0.151)

-0.012 (0.040) -0.011’ (0.005) -0.070 (0.160) a.332 (0.205)

-0.064 (0.045) -0.016’ (0.006) a.158 (0.195) a.406 (0.334)

-0.022 (0.041) -0.089 (0.064) 0.098’ (0.050) -0.050 (0.059)

-0.034 (0.040) a.079 (0.063) 0.106’ (0.050) 0.055 (0.059)

a.034 (0.056) -0.052 (0.077) 0.1688 (0.065) 0.026 (0.074)

-0.069 (0.069) -0.103 (0.114) 0.015 (0.086) 0.142 (0.106)

-0.085 (0.113) -0.106 (0.136) -0.064 (1.140)

-0.074 (0.113) -0.128 (0.137) 0.265 (1.106)

-0.097 (0.149) a.042 (0.185) -0.099 (1.451)

0.039 (0.180) -0.23 1 (0.213) -0.568 (1.797)

0.063 0.001 1165

0.065 0.001 1156

0.082 0.016 711

0.084 0.084 445

* The dependent variable equals one if the child had diarrhoea in the reference week. Standard errors in parentheses. +Column (1) uses characteristics of the reported head, while columns (2)-(4) use the characteristics of the working heads ‘p<.lO ip<.O5 ” Several working heads had missing values for education, hence the discrepancy in the sample sizes