Why head the household? Demographic effects on time use in Jamaica

Transcription

1 Why head the household? Demographic effects on time use in Jamaica Sudhanshu Handa Department of Public Policy CB#3435 University of North Carolina Chapel Hill, NC Sue Horton Division of Social Sciences University of Toronto at Scarborough 1265 Military Trail Scarborough, Ontario M1C 1A4 Canada Abstract It is well-known that household heads have strong influence on household decisions. What is less well documented is that this occurs at personal cost to the head, in terms of longer hours worked both in the market and in total, and (for female heads) poorer general health status. This holds true both in simple averages, and in multivariate analysis. The Jamaica data used also reject the hypothesis that headship is randomly or exogenously determined for males, and suggest that labour supply equations which treat headship as exogenous overestimate the wage elasticity for men by 15%. JEL Classification numbers I12; J12; J32 Keywords household time allocation; household headship; female headed households

2 Why head the household? Demographic effects on time-use in Jamaica 1. Introduction The head of the household is seen as playing an important role in some models of intrahousehold resource allocation. Several of the unitary household models assume that the head of the household acts as a dictator, who may be either benevolent or an effective altruist or a despot who subordinates and subjugates other household members. 1 Either way, the head in these models has considerable influence over the intrahousehold allocation of resources. In some empirical tests of collective models of intrahousehold allocation, the head also plays a key role. Female headed households have been found to have better child outcomes such as school enrolment and completion 2, lower incidence of moderate and severe malnutrition 3, more child-oriented expenditure patterns in terms of more nutritious foods and lower shares of expenditure on alcohol and tobacco 4, and better protection of consumption against economic downturns 5. Thus from both a theoretical and empirical perspective, the household head is posited to have considerable influence over intrahousehold allocation and the individual welfare of household members. But how about the household head herself? What are the welfare implications of headship for household heads? Despite the importance of the household head in theoretical models of family decision-making, as well as her observed importance in determining the welfare of individual household members (see empirical literature cited above), the welfare implications of headship for heads themselves has been neglected. Clearly there are strong normative assumptions about the role and importance of the household head in family matters. 1

3 To the extent they are true, these behaviourial assumptions are essentially a constraint on the behaviour of heads, with obvious welfare implications, for it is not inconceivable that the burden of headship could actually reduce the welfare of heads relative to other household members. In contrast to the many studies (some of which are cited above) that look at how headship (or characteristics of the head) affect the welfare of other household members (such as children), this paper focuses on how heading the household may effect the welfare of the head herself, and to our knowledge, is the first of its kind to do so. We assess the welfare implications of headship by comparing the consumption of leisure between heads and non-heads from a national household survey on time use in Jamaica. Simple averages suggest that being the household head is associated with working longer hours in the market and in total (on average 7-9 hours more in the market, and 8-12 more in total) than other household members of the same sex. More careful regression analysis suggests that this difference is even greater, holding constant other characteristics. This effect holds both for male and female household heads. Thus the greater decision-making power of the head over the consumption of household goods, as emphasized in the literature, contrasts with the lower consumption of leisure by the head. As a further indicator of individual well-being, we also analyse whether household heads are more likely to report being sick, to report being in poor general health, and to be limited at activities of daily living (ADLs). Simple means show that heads perform worse than non-heads in all indicators of health status. However, once controlling for age and other observable characteristics, the only significant difference that remains is for poor general health among females, where heads continue to report significantly higher incidences than non-heads. 2

4 A further question addressed in this paper is whether headship is determined exogenously. Using the highest (predicted) shadow wage in the household for identification, we test the hypothesis that headship is in fact endogenous. The data reject the hypothesis that headship is randomly or exogenously determined for males, indicating that the determination of who heads the household cannot be separated from decisions on consumption patterns, in particular the allocation of leisure. The opens the larger issue as to whether household composition is itself endogenous, a question which has concerned labor economists and demographers for some time, but which is not addressed here, although this paper can be seen as a step towards understanding the larger issues of family formation and composition and their influence on household welfare and economic development. The results in this paper have two main implications. First, they suggest that in models of household time allocation, the headship variable cannot necessarily be considered exogenous. Treating headship as exogenous affects the estimated impact of individual characteristics on time use and labor supply, and more seriously, overestimates the wage elasticity for men by 15%. Given the simultaneity of consumption decisions, this implies that future studies of the effect of male or female headship on the consumption of goods should test whether headship is in fact exogenous. Second, the results indicate that at least in this population, there is a considerable burden of headship in terms of reduced consumption of leisure. Previous work on household formation and intra-household resource allocation has viewed headship as a positive characteristic, due to the increased decision-making power and influence in household resource allocation that it brings. Results from this paper indicate that this increase in authority comes at a cost. 3

5 2. Literature, Theoretical Framework, and Empirical Model The standard theory of individual labour supply to the market posits that hours worked are a function of individual and household characteristics (X) such as age and household demographic composition, household non-labor income (I), and the individual s wage (w): H = f (X, I, w) (1) The individual works in the market (has non-zero hours) if the market wage exceeds some reservation value, which is itself determined by many of the same individual and household characteristics. Given the censoring of the sample (wages are only observed for those for whom market wages exceed reservation wages), a shadow wage can be estimated, correcting for the selectivity of participating in market work. New household economics models extend the basic framework above to deal with hours worked in home production as well as in the market, and also to consider the intra-household allocation of time. These models allow for production functions for home production which may (but do not necessarily) differ for men and women. These production functions define the marginal returns to additional hours spent in home production. Depending on the shape of the home production function, household members decide on the allocation of time to market and home activities. Since market wages for men typically exceed those for women (even holding constant individual characteristics), the household may choose to specialize such that men work longer hours in the market and women may undertake a greater proportion of household production 6. A basic household economics model of individual time allocation can be sketched out as follows: 7 Max Utility U = U(Z, L) (2) 4

6 where Z is home produced goods and L is leisure, subject to the following constraints: Z = Z(x, t; D) (3) household production function, where x is purchased commodities, t time, D household demographics T = H +t +L (4) time constraint, where T is total time endowment, and H market hours Y = w*h + I (5) income constraint, where Y is income, w the wage, and I unearned income p*x = Y (6) budget contraint, where p is the price of x The constraints (4)-(6) can be combined to derive the full income constraint w*l + w*t + p*x = w*t + I (7) Equation (7) says that full expenditures (left hand side) including leisure consumption cannot exceed full income. The individual chooses the optimal levels of market (H) and domestic work (t), and combination of home produced (Z) and market purchased goods (x), given all prices and household demographic composition, which is treated as exogenous. We will allow some aspects of household demographic composition (headship) to be endogenous in the empirical analysis below. Reduced form demand equations for Z, x, and leisure (defined as T-H*-t*) will then take the form Dj* = D(p, w, I; D) for j = L, Z, and x (8) Equation (8) is the basic model we estimate, using individual level data on Jamaican adults. Our dependent variable is the opposite of leisure, total time worked (H+t), although we also 5

7 provide some estimates of the determinants of market hours worked only (H), in order to calculate and compare conventional labor supply elasticities for men and women. Previous empirical studies of market labour supply in developed countries have found that unearned income has a negative effect on labour supply, and the effect of market wages can be to increase or decrease labour supply (depending whether income or substitution effects dominate), although the magnitude of the wage response varies substantially. In addition, female (uncompensated) wage elasticities tend to be higher than those for men due to the additional substitution effect out of home production The presence of small children in the household is negatively associated with women s market labour supply, although the issue of cause and effect in this relationship remains unresolved. 8 Studies for developing countries and of total (market plus domestic) labour supply are harder to find. Some findings from this literature include Behrman s study using Nicaraguan data that suggests that the relationship between fertility and female labor suppy is much weaker in developing countries due to the existence of extended families, and Malathy s analysis of household production in urban India, which shows that the shadow wage significantly reduces married women s time allocation to home production. Skoufias finds that gender and age composition of the household affects the allocation of time to work in the market and work in the home, particularly for women, in rural India. 9 Only one previous study of time use known to the author includes headship status of the individual, and that study, for Ghana, found that heads work longer hours in the markets than other individuals. 10 The headship variable has been found to have strong effects on consumption of goods (see references earlier), but headship need not be exogenously 6

8 determined. In the empirical work which follows, we test this, subject to the hypothesis that household composition is exogenously determined. It is quite possible that household composition is also not exogenous (although most models of consumption and labour supply assume that it is); nevertheless testing the exogeneity of headship is a useful first step towards understanding the larger issue of household formation. One previous study for Jamaica tests whether female headship is endogenously determined, by modeling the consumption of a woman and her children both in a separate female headed household, and as a component of a larger male-headed household. 11 However that model does not address the joint determination of headship and household composition more broadly, in part because opening the Pandora s box on exogeneity in these models leaves very few variables as truly exogenous (see for example Lundberg et. al. on the issue as to whether unearned income is truly exogenous: unearned income is frequently used as a variable in models of intrahousehold resource allocation). 12 Empirical Strategy We estimate a linear version of equation (8) of the following form: Time = α + β 1 *{demographic composition}+ β 2 *{union status}+ β 3 *{age} + β 4 *{shadow wage}+ β 5 *{headship status} + µ (9) where α is a constant, µ the error term, and β i are parameters to be estimated. The shadow wage for all individuals age 15 and over is predicted based on a selectivity (for participation) corrected (log) wage equation estimated separately for men and women. The wage equation included education, age, age squared, and geographic indicators (dummies for Kingston Metropolitan 7

9 Area/town/rural area). Probit estimates for observing a market wage include the same independent variables, with additional variables indicating household composition (number of male adults and female adults in addition to the individual, number of children under 5, and number of children 5 to 14), and individual information on conjugal status (three dummy variables indicating married /common law, visiting relationship, and single/none; and a dummy indicating whether or not the individual s partner was presently living in the household). Based on the selectivity corrected wage function, predicted wages (including the selection correction) were obtained for all individuals 15 and over. These predicted wages are lower than market wages averaged over the whole sample (i.e. individuals with lower market wages tend to select themselves out of the market see Table 1). The estimated wage equation is shown in the Appendix and additional information about this procedure is available from the authors upon request. Equation (9) is estimated separately for males and females by OLS, for total hours worked and market hours only, and with and without single person households. For market hours worked, we also provide TOBIT estimates given the large number of individuals who report 0 market hours worked. Summary statistics of the relevant variables used in the analysis are provided in Table Data and Descriptive Analysis Data The data for the study come from a special Employment and Time Use module of the 1993 Jamaican Survey of Living Conditions (SLC), an annual nationally representative household survey conducted by the Jamaican Statistical Institute. 13 Information was collected on details of 8

10 time use for all household members age 14 and over. Hours worked in the market here are defined to include hours in one or more occupations, as well as commuting time. Information was collected on hours spent in school or studying. Individuals were also asked how much time they spent in nine non-market (non income-earning) activities, and these were totaled to give domestic work. The nine categories are: food preparation, washing dishes and house cleaning, home improvement including lawn mowing, repairing appliances or the home, washing or ironing clothes, fetching water, gathering fuel, shopping or running errands, and other activities (kitchen garden, sewing). Total time spent working is the sum of domestic work and market work (including commuting). The employment and time use data were in addition to the detailed socio-economic information normally collected in the SLC. There are a total of 4683 individual observations with usable age data in this section of the SLC, coming from 1910 households. The demographic variables were constructed as follows. Ten-year age groupings were created (with the exception of the youngest and oldest groupings) to include individuals 0-4, 5-14, 15-24, 25-34, 35-44, 45-54, 55-64, and 65 and above. Schooling in Jamaica is compulsory up until age 14 (grade 9). Above age 24, virtually no hours are reported for studying in the sample. Retirement age in Jamaica in 1993 was 60 for women, 65 for men. For purposes of this paper, an adult is defined as 15 years old and above (based on the end of compulsory schooling): individuals in the age group are defined as young adults. Although data were collected on time use of 14-year olds, these were not included in the estimates since not all of the 14 year olds have completed compulsory schooling. Household structure and composition is varied in Jamaica. Of the 1910 households in the 9

11 whole sample, 61% had members less than 18. Of these households with children, 37% had one male over the age of 18 and one female over the age of 18, 22% had one or more females over the age of 18, 5% had one or more male over the age of 18, and 35% had three or more individuals over 18, including at least one male and one female. Only one individual younger than 18 lived without any individual over 18 - one 15-year old headed his own (uniperson) household. Headship is self-defined by the household. The key respondent was asked who the head of the household was, and the relationship of other members to this person. Handa explores the characteristics of heads further; according to his analysis heads are frequently (although not universally) important economic supporters of the household (in terms of cash income) and work longer hours in the market than other household members. 14 The probability of heading the household varies according to an individual s characteristics, and many of the same characteristics associated with working more in the market, are also associated with heading the household. Men are more likely than women to be household head (49.3% of males 15 and over head their household, versus 35.1% of females in the same age range). Less than half of the youngest and oldest age groups of adults (aged less than 35, and 85 and over) head households, as compared to 57% of individuals aged 35-44, and two-thirds or more of individuals aged % of heads report market wages as compared to only 41% of non-heads. By definition, household heads and non-heads are distributed differently across households of different sizes. All of the individuals in the 546 households with one adult are household heads, half of the 1364 individuals in two-adult households, and there are only 736 heads among the 2972 individuals living in households with more than two 10

12 adults. Descriptive Statistics Heads have higher hourly wages than non-heads. Of the 1397 households in which at least 1 member reported a market wage, the head reported the highest wage in 937 cases. Of course, a number of these were one-adult households where only the head reports a wage. Excluding these lone-adult households, there were still 614 households out of 1082 (again, where at least one individual reported a market wage) where the head had the highest hourly wage. Thus in households with at least two adults, and where at least one adult reported a market wage, 57% of heads as compared to 22% of non-heads were the highest wage earner in their household. Table 1 provides means of selected work related variables for heads and non-heads by sex. Males participate more in the labor market than females, and heads participate more than non-heads. The actual wages of men and women who participate in the work force are virtually the same, although there is some difference between male (J$29.64) and female heads (J$25.14). However the difference in the shadow wage between males and females for all individuals is large as is to be expected. Demographic variables have strong effects on patterns of time use in Jamaica, consistent with other studies. There are strong lifecycle effects, with the highest work hours being reported in prime working age (approximately ages 25-54). In this age group, men work hours per week in the market as compared to hours for women (Figure 1). Domestic work is primarily a female responsibility with hours per week between ages 20 and 65, whereas men in the same age range spend a little over 5 hours per week in domestic work (Figure 2). The combination is that women s work exceeds men s during the reproductive years (approximately 11

13 ages 20-50), with the greatest demands being when children are small, at which time women work on average 8 hours per week more than men. After age 50 this situation is reversed, and men s total work hours exceed women s. (Place Figures 1 and 2 about here) Being household head has a marked effect on total time use. Household heads work more hours in the market than non-heads, and this is true both for men and women. Male heads over the lifecycle work an average of 12 hours per week more than male non-heads (Figure 3), and the corresponding difference for women is 8 hours (Figure 4). Most of this is attributed to longer hours in the market: an additional 9 hours per week for men, and 7 more for women (Figures 5 & 6). Male heads also work longer hours in domestic work than do non-heads (Figure 7), although the reverse occurs for women in the prime age groups (Figure 8). (Place Figures 3-8 about here) 4. Results Basic OLS Estimates Columns (1) and (4) of Table 3 present coefficient estimates for our basic OLS model determining total hours (domestic and market) worked for men and women respectively. Even after controlling for age and household composition, household heads work on average 8 hours more per week than other household members and this difference is statistically significant. The shadow wage is also a statistically significant determinant of total hours worked for both sexes, and when evaluated at the respective means, the coefficients imply wage elasticities of 0.32 and 0.22 for men and women respectively. Columns (2) and (5) allow this price response to vary among heads and non-heads by introducing a dummy interaction term between headship and the 12

14 shadow wage. For both men and women this interaction term is large and negative, indicating lower price elasticities for heads, but only for men is the difference statistically significant at conventional levels. The coefficients in column (2) for males imply wage elasticities of 0.50 and 0.25 for non-heads and heads respectively; while for women the corresponding elasticities are 0.21 and Hence not only are heads consuming less leisure in absolute terms, their response to wage changes is also lower than the similar response for non-heads. In Table 4 we repeat these estimates using market hours as the dependent variable, and find the same general pattern. Both male and female heads work an average of 6 more hours per week than non-heads (columns 1 and 4), and as before, the shadow wage is a strong determinant of hours worked. The wage elasticities for men and women, based on the coefficient of shadow wage in column (1) and (4), are 0.45 and 0.58 respectively. Note that for market hours worked the wage elasticity is larger for women relative to men, while for total hours worked the reverse is true. Columns (2) and (5) allow for the wage elasticity to differ between heads and non-heads, and this proves to be especially important for men. For men, the elasticities come out to be 0.68 and 0.36 for non-heads and heads respectively, while for women the elasticities are 0.67 (nonheads) and 0.47 (heads), although the difference for women is again not statistically different. Overall, our results so far indicate a non-trivial burden associated with heading a household in Jamaica. For both market and total hours, heads work significantly more, and have lower wage elasticities than non-heads. The estimated difference in price effects are especially large for men, with calculated point elasticities almost 50 percent less for male heads compared to their non-head counterparts. There is an important gender pattern to the estimated elasticities. 13

15 Women are more price responsive to market hours relative to men, but less price responsive to overall hours (market plus domestic) worked relative to men. This result is consistent with the observed pattern of time use in Jamaica and elsewhere, where women tend to spend substantial time at domestic activity regardless of their labor force status, and has also been reported in the literature on female labor supply in more developed economies. Endogenous Headship: IV Estimates Most work on household consumption patterns, including children s health, educational, and nutritional status, maintain household structure and headship as exogenous. There are a few studies that attempt to control for the fact that unobserved characteristics associated with family formation may be correlated with observed outcomes such as budget shares or children s health, and these have shown that this correlation can significantly alter the measured impact of household characteristics on such outcomes. 15 In the present case, our concern is that unobserved characteristics (at either the household or individual level) associated with who becomes household head, also determine the allocation of time among household members, thus leading to inconsistent estimates of the impact of headship on leisure consumption. In practice this correlation can lead to either an upward or downward bias in the estimated impact of headship on hours worked. For example, unobserved individual level heterogeneity, where heads are more motivated or talented and would work longer hours anyway, will lead to overestimates of the impact of headship on labor supply in a standard OLS framework. On the other hand, household level unobservables such as the distribution of native abilities or talents among household members could require the head to work longer hours than would otherwise be captured via OLS 16, leading to downward bias in the 14

16 OLS framework. We address this potential problem using instrumental variables, where our key identifying variable is a dummy variable indicating whether or not the head has the highest shadow wage in the household. The head has the highest shadow wage in 71.5% of households (60% if all oneadult households are excluded). Or put differently, of those adults living in households with two or more adults, 60% of heads but only 13% of non-heads have the highest shadow wage in their household. The identifying assumption is that while the actual value of the shadow wage influences hours worked, the relative value of the shadow wage determines household headship for a given household composition. In addition to this identifying variable, we use two other dummy variables as potential instruments: whether the head has the highest level of education in the household, and whether the head is the oldest household member. These two variables provide information on the possible status and talent of the head in relation to other household members. We carry out a test for overidentifying restrictions to check the validity of the instrument set. Results of this test (available upon request from the authors) show that for men, highest shadow wage and highest age are valid instruments, while for females highest shadow wage and highest education level are valid. In addition to these identifying variables, all the exogenous variables in the labor supply equation are also included in the first stage equation (i.e. we use two-stage least squares). Columns (3) and (6) of Table 3 provide the two-stage least squares estimates of the determinants of total hours worked for men and women respectively. For both men and women, the coefficient of headship virtually doubles; male heads now work almost 19 hours per week more relative to non-heads, while female heads work on average 16 hours more per week than 15

17 their non-head counterparts. These estimates imply a significant negative association between unobservable characteristics influencing household headship and time allocation for the head. One possibility is that of the bad draw mentioned earlier, where heads are essentially required to pick up the slack of other household members by increasing their own time at work. An empirical framework that does not allow for these intra-household dynamics does not pick up the full effect of headship on time allocation. Based on the 2SLS estimates in Table 3, the calculated wage elasticities become 0.27 and 0.21 for men and women respectively. While the elasticity for women is virtually unchanged, failure to account for the endogeneity of headship for males results in an over-estimate of the wage elasticity by 16 percent (0.32 versus 0.27). When we allow the shadow wage to vary by headship status by interacting the shadow wage with the predicted probability of being the head, we encounter point elasticities of 0.17 and 0.06 for male and female heads respectively, considerably lower than those implied by OLS (results not shown). Columns (3) and (6) of Table 4 present the 2SLS estimates for the determinants of market hours worked, and again, the coefficient for headship is 2 or 3 times the value of the OLS coefficients. The reductions in the implied point elasticities are not as great in this case. For men, the wage elasticity declines by 11 percent from 0.45 to 0.40, while for women it remains virtually unchanged. However when we include the interaction between predicted headship and the shadow wage, the coefficient of this variable is statistically significant for both men and women, and the implied elasticities for heads fall even further to 0.28 (males) and 0.29 (females) (These results are available upon request). Excluding Single Person Households 16

18 Since the intra-household allocation of time in single person households is likely to be structurally different than in multi-person households, we repeat our estimates in Tables 3 and 4 excluding these households from the sample. Table 5 presents these estimates for the determinants of total hours worked. The OLS estimates of the impact of headship on total hours is slightly lower for men and approximately the same for women, and the corresponding elasticities are 0.36 and 0.22 for men and women respectively which is higher than the 0.32 for men estimated from Table 3, and exactly the same for women in Table 3. (The relevant means for calculation of these elasticities are provided in Table A2). When the price response is allowed to vary by headship status (columns (2) and (4) of Table 5) the elasticity drops to 0.31 for men (a 14 percent decline) but remains the same for women. These are exactly the same pattern of results that were found in Table 3 for the full sample, although the levels are slightly different for men. This is because our reduced sample excludes more men than women; only 5 percent of women live in single person households compared to 13 percent for men. The 2SLS estimates in columns (3) and (6) of Table 5 also follow the pattern encountered in Table 3. The point estimates for the number of hours worked on average by heads increases by threefold, and the corresponding wage elasticity declines by 14 percent for men (from 0.36 to 0.31) but remains the same for women. Table 6 presents the determinants of market hours worked on the reduced sample and these results also mirror the pattern found in Table 4. Both the OLS and 2SLS coefficients of the effect of being head are in the same range as those found in Table 4, and the elasticities for women (both heads and non-heads) are virtually identical to those calculated from Table 4. For men, the drop in the wage elasticity as we move from OLS to 2SLS is 12 percent (0.49 to 0.43), 17

19 and the calculated elasticity for male heads from column (2) is 0.40, slightly higher than the 0.36 calculated from column (2) in Table 4. In summary, the results on the impact of headship on time allocation, and the corresponding wage elasticities, are robust to the exclusion of single person households. The only noticeable difference in the results is a slight increase in the wage elasticity for men in the select sample of individuals living in multi-person households. Sensitivity Analysis: Tobit Estimates of Market Hours Worked While only 7 percent of adults in our sample report 0 total hours worked in the reference week, 38 percent report 0 market hours worked. We have estimated the regressions reported in Tables 4 and 6 explicitly allowing for left censoring using Tobit. A summary of the coefficients of the variables of interest are presented in Table A2 and A3 of the Appendix. In the full sample of adults (i.e. including single person households), the pattern of estimates using OLS and Tobit are the same as for OLS although there is some difference in the levels, especially for women (who report more 0 market hours worked). For both men and women, heads work significantly longer hours than non-heads, and wage elasticities are lower for heads than non-heads. For example, the wage elasticities calculated from column (2) of Table A2 for male heads and non-heads respectively are 0.49 and 0.96; for women these numbers are 0.69 and 1.48 respectively. And as before, when we treat headship as endogenous, the estimated hours of work for heads increases, and the corresponding wage elasticities decline, although the decline is much larger for men (11 percent) than for women (2 percent) (see Table A2). When we exclude single person households (Table A3), we obtain the same pattern, although estimated hours worked for heads is slightly higher and associated wage elasticities are lower (for men only). 18

20 Results by Family Structure: Children versus No-Children There is a large literature that attempts to measure the impact of fertility on female labor supply. 17 Following this tradition, we present some estimates of the possible effect of the presence of children on the time allocation of household heads, hypothesizing that this relationship is likely to be different for female heads relative to male heads. As in most studies of this type, we treat demographic composition (more specifically, fertility) as exogenous, and estimate separate regressions for household with and without children (defined as individuals less than age 15). Table 7 presents a sub-set of these estimates for the determinants of total hours worked. For each set of estimates we present both OLS and IV coefficients of the effect of headship, the shadow wage, and the interaction of the two, on total hours worked. The association between total hours worked and the corresponding wage elasticity is different for male and female heads, depending on family structure. In almost all cases heads work significantly more hours than other adults, but for males, this difference is especially large in columns (2) and (4). In families without children, the IV coefficient in column (4) indicates an implausible 52 additional hours per week for male heads in families with no children. For females, the opposite is the case; large additional hours and significantly lower wage elasticities are found for female heads living in households with children. Table 8 replicates these estimates using market hours as the dependent variable, and the pattern of the results is essentially unchanged. For male heads, large additional hours worked and significantly lower wage elasticities occur in households without children, while for female heads, this occurs in households with children. (Tobit estimates are presented in the Appendix 19

21 Table A3). The literature on female headed households in developing countries has recognized that adult women may purposefully trade-off lower overall consumption for greater control of resources and decision-making power in deciding to head their own household. 18 The results presented in this section provide some indication of the cost of this decision to female heads in terms of their consumption of leisure and ability to respond to price changes. 5. Alternative Welfare Measures: Adult Health Status Thus far in our analysis of the welfare implications of headship for household heads we have focused on leisure consumption, and have found that both male and female heads consume significantly less leisure and have lower wage elasticities than other adults. The Jamaican Survey of Living Conditions (SLC) contains a health module which allows us another opportunity to compare the welfare of household heads relative to non-heads. The standard health module contained in all rounds of the SLC asks whether the respondent was sick or injured in the last 4 weeks, whether the illness or injury began prior to the last 4 weeks, and the number of days ill or sick. We use these three variables from the 1993 SLC as outcomes, and compare whether adult heads of household reported being sick more often, and for longer periods, than other adults. In addition, the 1989 SLC contained a special module on adult health, and included a question on self-reported general health status, as well as self-reports on a set of 7 activities of daily living (ADLS). 19 Based on the self-report of general health status we create a dummy variable indicating whether the individual reported themselves to be in poor or fair general health. The ADLs are used to construct two indexes. The first, less severe index, contains information on vigorous and moderate activities, walking uphill, and bending. The second, more severe index, contains information on limitations in walking one mile or 100 yards, 20

22 and daily activities. In each case, we construct a dummy variable equal to one if the individual was moderately or severely limited at performing any of the activities contained in that index. 20 Illness in previous 4 weeks In Table 9 we present means of the three illness related variables from SLC 1993, by sex and headship status. For both males and females, heads reported a higher incidence of illness in the last 4 weeks relative to non-heads. For males, the difference in reported incidence is versus 0.081, while for females the reported incidence rates are and for heads and non-heads respectively. The same pattern emerges for family heads (i.e. excluding single person households). Heads also tend to report more sick days than non-heads (14.2 versus 11.8 in the case of men, for example). Since heads have different demographic profiles than non-heads, Table 10 presents results of a descriptive regression where we estimate the net difference in mean incidence rates between heads and non-heads, controlling for age, education, region of residence, and a measure of household resources. 21 Row 1 of Table 10 shows that the net difference in incidence of sickness between heads and non-heads is statistically insignificant. The only significant difference is found in the propensity for the sickness to begin earlier in the case of male heads relative to non-heads. Activities of Daily Living and General Health Status Table 11 presents means by headship status for the incidence of fair or poor general health status, as well as the incidence of reporting moderate or severe limitation in each of the two ADL indexes. For all three health status measures, heads are more likely to report being in worse 21

23 health status than non-heads. Note also that the levels of incidence are higher for women in general, a result that is consistent with previous work on Jamaica and in other countries. After controlling for observable household and individual characteristics, the mean differences in incidence rates are eliminated for the ADL measures, but persist for self-reported general health (Table 12). In the case of general health, after controlling for observables, male heads are now less likely to report fair or poor general health relative to non-heads, with the difference being 2 percentage points (columns (1) and (2) of Table 12). For women on the other hand, the higher incidence of fair/poor general health among heads persists even after controlling for observed differences in age, education, region of residence, and access to resources. The mean difference in incidence is estimated to be 2-3 percentage points from columns (3) and (4) of Table 12. In conclusion, the results from the analysis of adult health status show that heads of household are in poorer health relative to other adults, but that this difference is due to other factors (primarily age) and not headship per se. The one exception is for self-reported general health, where even after controlling for observables, differences continue to persist between heads and non-heads. For males, heads report a lower incidence of fair/poor general health relative to non-heads, while female heads continue to have higher incidences of fair/poor general health relative to female non-heads. 6. Discussion and Conclusions The household head plays an important role in several theoretical models of household decisionmaking, while empirically, the characteristics of the head have been shown to be an important determinant of individual household member well-being such as children s health and 22

24 educational status. Furthermore, the literature on female headed households has pointed out that despite their relative poverty, female headed households may be formed voluntarily, an adult woman opting for lower overall well-being in order to gain decision-making freedom and authority over the allocation of goods within the households. Interestingly, none of these various strands of research have considered the impact of headship on the household head herself. Using data on time allocation from Jamaica, we show that household heads consume significantly less leisure than other adults, and have significantly lower wage elasticities. When headship is treated as endogenous, this cost of headship rises, especially for male heads, for whom wage elasticities decline by approximately 15 percent. The size of this cost for male and female heads depends on family structure. For male heads, mean hours are higher and wage elasticities lower in households without young children, while for female heads, mean hours worked are higher and elasticities lower in households with children. This result is especially important given that female headed households are much more likely to have resident children, both in Jamaica and worldwide. We supplement our analysis on the welfare of heads by comparing several dimensions of adult health status, including incidence of recent illness, self-reported general health, and physical health status as captured by ADLS. In all these dimensions of health status, heads are in poorer health than other non-head adults. However, when controlling for observed differences in household and individual characteristics, these differences disappear, except for general health for women, where the heads continue to report incidence rates 14 percentage points higher than those for non-heads. (Interestingly, for men, the controlled means indicate male heads to be in better general health than non-heads.) 23

25 Our results have several implications for the general literature on household formation, headship, and intra-household resource allocation and labor supply. First, we address an important gap in the previous literature by illustrating the potential burden that comes from heading a household. Recent thinking on household formation, especially in relation to female headship, has stressed the notion that female headship is endogenous, and that women may choose to head households to achieve greater decision-making power. This thinking has ignored the potential welfare consequences of headship for the head herself. Results from Jamaica show that these consequences are non-trivial, and that the gains due to increased authority may come at the expense of private consumption of heads. Second, our empirical methodology of allowing for endogeneity of household headship reveals that failure to do so can bias estimates of intra-household allocation decisions (in our case--decisions on time use). In our data, treating headship as endogenous leads to a 15 percent reduction in the wage elasticity for men, and a 2-3 fold increase in the estimated impact of headship on total hours worked. These results indicate that a similar methodology should be applied to other studies that attempt to measure the impact of family structure on household consumption decisions. To this end, variables that measure the social status and productivity of the head relative to other household members, such as the variables used here, may be fruitful instruments. Finally, an interesting and important observation is the impact on family structure of raising women s productivity and wages, especially relative to men, as the shadow wage is an important correlate of household headship. Endnotes 1. Haddad L, Hoddinott J, Alderman H (1997) Introduction: the scope of intrahousehold resource allocation issues. In: Haddad L, Hoddinott J, Alderman H (eds.) Intrahousehold 24

26 resource allocation in developing countries. Baltimore, Johns Hopkins; Sen, A (1983) Economics and the family. Asian Development Review 1: For sub-saharan Africa Lloyd CB, Blanc AK (1996) Children s schooling in Sub- Saharan Africa: the role of fathers, mothers, and others. Population Development Review 11 (3): ; in Jamaica Handa S (1996) Mother s education and child attainment: a test of the bargaining power hypothesis. Oxford Bulletin of Economics and Statistics 58 (1): For Kenya, Kennedy E, Peters P (1992) Household food security and child nutrition: the interaction of income and gender of household head. World Development 20 (8): For Jamaica, Horton S, Miller B(1989) The effect of gender of household head on food expenditure: Evidence from low-income households in Jamaica. University of Toronto, (mimeo.); Handa S (1996) Expenditure behavior and children s welfare: an analysis of female headed households in Jamaica. Journal of Development Economics 50 (1): In Cote d Ivoire, Grootaert C, Kanbur R, Oh G (1997) The dynamics of welfare gains and losses: an African case study. Journal of Development Studies 33 (5): Evenson Robert (1978) Introduction to Symposium on household economics. Philippine Economic Journal 36: As Killingsworth M, Heckman J (1986), Female Labor Supply: A Survey, Chapter 2 in Ashenfelter and Layard (1986)Handbook of labor Economics Vol.1, Amsterdam: North-Holland point out, the time allocation approach does not reveal additional predictions on market work that cannot be derived within the conventional individual labor supply model. However the former model does provide insights on time allocation to domestic work, as well as the intrahousehold allocation of leisure. 8. See Chapters 1 & 2 in Ashenfelter O, Layard R (eds.) (Note 7 above) for a review. In a recent paper, Angrist J, Evans W (1998) Children and their Parents Labor Supply: Evidence from Exogenous Variation in Family Size. American Economic Review 88(3): use exogenous variations in fertility to identify the impact of fertility on female labor supply. Their IV estimates are smaller than conventional OLS, and they find almost no effect of fertility on labor supply for American women with college degrees. 9. Behrman, Jere & Barbara Wolfe, Labor force participation and earnings determinants for women in the special conditions of developing countries. Journal of Development Economics 15: ; Malathy R (1994), Education and Women s Time Allocation to Non-Market Work in an Urban Setting in India. Economic Development & Cultural Change 42(4): ; Skoufias E (1993) Labor market opportunities and intrafamily time allocation in rural households in South Asia. Journal of Development Economics 40:

27 10. Lloyd CB, Gage-Brandon AJ (1993) Women s role in maintaining households: family welfare and sexual inequality in Ghana. Population Studies, 47: Handa S (1996b) The determinants of female headship in Jamaica: results from a structural model. Economic Development and Cultural Change, 44 (4): Lundberg SR, Pollak R, Wales T (1997) Do husbands and wives pool their resources. Journal of Human Resources 32: These data are described in Planning Institute of Jamaica (1995) Jamaica Survey of Living Conditions: Report Kingston: Planning Institute of Jamaica and The Statistical Institute of Jamaica. 14. Handa Sudhanshu (1994) Gender, Headship and Intra-Household Resource Allocation. World Development, 22(10), Handa, Sudhanshu (1996) Expenditure Behavior and Children s Welfare: an analysis of Female Headed Households in Jamaica, Vol.50(1): McElroy, Marjorie (1985) The joint determination of household membership and market work: The case of young men. Journal of Labor Economics, Vol. 3(3): For example, in a household that had a poor allocation of (unobservable) talents and abilities among its non-head members, the head would be required to work longer hours than the observed distribution of household characteristics (such as age and education) would predict. 17. See Ashenfelter and Layard, note Female headed households are more likely to have children than male headed households, and Jamaica is no exception. In our sample of households, 67 percent of FHHs have at least one child under 15, compared to 50 percent for male headed households. 19. The 7 activities are bending, walking 1 mile, walking 100 yards, climbing stairs, moving a table (moderate activities), lifting heavy objects (vigorous activities), and brushing teeth General health is coded into 5 categories: excellent, very good, good, fair, poor. 20. The ADL information from this survey have previously been used by Handa & Neitzert (1998) Chronic Illness and Retirement in Jamaica. LSMS Working Paper No. 131, Washington DC, The World Bank, and Strauss J, Gertler P, Rahman O, Fox K (1993) Gender and life-cycle differentials in the patterns and determinants of adult health. Journal of Human Resources 28(4): The two ADL indexes are described in more detail in Handa & Neitzert. 21. We control for household resources with log household per capita consumption expenditures, since income is not available in the SLC, and consumption is thought to better approximate permanent income. Consumption expenditures are instrumented using housing characteristics including access to electricity and water, and property income. 26