S E IZA DP No. 4879 I R E S R E The Dynamics of Women’s Labour Supply in P Developing Countries A P N Sonia Bhalotra O Marcela Umaña-Aponte I S S U C S I April 2010 D Forschungsinstitut zur Zukunft der Arbeit Institute for the Study of Labor The Dynamics of Women’s Labour Supply in Developing Countries Sonia Bhalotra University of Bristol (Economics, CMPO and Townsend Centre), University of Oxford (CSAE and QEH), CHILD and IZA Marcela Umaña-Aponte University of Bristol Discussion Paper No. 4879 April 2010 IZA P.O. Box 7240 53072 Bonn Germany Phone: +49-228-3894-0 Fax: +49-228-3894-180 E-mail: [email protected] Any opinions expressed here are those of the author(s) and not those of IZA. Research published in this series may include views on policy, but the institute itself takes no institutional policy positions. The Institute for the Study of Labor (IZA) in Bonn is a local and virtual international research center and a place of communication between science, politics and business. IZA is an independent nonprofit organization supported by Deutsche Post Foundation. The center is associated with the University of Bonn and offers a stimulating research environment through its international network, workshops and conferences, data service, project support, research visits and doctoral program. IZA engages in (i) original and internationally competitive research in all fields of labor economics, (ii) development of policy concepts, and (iii) dissemination of research results and concepts to the interested public. IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion. Citation of such a paper should account for its provisional character. A revised version may be available directly from the author. IZA Discussion Paper No. 4879 April 2010 ABSTRACT The Dynamics of Women’s Labour Supply in Developing Countries* This paper investigates cyclicality in women’s labour supply motivated by the hypothesis that it contributes to smoothing household consumption in environments characterized by income volatility. We use comparable individual data on about 1.1 million women in 63 developing and transition countries merged with country-level panel data on GDP during 1986-2006. The scope of these data is unprecedented in the small but growing literature on labour markets in developing countries. We find that the within-country relationship of women’s employment and income is, on average, negative in Asia and Latin America but positive in Africa. We suggest that amongst reasons why African women behave differently are that the conventional family structure with income pooling is less the norm, there are fewer opportunities for paid employment, and aggregate income shocks are more closely tied to rainfall variation. The findings are robust to controls for country-specific trends and potentially correlated shocks. In Asia and Latin America, characteristics that strengthen counter-cyclical responses include low education, being married, being married to men with low education, low wealth, no landownings, rural residence and fertility. These findings suggest that insurance motives underpin the dynamics of women’s work participation. Examination of cyclicality in the distribution of employment across types suggests that recessions in every region are associated with a rise in self-employment amongst women. In Asia and Latin America, there is a parallel rise in paid employment and a sharp drop in non-employment. In Africa, there is a decline in paid employment which overwhelms the rise in self-employment and this is how total employment comes to decline. The results have potentially important implications for understanding labour markets, fertility timing and child outcomes. JEL Classification: J22, J13 Keywords: insurance, women’s labour supply, added worker effect, business cycles, dynamics, Africa, Asia, Latin America Corresponding author: Sonia Bhalotra Department of Economics University of Bristol 8 Woodland Road Bristol BS8 1TN United Kingdom E-mail: [email protected] * This paper has benefited from presentation at the CSAE Oxford conference (March 2009), the IZA/World Bank Employment and Development conference (Bonn, May 2009), the Gender and Development Department of the World Bank (DC, June 2009), the European Society of Population Economics conference (June 2009), the Regulating for Decent Work conference at the ILO (Geneva, July 2009) and CMPO (Bristol, July 2009). We are grateful to Alex Barr for excellent assistance with setting up the data and to David Newhouse and Jon Temple for helpful comments. 1 Introduction Income volatility is endemic in poor countries (Pritchett; 2000; Koren and Tenreyro; 2007). During 1960-99, the median of the standard deviation of annual growth rates was more than three times higher for low-income countries than for OECD member countries (Malik and Temple; 2009). Governments in these countries often face severe fiscal constraints (Fallon and Lucas; 2002) and the tightening of these constraints during economic downturns often leads to cuts in state social spending (Lustig and Walton; 1998; Woo; 2005; Bhalotra; 2009; Loayza et al.; 2007) in contrast to the tendency in richer countries for social spending to rise to smooth over potential dips in household consumption (Lane; 2003). Poor households have limited access to credit (Banerjee et al.; 2007) and formal insurance mechanisms such as unemployment benefits (van Ours and Vodopivec; 2006). As a result, even if income shocks are transitory, they often have severe and irreversible consequences. They may throw families into poverty traps, for example, through the co-dependence of wage income and adult health (e.g. Ray; 1998; Dasgupta; 1997; Eriksson et al.; 2005), raise the death toll amongst newborns (Baird et al.; 2007; Bhalotra; 2009) or leave the survivors amongst exposed children scarred with the prospect of greater later-life morbidity and lower educational attainment (van den Berg et al.; 2006; Ferreira and Schady; 2009). A vast literature describes how households attempt to cope with idiosyncratic shocks through asset decumulation or reliance upon informal insurance networks (Morduch; 1995; Sk- oufias;2003). Howevermanyhouseholdshavenoassetstosellandriskpoolingarrangementsare challenged by aggregate shocks that impose covariant risks on members (e.g. Townsend; 1995). Households may then fall back on the one asset they have, which is their labour. Understand- ing the dynamics of employment is therefore key to understanding the dynamics of poverty. We focus upon women’s labour supply, investigating the extent to which it rises in economic downturns and reverts in upturns. The hypothesis that counter-cyclical employment performs an insurance function is further investigated by allowing interactions between aggregate shocks and indicators of liquidity constraints at the household level. Even if women desire to increase their work participation in downturns, they may be limited by a decline in labour demand. We investigate transitions between unemployment and types of employment, illustrating the role of self-employment. This exercise also illuminates the differences in employment behaviour that we find across the three developing country regions. The stylized fact for richer countries is that women’s labour supply is pro-cyclical, rising in upturns when the offered wage is more likely to exceed the opportunity cost of women’s time; for example, see Joshi (1981) for the UK, Killingsworth and Heckman (1987) for the US and Darby et al. (2001) for other OECD countries. There is no stylized fact regarding the cyclicality of women’s work in poorer countries. There is some evidence from historical data that women’s 2 employment rose during periods of economic crises and industrial restructuring in 19th century England (Anderson; 1974; Scott and Tilly; 1978, p. 144) and New England (Hareven; 1982, 1990; Lamphere; 1987). Scott and Tilly (1978, p. 144) argue that “the reallocation of women’s labourbetweenproductiveandreproductiveworkwaspartofawidespreadeconomicadaptation ofhouseholdsthatemergedtomaintainhouseholdlivingstandardsunderrestructuredeconomic conditions”. There is some compelling evidence that the economic crisis in 1997/8 in Indonesia threw women into employment, even as male employment rates fell, and that the added women workers were primarily in the informal sector (Thomas et al.; 2003, also see Section 5 below). However, in terms of the proportional decline in average hourly wages, the Indonesian crisis was as severe as the Great Depression of the 1930s in the US or the collapse of Soviet Union in the early 1990s (Fallon and Lucas; 2002) and it is unclear how well these findings would generalise to the case of smaller shocks or alternative institutional settings. We use a micro-dataset of unprecedented scope in this literature that contains compa- rable information on 1.1 million ever-married women (and a further 0.1 million never-married women)ofage20-49interviewedin129surveyroundsconductedin43countriesacross21years, 1986-2006. The micro-data are merged by country and year of interview with cross-country panel data on income (GDP) and other relevant macroeconomic variables. The macro-panel is exploited to create country-year specific indicators of the state of the business cycle. The micro-data are exploited to investigate heterogeneity in the response parameters predicted by theory, for example, with respect to wealth and the stage of the lifecycle (Gruber and Cullen; 1996; Attanasio et al.; 2005). We report a range of specifications designed to balance the trade- off between robustness and efficiency that has been emphasised in the recent literature on growth econometrics (Durlauf et al.; 2005) and explore robustness of the estimates to controls for trended unobservables and country-year varying shocks. The results show that economic fluctuations are mirrored in significant fluctuations in women’s work. The average tendency is for African women to fall out of employment in recessions, presumably to a very vulnerable position. In contrast, Asian and Latin American women’s employment increases in recessions and heterogeneity in the response parameter by a number of individual characteristics indicates that this is part of a household coping strategy. We suggest that amongst reasons that African women behave differently are that the conven- tionalfamilystructurewithincomepoolingislessthenorm(Udry;1996), opportunitiesforpaid employment are limited and aggregate income variation is more closely tied to rainfall variation with the implication that income shocks strike the sector where women most naturally seek employment (see Section 5.1). Economic fluctuations generate considerable churning across job types, which average employment rates mask. Downturns in Asia and Latin America are associated with women moving from non-employment to both self employment and paid em- ployment. In Africa, the important shifts are out of paid employment and into self-employment 3 and non-employment. Previous cross-sectional research shows that women’s labour supply tends to be U-shaped in economic development (Schultz; 1988; Goldin; 1995; Mammen and Paxson; 2000). The upturn at higher levels of development is thought to be driven by gains in women’s education and the emergence, through structural change, of jobs “suitable” to women. The upturn at low levels of development arises because families often need the income from women’s work, and their participation is facilitated by the prevalence of family farms and enterprises. This paper presents a dynamic and more disaggregate picture of life at the low end. On average about 50% of women work in our sample of developing countries. The 50% who do not work include those who are wealthier and/or married to educated men (income effect) and those for whom the opportunity cost of work is high because of high fertility (substitution effect). Our estimates show that transitory participation induced by the business cycle brings in a more diverse set of women, including asset-poor, less educated, younger women with children under the age of five. This invites consideration of the effects that this has upon the timing of fertility and of investments in children. Studies of poverty dynamics have paid limited attention to the role of macroeconomic fluc- tuations (Baulch and Hoddinott; 2000; Ravallion; 2001) and studies of the effects of macroe- conomic crises on poverty in poorer countries have paid limited attention to the mediating influence of labour market dynamics (Fallon and Lucas; 2002; Skoufias; 2003; Conceic¸a˜o et al.; 2009). This is topical given the ongoing world recession, thought to be the worst since the Great Depression. Growth in developing countries was 1.2 percent in 2009, compared with 5.6 1 percent in 2008 (World Bank; 2010) . In fact, developing countries have been routinely ravaged by crises stimulated by natural disasters, wars, debt, commodity price shocks, financial collapse and the spread of HIV/AIDS. This paper is therefore more broadly relevant to growing interest in the effects of income volatility in poor countries (Koren and Tenreyro; 2007) and especially its effects on human capital investment at critical periods of childhood (Almond; 2006; Cunha and Heckman; 2007; Ferreira and Schady; 2009; van den Berg et al.; 2009), in the transmission of which women’s labour supply is an under-studied mechanism (Basu and Basu; 1991; Rose; 2000; Bhalotra; 2009; Ag´enor and Ag´enor; 2009). 2 Background Themotivatinghypothesisisrootedinaliteratureontheaddedworkereffect(AWE)thatorigi- natedaroundtheGreatDepression(seeHumphrey;1940;Woytinsky;1940). Thiscastsmarried 1Uncertainty concerning the likely impact on women’s labour market status is evident in the media, not only for developing countries but also in, for example, the UK and the US (Appendix B includes some relevant quotes from the media) 4 women as secondary workers who temporarily increase labour supply when their husbands suf- fer unemployment. The underlying theory is developed in Ashenfelter (1980), Heckman and MaCurdy (1980), Lundberg (1985), Ehrenberg and Smith (2003) and Borjas (2005). In the simple static model, women raise labour supply following a transitory dip in family income if their leisure is a normal good and this effect is enhanced if the increased non-market time of the husband lowers the opportunity cost of market work for the wife through substitution in home production. Given evidence that cross-substitution effects are typically small, the AWE relies upon wealth effects. In a lifecycle setting, as long as the income loss from unemployment is small relative to the husband’s lifetime earnings, AWE effects are expected to be small and con- centrated amongst families that face liquidity constraints or fixed consumption commitments (Mincer;1962;Lundberg;1985). Unemploymentinsuranceinrichercountrieswilltendtocrowd out any need for wives to provide insurance (Gruber and Cullen; 1996). Consistent with these predictions, studies set in the US and UK find small if any added worker effects (Stephens; 2002). In aggregate data, the discouraged worker effect tends to dominate and women’s work is pro-cyclical (see references in the previous section). As indicated earlier, there is typically no unemployment insurance in developing countries and a vast fraction of households face subsistence constraints that they are not equipped to borrowtosmoothover. Inthissetting, wemayexpecttoseedominantwealtheffectsand, inthe aggregate, counter-cyclical labour supply, see, for example, Barzel and McDonald (1973) and Stern(1984),whoshowtheimportanceofassetlevelsandsubsistenceconstraintsindetermining the shape of the labour supply curve. The tendency for women in poorer countries to act as secondary workers is likely to be strengthened by traditional family structures and low levels of education and skill amongst women. The informal sector that characterises developing countries facilitates transitory work spells as entry and exit barriers and search costs are low (Basu et al.; 2000). The literature on household insurance mechanisms in developing countries has considered child labour as an insurance mechanism (Jacoby and Skoufias; 1997; Skoufias; 2003) but it contains relatively limited investigation of the role of changes in women’s labour supply. 3 Data and Descriptive Statistics 3.1 Data The micro-data are assimilated from 153 Demographic and Health Surveys (DHS) conducted using a similar questionnaire in 63 developing countries between 1986 and 2006. There are no similar cross-country micro-data on employment and the DHS data have not been used for this purpose before. The potential sample contains 1,777,063 observations on women’s work, 5 2 education and demographics . Countries and survey years in the sample are in Table 1. For 20 countries, only one survey is available. There are 13 countries with two surveys, 17 with three, 9 with four and 4 with five. We group the countries into five regions, sub-Saharan Africa (henceforth Africa), Asia, Latin America, the Middle East and North Africa and the former USSR. About 33.5% of women in the sample are located in sub-Saharan Africa, 33.8% in Asia and 22.8% in Latin America. Another 8% are in the Middle East and North Africa and 2% in the former USSR. Although we present some results for the latter two, the discussion will refer to the 90% of women in the three developing country regions and these are the regions that are pooled for the estimation. We construct nine sub-regions within these three regions to capture cultural and institutional heterogeneity (Table 1). The surveys interview women of reproductive age, typically 15-49 years. We keep women aged 20-49. Of the 153 surveys, 124 interview a representative sample of women. These include all countries in Africa, Latin America and ex-USSR. In the other 29 surveys, which pertain to 15 countries in Asia and the Middle East, the sample was restricted to ever-married women (Table 1). The main analysis is on the sample of ever-married women although we allow heterogeneity as a function of current marital status and fertility. There are no data on hours of workorwages. Wagedatafordevelopingcountriesarenotveryusefulgiventhatunpaidworkor self-employment are widespread. The data contain information on the education of the woman and her partner, and education may be thought of as an indicator of the permanent component of individual wages. The business cycle variation that we analyse creates transitory variation in wages and the specifications estimated will allow the impact of this on women’s employment to vary with her education and that of her husband. Information on years of education is used to construct dummy variables indicating whether the individual is uneducated (none), has some education but less than secondary (some), or has completed secondary or higher education (high). A household wealth index is constructed using data on ownership of assets and this is used to create indicators for the quartile of the wealth distribution that the household falls into (see the data appendix). The microdata are merged by country and year of interview with country panel data on GDPpercapitainconstantprices, inflation, termsoftrade, indicatorsoffinancialdevelopment, rainfall and civil conflict. Definitions and sources of these variables are in the data appendix. 2See www.measuredhs.com where the data, documentation and reports are available by country. We have harmonisedthedataacrosscountriesandsurveyyears. Forexample,wehaveadjustedthedatesintheEthiopia andNepalsurveysforthefactthattheyfollowtheJulianandtheNepalicalendar(VikramSamwat),respectively. Details are in the data appendix. 6 3.2 Descriptive Statistics This section profile’s women’s employment in developing countries, providing what would ap- pear to be the first broadbrush sketch representative of developing countries. The distribution of women in the sample by country and region is in Table 1 and distributions by sector of employment and individual characteristics are in the data appendix. Summary statistics for all variables analysed are in Table A3. Averaging across the sample, about 49% of women work compared with about 98% of men, so there is clearly less room for manoeuvre amongst men. The employment rate of men does not vary much across regions, except that it is lower in the ex-USSR (Table A1). Women’s employment exhibits considerable variation by region and characteristics (Tables 2 and A1). The employment rate of women aged 20-49 is 64% in Africa, 43% in Asia and 50% in Latin America. Self employment dominates paid employment amongst African women but the reverse is the case in Asia and Latin America (Table A2). The employment rate is U-shaped in women’s education but decreases monotonically in their partner’s education (Table 2). The between-country variation shows a significantly negative relation between women’s work and GDP (Figure 1). Plots of the between-country relationship by continent show that African countries drive the overall negative relationship, the relationship being negative but flatter in 3 Asia and positive in Latin America- and also in the Middle East and ex-USSR (Figure 2) . Pooling the within and between country variation (Table A1) we see a negative relation between women’s employment and each of GDP and household wealth. A greater percentage of women are employed in recessions as opposed to booms, 54% against 49%, while men’s employment is similar in the two regimes (Tables 2 and A1). The relationship with GDP is sharpest at the low end - the bottom 20% of country-year observations in the GDP distribution are associated with sharply higher rates of employment (66%) than the rest of the distribution. The relationship with household wealth is sharpest at the high end- women in the top quartile of the household wealth distribution are much less likely to work (46%) than elsewhere in the distribution. Women’s employment rates are 5 points higher in rural than in urban areas. Overall, the correlations are consistent with the hypothesis that women are more likely to work in poorer circumstances (indicated by partner’s education, wealth, GDP, rural), and we know that incomes are more volatile in poorer circumstances. The employment rate of unmarried women is 18 points higher than that of married women, consistent with married women being secondary workers. Amongst married women, employ- ment is 4 points lower if there is at least one child under the age of five in the household, reflecting the widely established tendency for the opportunity cost of work to rise with fertil- 3Following Boserup (1970) seminal work describing how industrialization alters women’s traditional social andproductiveroles,theliteraturedocumentsatendencyforwomen’sparticipationtodeclinerelativetomen’s (Scott; 1986). 7 4 ity Employment rates are increasing in age (cohort), possibly reflecting the fact that younger women are more likely to have young children at home. In Figure 3 we plot employment rates for single mothers, single non-mothers and (currently) married mothers against GDP. Single mothers are the most active group across income levels, consistent with their bearing the pri- mary responsibility for dependents. It is striking that married women with children (in this casenotnecessarilyundertheageoffive)work morethansinglewomenatlowbutnotathigher levels of GDP. This is indicative of their employment being distress-driven. The correlations in these data incorporate unobserved heterogeneity. The more formal analysis to follow uses country and time variation in income to identify the extent to which the dynamics of women’s work is driven by distress. 4 Model Specification and Results We model women’s employment as a function of business cycle variation which, it is implicit, creates demand induced declines in the income of primary earners. Studies of the added worker effect have tended to model women’s work participation as a function of the labour market status of their husbands (Lundberg; 1985; Stephens; 2002) but this encounters endogeneity and selection problems (Gruber and Cullen; 1996) which the modelling of women’s employment as a function of the business cycle avoids.. Our approach has the following further advantages. First, it accommodates changes in earnings across multiple earners, which is pertinent given the extended family structures in poorer countries. A father-in-law or a brother-in-law may well be the main earner and if, say, recessions induce young men to enter the labour market early or elderly men to delay retirement then in the antecedent model their labour supplies would appear as omitted variables. Second, our approach captures the impact on women’s work not only of unemployment amongst household members but also of under-employment and wage decline. This is relevant in poorer countries where unemployment is unaffordable and hence ill-defined, and wages tend to carry most of the impact of economic crises (e.g. Thomas et al.; 2003). Also, using US data, Maloney (1987) shows that it is only once employment constraints on the desired labour supply of both husband and wife are allowed that an added worker effect emerges. A third reason to prefer the direct modelling of aggregate income variation is that this is empirically relevant given the considerable variation in developing countries (Section 1). It is important to emphasise that we only seek to answer the question that the reduced form permits 4The labour economics literature focuses on substitution effects created by fertility. In developing countries, it is important to recognise that poor health and nutrition have productivity effects (Fogel; 1994; Ray; 1998, chapter8)which, setagainstthelabourintensityofhomeproductionmayprohibitworkparticipation. Women may have to spend years in cycles of reproduction especially in areas of high mortality (Dasgupta; 1993, e.g.) or many hours per day collecting firewood and water (Cain; 1986) or walking their children to faraway clinics (Mwabu et al.; 1993). 8