In this chapter I have presented an attempt to estimate the effect of rural out-migration on rural wages. I find a strong positive and robust effect of rural out-migration on rural wages in Brazil during the period 1991-2000. Using a cohort analysis my results suggest that rural out-migration flows between 1991 and 2000 have increased rural wages in Brazil by 6.5%. One concern is that rural out-migration may be accompanied with changes in the workforce composition in rural areas since those who migrate do not constitute a random sample of the initial rural population. I find that changes in the workforce composition do account for some of the wage increase due to larger out-migration rates. Controlling for observable measures of workforce composition such as educational attainment and gender composition, I find that the wage effect of rural out-migration flows between 1991 and 2000 drops from 6.5% to about 3%. Due to data limitation, the analysis in this chapter did not measure the short-run effect of rural out-migration on wages by taking into account dynamic adjustments in physical capital. Future research in this areas could she light on which rural population groups lose or gain in the short-run as rural out-migration increases.Because of their dependence on rain-fed agriculture a large proportion of households in developing countries are particularly vulnerable to rainfall shocks.
Moreover the usual mechanisms for smoothing income or consumption may be missing or limited in such economies. In addition,flower bucket since shocks such as weather and pests are likely to affect the income of all households, they cannot be insured locally through non-market mechanisms. Households and individuals inability to transfer resources across time and states of the nature may lead them to adopt coping strategies that are detrimental to asset and human capital accumulation. For instance, a negative income shock may lead households to draw on monetary or liquid savings to smooth income and consumption. Households who experience a negative shock can also smooth income by increasing labor supply or reducing spending in some food or investment groups. Increasing labor supply may entail putting children to work, while the need to reduce spending can result in less health and schooling investment. Jensen finds that among households who experienced a rainfall shock in Cote d’Ivoire, enrollment rates and child growth drop considerably relative to one year before the shock. Jacoby and Skoufias find that household income fluctuations in India lead to year-to-year variations in school attendance. Beegle et al. find evidence of increased child labor following crop losses using longitudinal household data from Tanzania. Examples include Funkhouser for the debt crisis in Costa Rica, Thomas et al. who examines household response to the financial crisis in Indonesia and Rucci for the Argentine peso crisis.
However considerably less attention has been devoted to the medium to long-run consequences of income shocks on children’s schooling in developing countries. In the presence of state dependence, an income shock may have permanent effect on a children schooling. De Janvry et al. find strong evidence of state dependence in child school enrollment using a panel of households in rural Mexico. I contribute to this literature in this paper by examining to which extent short-run drops in enrollment rates affect medium-run enrollment decisions and years of education using a period of drought in Southern Africa. This paper is related to Meng and Qian’s analysis of the long term link between famine and educational attainment in China; but it also differs in several ways. The two papers have different notion of “long term”. While Meng and Qian analyze the impact of the 1959-1961 Famine in China over 30 years after, my medium to long-run analysis is carried out within the 10 years following the droughts. This allows me to observe individuals before they have completed their schooling which, in turn, permits an analysis of their school enrollment decisions adding to our understanding of their educational attainment. Following this approach, I show that individuals exposed to the droughts tend to stay in school at older ages, and that this leads to partial catch-up in educational attainment, especially in regions where the intensity of the shock was not too high. The policy implication of this finding is that simple means testing could help target policies aimed an dampening the negative effect of agricultural shocks. In 1991/1992 and 1994/1995, the southern region of Africa experienced two major droughts.
The 1992 drought was qualified as one of the most devastating droughts in the region and followed a 60-year low rains and over two million cattle lost. Figure 3.1 shows the logarithm of the ratio between rainfall and its mean between 1940 and 1995. The graph confirms that during this 26 year period rainfall was lowest in 1992 and 1995. In 1991/1992 rainfall was about 65.8 per cent of average rainfall in the sample period, while in 1994/1995 rainfall was about 66.5 of average rainfall. These low rains have had a large negative impact on food production. Figure 3.2 shows an index of food production in Zambia between 1960 and 2005 and its deviation from a quartic time trend. In 1991/1992 food production was 14 per cent below the trend. While 1992/1993 and 1993/1994 were relatively good years, in 1994/1995 food production dropped by 8 per cent relative to the estimated production. 2 . In this paper I use data from Zambia to explore the short-run and medium-run consequences of rainfall shocks on children’s schooling. I use three cross-sections of the Zambian Demography and Health Survey, one collected during the first months of the 1991/1992 rainy season and two other surveys in 1996 and 2001/2002, to investigate the impact of the droughts on school enrollment and years of schooling of children exposed to the drought. A major drawback of the ZDHS is the absence of measures of income and consumption,square flower bucket or information on children’s time allocation or work inside or outside of the household. Thus I focus on a reduced form analysis where I compare the schooling of school aged children affected by the drought with children of the same age-group interviewed before the drought. I control for trends in schooling using enrollment and years of schooling of older youth and adults. The estimates using this strategy are biased if other aggregate shocks correlated with schooling took place during the same period. To address this issue, I employ a triple difference strategy to confirm that the effects on schooling can be attributed to the drought. I use rainfall data from actual rainfall gauges and exploit variation across provinces in the intensity of rainfall deficit during the peak of the drought. My triple difference approach consists in comparing differences in schooling across highly and moderately affected provinces between school-aged children affected by the drought and children of the same age-group interviewed before the drought. I find that exposure to the drought reduced enrollment rates by 10 percentage points and years of schooling by 8 percentage points in the short-run. I also find some evidence of partial catch-up in the medium-run in provinces moderately affected by the drought which suggests that children exposed to the drought remained in school at older ages. However, in the provinces most affected by the drought, I find no evidence of such accumulation of delayed education.
Within such provinces, young children who were in school during the drought were up to 7 percentage points less likely to be enrolled in school five to six years after the drought. Given the existing literature, the medium-run consequences on delayed entry in the job market, forgone earnings, lower wages might be large. These findings have important policy implications. They suggest that technologies to reduce the impact of rainfall shocks and safety nets may have large benefits in reducing delays and increasing the rate of human capital accumulation. Moreover education policies should target regions and individuals exposed to agricultural or income shocks in order to limit drops in enrollment rates and facilitate the return of students who temporarily left school. The remainder of this chapter is organized as follows. In section 3.2 I develop a simple dynamic model of income shock and investment in school to motivate the empirical analysis. Section 3.3 describes the timing of the data collection and the rainfall season in Zambia and some basic summary statistics. Section 3.4 explains the empirical strategy, and section 3.5 presents the main results and some robustness checks. Section 3.6 concludes. The onset of the rainy season in Zambia is normally during October or November. Rains are usually recorded up to March or April of the following year. Since I am using the 1992 ZDHS to control for an age-group’s pre-drought schooling, my estimates of the impact of the 1991/1992 and 1994/1995 droughts will be biased if, for instance, households started reducing investment in schooling during the first months of the raining season. My estimates will also be biased if data collection occurred too early relative to the raining season for households to be affected and to adjust their schooling decisions. As a consequence it is important to understand the timing of the ZDHS data collections relative to the onset and offset of the raining season in Zambia; the diagram below provides a basis for this. Data collection of the ZDHS-1992 was carried out between January and May 1992 meaning that the issue that households may have started reducing investment in schooling during the 1991/1992 raining season. In the analysis below I present a robustness check to judge the severity of the concern. The robustness of the results are tested by restricting the ZDHS-1992 sample to households surveyed during the 1991/1992 raining season, period during which the impact of the drought on harvests was not yet felt. On the other hand, the ZDHS-1996 data was collected between July and December 1996, so 3 months after the end of the poor 1994/1995 raining season suggesting that I should be able to capture the impact of the two droughts on schooling. Primary and secondary education in Zambia is divided in three levels : Primary education , Junior Secondary , and Upper Secondary. My analysis focus on school enrollment, and years of schooling for individuals 6 to 40 years old living in rural areas. I first present some graphical analysis of educational attaintment before and after the drought to motivate the empirical strategy. Figures 3.3 to 3.6 report school enrollment rates and average years of schooling by age for the ZDHS in 1996 and 1992. The general patterns reported here are similar using non-parametric regression of enrollment rate and years of schooling on age. Figures 3.3 to 3.4 show enrollment rates in the full sample and for males and females separately. Enrollment status is only recoded for residing household members between 6 to 24 years old. Figure 3.3 shows that for individuals from 6 to 13 years old, current enrollment in school is higher in 1992 relative to 1996. For individuals ages 15 to 19 years there is a less clear ranking of enrollment rate between individuals interviewed in 1992 and 1996. For individuals 20 to 24 years old, the difference in enrollment rates in 1992 relative to 1996 is much smaller. Turning to years of schooling, figures 3.5 to 3.6 show years of schooling in the full sample, and for males and females separately. For individuals 6 to 13 years old, years of schooling is larger for individuals interviewed in 1992 relative to those interviewed in 1996. There is no difference in education between 1992 and 1996 for individuals 14 to 30 years old. For older cohorts, individuals interviewed in 1996 have slightly higher schooling consistent with the overall increase in education over time. Taken together these graphics suggest a strong negative impact of the drought on individuals that were 6 to 13 years old in 1996. This cohort of children is a judicious group of interest to analyze the schooling consequences of the drought for two main reasons. First, the threshold of 13 years old is important as it represents the final year of primary education for children who started school at age 6 and had a normal progression through school. Moreover about 91 per cent of the sample has completed 7 years of schooling or less. Second, exposure to drought for children ages 2 to 5 years old is likely to affect children’s educational attainment.