For instance Alderman, Hoddinott and Kinsey show that exposure to drought for children aged 1 to 3 years affect readiness to school, and progression through school. Figures 3.7 to 3.10 further explore the relation between exposure to drought and human capital accumulation five years after the drought period. Individuals 6 to 13 years old in 1996 were aged 11 to 18 years old in 2001/2002. Figures 3.7 to 3.8 show that while individuals 11 to 12 years old in 2001/2002 are slightly less likely to be enrolled in school relative to individuals 11 to 12 years old in 1992, the pattern is reversed for individuals 13 to 24 years old suggesting that individuals have caught-up for their delayed schooling during the drought years by staying more in school. This hypothesis is further investigated by analyzing years of education for people . Figure 3.9 shows years of education for individuals aged 11 to 40 years in 2001/2002 and in 1992. Overall there is no difference in years of schooling for individuals ages 11 to 14 years interviewed in 1992 or in 2001/2002. Individuals aged 15 to 22 years when interviewed in 2001/2002 have slightly higher years of education than individuals 15 to 22 years old when interviewed in 1992. However the same pattern emerges for individuals aged from 23 to 40 years. This would suggests that controlling for the trend in education, individuals 11 to 22 years old interviewed in 2001/2002 have lower years of education than the same age-group of individuals interviewed in 1992,black flower bucket especially in the sample of females. In all the analysis I restrict the sample to individuals residing in rural areas whose livelihood are most likely to depend directly on good rainfall .
Even though households living in urban areas are affected by the drought through increased food prices, household residing in rural areas are both experiencing a drop in income and higher prices. As a first step I use the ZDHS-1996 and the ZDHS-1992 to examine the short-run effects of the droughts in 1992 and 1995 on schooling. The basic model I estimate relates enrollment status or years of education to a dummy equal to one if the individual was interviewed in 1996 and zero if the individual was interviewed in 1992, a second dummy equal to one if the individual was from 6 to 13 years old in 1996 or in 1992, and the interaction between these two dummy variables.Table 3.2 presents baseline results for the estimation of equations and . Regression results for the short-run effect of the drought are given in Panel A; while results for the medium-run consequences of the drought are presented in Panel B. Columns 1 to 3 show results using enrollment status as the dependent variable. Columns 4 to 6 show regression results using years of educations as the variable of interest. The difference-in-differences estimate is the coefficient on the interaction between the dummy for individuals ages from 6 to 13 and the dummy for individuals interviewed in 1996 for the short-run analysis; and the the coefficient on the interaction between the dummy for individuals ages from 11 to 18 and the dummy for individuals interviewed in 2001/2002 for the medium-run analysis. As in Panel A, the dependent variable in columns to is enrollment rate of individuals 11 to 24 years old at the time of the survey, while in columns 4 to 6 the dependent variables is years of schooling for individuals 11 to 40 years old at the time of the survey. In column 1 the difference-in-difference estimate is .044, a large and highly significant effect. This result suggests that children that were exposed to the drought may have stayed in school at later ages to catch-up on their delayed schooling.
The same result holds in column 2 and 3 where I condition on a province and age fixed effect and control for a range of individual and household characteristics and their interaction with the dummy for individuals interviewed in 2001/2002. Turning to the medium-run consequences of the drought on accumulated years of the schooling, the results in column 4 suggest that individuals 11 to 18 years old in 2001/2002 have -.173 years of schooling than individuals 11 to 18 years old in 1992. This point estimate suggests that the drought has reduced the stock of schooling of children exposed to the drought even 5 years after the drought but with a tendency of catch-up and return to pre-drought levels of schooling. Average years of schooling among individuals ages 11 to 40 in 2001/2002 or 1992 is about 5.646 years of school. So the estimated coefficient corresponds to a 3 percentage point decrease in years of education. This medium-run estimate is three times smaller than the short-run estimate effect of the drought. In column I condition on a province and age fixed effect so that only within region variation for a given age are used for identification. This makes the point estimate slightly smaller and imprecisely estimated. In column I further control for a set of individual and household characteristics and their interaction with the dummy for individuals interviewed in 2001/2002. The point estimate increases in magnitude to -.273. This latter point estimate corresponds to 4.8 percentage points decrease in years of schooling at baseline.Overall this analysis show that children exposed to the drought have accumulated less schooling during the drought. There is also evidence that such individuals have partially caught-up on their delayed education probably by staying enrolled in school at older ages. The results in Table 3.3 check that the result is robust to the timing of ZDHS-1992 and explore the heterogeneity of the consequences of the drought.
As in Table 3.2, Panel A shows results for the estimation of while Panel B shows regression results for . All specification controls for a province and an age fixed effect and a set of control variables. Columns to examine the impact of the drought on school enrollment while the results in column to show results for years of schooling. In columns and I restrict the analysis to individuals interviewed early during the raining season. Turning first to the results in Panel A, the coefficient estimate in column is higher than the corresponding coefficient in Table 3.2 but still falls within the confidence interval. The coefficient estimate for years of education in column is smaller in magnitude than the corresponding coefficient in Table 3.2 but the two estimates are not statistically different. Similarly the medium-run estimates using the sub-sample of individuals interviewed early during the 1992 raining season are close to the estimates in the full sample. Columns and restrict the estimation to males while columns and show estimation results in the sub-sample of females. In the short-run, females’ school enrollment and years of education are more negatively affected by the drought. But the male-female difference are not statistically significant. However in the medium-run these differences are much larger. Columns and of Panel B show that while in the medium-run men’s education has converged to pre-drought levels, women’s education is still -.525 years of schooling below pre-drought levels. This is a large effect given that average years of schooling among female aged 11 to 40 years in 2001/2002 or in 1992 is about 5.19 years of education. In percentage terms,square black flower bucket the estimated coefficient corresponds to a 10 percentage point decrease in schooling. Table 3.4 presents results of the triple difference estimators. The key variable of interest is the triple interaction between the young cohort dummy, the dummy equal to one for individuals interviewed either in 1996 or 2001/2002 and zero for individuals interviewed in 1992, and one of the two measures of rainfall deficit. All models include the three double interactions between : the young cohort dummy and the dummy for individuals interviewed either in 1996 or 2001/2002, the young cohort dummy and one of the two measures of rainfall deficit, and lastly the dummy for individuals interviewed either in 1996 or 2001/2002, the young cohort dummy and one of the two measures of rainfall deficit. I also control for a series of individual and household characteristics and their interaction with the the dummy for individuals interviewed either in 1996 or 2001/2002. Columns to show results for school enrollment; in columns and I restrict the estimation to males and females respectively. Columns to show results for school enrollment; in column 5 and 6 the estimation is restricted to males and females respectively. Panel A shows results for the short-run impact of the drought while Panel B presents regression results for the medium-run effects. Each row corresponds to a separate regression. The results in column show that increasing rainfall deficit by 10 per cent reduces school enrollment by .021 in the short-run and by 0.034 in the medium-run. Using the second measure of rainfall deficit, the coefficient is still negative but not precisely estimated. Column and show that the effect is slightly larger for males using the first measure of rainfall deficit. So the results indicate that the negative impact of the drought on enrollment rates is stronger in the provinces with the lowest rainfall realization in 1992 relative to the period 1980-1991.
However the effect in the medium-run are of different sign than the difference-in-differences results. This suggests that the previous analysis of the medium run consequences of the drought are likely to have been biased by other aggregate shocks correlated with enrollment rates. Overall I find that the drought has had strong short-run and medium-run consequences on school enrollment : a 10 per cent drop in rainfall relative to the mean results in a 4 percentage points reduction in enrollment rates in the short-run and a 7 percentage points reduction in the medium-run. Next, I turn to the results for years of schooling. The results in column show that increasing rainfall deficit by 10 per cent reduces years of schooling by 0.032 in the short-run and 0.076 in the medium-run. Using the second measure of rainfall deficit I find that for individuals who were 6 to 13 years old in 1996 and are living in the four provinces which experienced the highest rainfall deficit, years of schooling drops by .272 in the short-run. The coefficient estimate in the medium-run is of similar magnitude but is imprecisely estimated. The results also show that the effect is mainly driven by females. In the short-run females who were 6 to 13 years old and living in the four provinces which experienced the highest rainfall deficit lose up to .435 years of schooling. These findings suggest that the drought has had a larger impact on male enrollment rates but female years of schooling were affected the most. A potential reason for this might be the differential cost of re-entering school after drop out that females face. If this is the case then it is possible that more male drop out of school following a drought but females are less likely to return to school after the shock has been experienced. In this paper I have investigated the impact of the 1992 and 1995 droughts in Southern Africa on investment in schooling and years of schooling. Using three cross sectional data in Zambia collected at the start of the 1992 raining season, in 1996 and in 2001/2002, I am able to analyze the short-run and medium-run consequences of the drought on school enrollment and years of education. My empirical strategy compares young children interviewed after the drought with children of the same age-group interviewed before the drought. This empirical strategy allows me to use data collected in 1992 to control for pre-drought schooling levels and uses older cohorts to control for trends in enrollment rates and educational attainment. Using this difference-in-differences strategy I find that young children exposed to the drought are .048 points less likely to be enrolled in school and loose up to .35 years of schooling in the short-run. In the medium-run I find evidence of catch-up through enrollment in school at older ages. Acknowledging that other aggregate shocks correlated with schooling can generate these findings, I employ a triple difference strategy to confirm that the negative impact on schooling can be attributed to the drought using rainfall data from actual rainfall gauges. One drawback of the data I use is the lack of information to investigate the mechanisms through which the drought impacted enrollment rates and years of schooling.