Effect of current productivity shocks

The effect of current shocks on education outcomes can be estimated with the following specification:

Eijty = β₀+ β₁Pj,y−1+ β₂SP EIj,y−1+ γX_ijy + δ_j + µ_t+ ν_y+ _ijty (3.22)

where i denotes the child of age t living in location j during the survey year y.

Since households geo-coordinates are available only in the LSMS data, j designates the geographical units (50km* 50km) where the household lives in the LSMS estimations, while j designates the district of residence in the Uwezo estimations. The parameters δ, µ and ν are location, age and year fixed-effects, respectively. The error term ijty is clustered by location j, and X_ijy is a set of household controls such as the number of adults and children in the household, the number of boys among siblings, and age and education of the household head. E_ijty is a large set of education outcomes that measures education decisions and educational achievement. I regress current education outcomes E_ijty on the lagged climate variable SP EI_j,y−1 and on the lagged aggregated price index P_j,y−1.

By adding region and year fixed effects, this estimation strategy compares children from the same location in different rounds of the survey. It captures the causal effect

27This index can be constructed at the household level. However, the area allocated to each crop may be endogenous at the household level, while at a larger scale (community or district level) Sc,j,2000 is representative of the geographical conditions suitable for different crop’s cultivations.

of productivity shocks on education outcomes if several assumptions are satisfied. First, SP EI_j,y−1and P_j,y−1should change the labor productivity (see sub-section3.4.4). Second, the shocks should be purely exogenous (see sub-section 3.3.2 and 3.3.3 for further discussion) and finally, they should not be correlated with unobserved variables that would explain education outcomes. This question will be addressed later in section 3.5.

To estimate the effect of productivity variations on children’s education and activities (whether the child works, whether the child is enrolled in school, whether the child has dropped out of school and what is the highest grade achieved), I use the LSMS data and I restrict the sample to school aged children. In TableA3.17, I observe that continuous price and climate variables have no significant impact on either education or work decisions.

Table 3.1: Effect of positive shocks on children’s activities Work Enrolled Dropout Grade

(1) (2) (3) (4)

Positive Price Shockt−1 0.058* -0.035** 0.004 -0.063 (0.033) (0.017) (0.011) (0.082) Positive Rainfall Shock_t−1 0.084** 0.001 0.014* -0.124***

(0.033) (0.014) (0.008) (0.045) Negative Price Shockt−1 -0.013 -0.004 -0.006 0.006

(0.025) (0.014) (0.009) (0.074) Negative Rainfall Shockt−1 0.006 0.009 -0.004 -0.034

(0.028) (0.017) (0.008) (0.045)

R-squared 0.167 0.154 0.084 0.297

Observations 12,677 11,625 11,230 10,588

Localities F.E × × × ×

Year F.E × × × ×

Sources: LSMS-ISA from 2008, 2010 and 2012. Note: Standard errors, clustered by geographical units (0.5°×0.5° of precision), are reported in parentheses.

Controls are survey month dummies, the number of adults and the number of children in the household, age dummies, the gender and the birth order of the child, the number of boys among siblings, the age and the education of the household head. ***,**,* mean respectively that the coefficients are significantly different from 0 at the level of 1%, 5% and 10%.

Moving to the non-linearity of productivity shocks, I observe that positive rainfall shocks increase the probability of working, increase the probability of droping out of school, and lowers the grade achievement by 0.11 years (see Table 3.1). Although all coefficients are not significant, these results go in the same direction and suggest

144

that contemporaneous productivity shocks have a counter-cyclical impact on education decisions (^{∂E(w2,γ,∆,I}_∂w^{2(w1,w2γ,∆))}

2 < 0). In contrast, negative productivity shocks have no significant impact on education decisions. According to the model’s predictions, these results indicate that the substitution effect dominates the income effect and that children are encouraged to work more and to decrease their demand for education when they become more productive.

Based on the theoretical framework, these effects are also expected to vary with households’ wealth. To explore this heterogeneity, I compute the household consumption following the guideline of Deaton and Zaidi (2002).²⁸ Table A3.9 shows that results are very similar for rich and poor households. Most coefficients are not signficantly different, and suggest that positive productivity shocks are detrimental to education decisions at all wealth levels.

I can also examine whether the effects are heterogeneous across children’s age and across gender. TableA3.10shows that the effects of positive shocks on labor and education decisions are very close for boys and girls and are not statistically different. To consider heterogeneity by age, I define the 7-13 years old group as children of primary education age and the 14-16 years old group as children of lower secondary age. As pointed out in section 3.2, it is unclear whether the effects of productivity shocks will be more pronounced for the younger cohort or the older cohort. I see from Table A3.11, that older children are more likely to work and less likely to pursue their education, meaning that the the counter-cyclical relationship between productivity shocks and education decisions is strengthened when children get older. This result is consistent since the substitution effect should be larger when children become more productive.

Thus, the LSMS results suggest that positive productivity shocks increase child labor and are unfavorable to education achievement. If these shocks provoke erratic attendance, they should also decrease children’s cognitive skills. To test this hypothesis, I use the

28This consumption variable is composed of four sub-aggregates, food items, non-food items, housing consumption and consumer durables. In order to create a consumption variable independent from current shocks, I exclude all current consumption items such as food consumption and current non-food items that could have been affected by productivity shocks.

Uwezo data and I regress test scores on price and climate shocks.

Table 3.2: Effect of Contemporaneous Shocks on Test Scores Swahili Maths Swahili Maths

(1) (2) (3) (4)

Positive Price Shock_t−1 -0.007 -0.012 -0.015 -0.020 (0.016) (0.018) (0.016) (0.020) Positive Rainfall Shock_t−1 -0.029* -0.036* -0.023 -0.032

(0.017) (0.020) (0.018) (0.022) Negative Price Shock_t−1 -0.022 0.019 -0.020 0.010

(0.028) (0.026) (0.024) (0.021)

Drought_t−1 -0.001 0.003 0.006 0.010

(0.013) (0.016) (0.013) (0.015)

Sources: Uwezo data from 2011 to 2014. Note: Standard errors are clustered at the district level and are reported in parentheses. Controls are years dummies, the number of adults and the number of children in the household, age dummies, the gender and the birth order of the child

” the age and the education of the household head. ***,**,* mean respectively that the coefficients are significantly different from 0 at the level of 1%, 5% and 10%.

Table 3.2 presents the results and shows that only positive rainfall shocks decrease Swahili and maths scores, and they do so by 0.03 standard deviations. These effects are no longer significant when I restrict the sample to enrolled children, probably because children who stay at school during positive shocks are positively selected.

In conclusion, households take advantage of a labor productivity increase by calling on child labor. This decision interacts with education enrollment and achievement but has very little effect on schooling performance.