In this section I analyze the underlying mechanisms behind the increase in the share of the population earning zero income. There are multiple likely underlying mechanisms behind the increase in zero income. One could be that the workers lost their employment, whereby we would see an increase in unemployment. The lower demand for labor following the destruction of harvest (FAO, 2015) could have led to lower wages, and for the informal workers of Tabasco a possible fall to below minimum wage levels. If this were the case, we would see an increase in the number of workers having employment but receiving zero (or below the minimum wage) income.
Another possibility is that younger cohorts enrolled in school again as the return to labor decreased and thereby the opportunity cost of education. A last possibility is that the reduced demand for labor made people redraw from the labor market, increasing the unavailable share of the population. To test whether any of these labor market changes were the underlying mechanisms, I use equation 3 to estimate the differences across the two treatment measures after the flood. The results are presented in Table 5. The intensive treatment measure of log distance from the flood border show the expected decrease in employment without receiving a work income above the minimum wage, lower enrollment, fewer people not available for the labor market, and fewer receiving economic support the further away from the flood we move.
None of these results are strongly or at all significant. Looking at the extensive treatment measure of being flooded or not, the results are stronger, indicating that the extensive margin is more important than the intensive. I find that being inside the flooded area increase the likelihood of having a job giving zero income and being enrolled in school. These results are in correspondence with the theory that a large natural disaster reduces productive employment possibilities. It is a common result from developing countries that there is little unemployment (Ball et al. (2011) and Fields (2011)), which is supportive of the result here where there appears to be no effect on unemployment or participation in the labor force. Additionally, I find borderline significant positive effects on the enrollment rates and receipt of economic support.
This is in correspondence with the theory that investment in human capital increases when the outside option decreases. The increase in recipients of economic support may also explain the
lack of mitigative actions to the lack of return to labor.
Table 5: Effect on connection to the labor market
(1) (2) (3) (4) (5)
Employed with no pay Unemployed Enrolled Not available Receive economic support Panel (a)
Distance*post -0.0158* 0.000811 0.00462 -0.00963 -0.0103
(0.009) (0.007) (0.017) (0.015) (0.021)
Adj. R2 0.0286 0.0186 0.499 0.352 0.214
N 131734 131734 60428 131734 52046
Mean DV 0.0945 0.103 0.436 0.481 0.290
Panel (b)
Inside*post 0.0237** 0.00969 0.0366* -0.0117 0.0836*
(0.011) (0.010) (0.021) (0.016) (0.042)
Adj. R2 0.0286 0.0186 0.485 0.352 0.214
N 131734 131734 131734 131734 52046
Mean DV 0.0945 0.103 0.207 0.481 0.290
OLS estimates. All estimations include locality, quarter, municipality*year fixed effects, and the baseline controls.
Standard errors in parentheses
∗p < .1,∗ ∗p < .05,∗ ∗ ∗p < .01
In the main analysis I focus on the individual income. However, if the household rearranged the income generation across the members so that one member became the primary breadwin-ner, we would see this increase in persons earning zero work income. If, for example, one from the household started to work within the household to rebuild the assets of the family, he/she would be identified as unemployed, not available at the labor market, or working domestically receiving zero income. Such a respond would result in a smaller effect on the household income, and thus the consumption possibilities, although also a higher variation within the household.
Table 6 shows the effect of the flood on household income confirming the result from 2, namely a significant and sizable negative effect of the flood both using the binary measure and the distance measure. This indicates clearly that the affected households were not capable of mit-igating the negative income shock by changing the labor supply of the household, but on the contrary the flood had even stronger impacts on household income. On the extensive margin I also find a decrease in the variance of the household income, indicating that they were not com-pensating the lower income by rearranging income generation across members, and an increase in the share of households where no one had a strictly positive income.
Table 6: Effect on household income
(1) (2) (3) (4) (5) (6)
HH income Var of HH income Zero HH income HH income Var of HH income Zero HH income Inside*post -0.136*** -0.0285*** 0.0464**
(0.047) (0.009) (0.023)
Distance*post 0.0877*** 0.0101 -0.0325
(0.027) (0.007) (0.020)
Adj. R2 0.240 0.132 0.105 0.240 0.131 0.105
N 44184 39310 44184 44184 39310 44184
OLS estimates. All estimations include locality quarter, municipality*year fixed effects, the mean household baseline controls, and number of persons in the household between 12-97. HH stands for Household, and Var is the variance. Robust standard errors in parentheses∗p < .1,∗ ∗p < .05,∗ ∗ ∗p < .01
Looking more into the distributional effects of the drop in productive employment by in-teracting the treatment measures with groups expected to be affected to a larger or smaller degree revealed that the effect on the intensive margin was strongest for males (Table 7). I find no differences across educational attainment, and only weakly on being formally or informally employed. With respect to the external margin of being flooded or not, males had a much higher risk of zero work income in the flooded area.
Table 7: Heterogeneous effects on earning below the minimum wage
(1) (2) (3) (4) (5) (6) (7) (8)
Dependent variable: Zero work income [0;1]
Distance*post -0.0301*** -0.0366*** -0.0238*** -0.0369***
(0.009) (0.010) (0.009) (0.012)
Inside*post -0.0160 0.0182 0.0404** 0.0482**
(0.016) (0.018) (0.020) (0.023) Male -0.380*** -0.363*** -0.357*** -0.0305*** -0.377*** -0.363*** -0.357*** -0.0302***
(0.014) (0.012) (0.012) (0.007) (0.014) (0.012) (0.012) (0.007)
Male*Distance 0.00868**
(0.004)
Male*Affected 0.0906***
(0.022)
Age -0.0452*** -0.0457*** -0.0445*** -0.0313*** -0.0452*** -0.0452*** -0.0445*** -0.0313***
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Age*Distance 0.000314***
(0.000)
Age*Affected 0.000232
(0.000)
Years of schooling -0.00999*** -0.00990***
(0.001) (0.001)
Years of schooling*Distance -0.000124
(0.000)
Years of schooling*Affected -0.00200
(0.002)
Formal employment -0.126*** -0.109***
(0.010) (0.007)
Formal employment*Distance 0.00740*
(0.004)
Formal employment*Affected -0.0182
(0.015)
Adj R2 0.287 0.287 0.292 0.119 0.287 0.286 0.292 0.119
N 129815 129815 129757 63484 129815 129815 129757 63484
Mean DV 0.611 0.611 0.611 0.198 0.611 0.611 0.611 0.198
OLS estimates. All estimations include locality, quarter, municipality*year fixed effects, and the baseline controls.
Standard errors in parentheses
∗p < .1,∗ ∗p < .05,∗ ∗ ∗p < .01
8 Robustness checks
In this section I test the validity of the main results of a decreased work income in the immediate aftermath of the flood and a longer-lasting increase in the share earning zero income.
One concern is whether the results are driven by migration changing the composition of workers in the flooded area. As depicted in Figure 1, there are no indications of changing population sizes across the flood border. A more formal test of selection away from (or into) the flooded area is to look at demographic differences across treatment status using equation 3 (Bharadwaj et al., 2018). The results are presented in Appendix Table A.11 with age, male, years of schooling, and marital status as dependent variables. There are no differences in demographic characteristics relative to the flood except for a decreased share of males inside the flooded area. This indicates that there is a possibility of men migrating to find work somewhere else due to the lower productive employment possibilities. However, as neither the population size nor any of the other demographic variables were affected the result of men migrating is not robust.
Next, I test the validity of treatment status. Despite the fact that localities closer to the flood may have been affected by changed access to markets and other factors, I do not expect this to be a major factor in comparison to those directly affected by the flood. Splitting the sample into the two groups of within and outside the flood border, I check whether the flood had significantly different effects within the groups relative to the distance to the flood border. The results are presented in Appendix Table A.12. While there were no significant differences across the localities outside the flood border relative to the distance to the flood border, the main results of lower income for the most affected and an increased share of zero income earners prevail for those inside the flooded area. This confirms that while the extensive margin dominates, the intensive margin across the affected localities is significant.