The model is estimated using the sample of children ages 24 to 59 months living in rural areas. In these and in all other models standard errors are clustered at the treaty basin level. In the next-to-last two rows we report the F-statistic of the first stage for the “own dam” coefficient and for the “upstream dam” coefficient . The instruments for irrigation dams appear sufficiently strong to avoid bias from weak instruments. However the instruments for all “own dams” are small, and the bias that could result from this should be kept in mind when interpreting the results. I also run the Sargan-Hansen test to check that the instruments are relevant. The null hypothesis here is that the instruments are valid, i.e. they are uncorrelated with the error term and correctly excluded from the second stage. The p-values of the tests are when we estimate the model using irrigation dams only and and when including all types of dams. These p-values confirm that the instruments are relevant. We first analyze how dams with some irrigation purpose have affected the nutritional status of children in the 6-digit river basin where the dam is located and downstream. The first column of table 1.4 shows estimates from models which include controls for demographics and rainfall. The IV estimate indicates a large and significant decrease in the nutritional status of children in the 6-digit river basin where a dam is located and an increase in downstream 6-digit river basins. An additional dam reduces child height-for-age z-score in the 6-digit river basin basin where it is located by point of standard deviation and increases the probability of being stunted by points. In contrast an additional dam increases height-for-age z-score in downstream 6-digit river basins by points and reduces stunting by .03 points,french flower bucket but this last coefficient is small and insignificant.
These effects are large given that average height-for-age and stunting are -1.91 and 47 per cent respectively. The ratio of the effect of an additional dam in the 6-digit river basin where it is located to the mean are 18 per cent for height-for-age and 34 per cent for stunting. The same ratio for downstream 6-digit river basins are 8 per cent for height-for-age and 6 per cent for stunting. Columns and show estimates conditional on children’s gender, a dummy for twin birth, mother’s age, mother’s years of education, a dummy for a child living in a female headed household, the number of household members and the number of children in the household less than 5 years old. The coefficient on “own dam” drops by less of a standard error while the coefficient on “upstream dam” increases slightly. Columns and further control for rainfall during the survey year and the two years prior to the survey year. This has little effect on the coefficient estimates. Next we turn to the impact of all types of dams on weight-for-age and malnourishment. We find that dams with some irrigation purpose have, on average, little effect onweight-for-age in the 6-digit river basin where they are built but increase the incidence of malnourishment by points or 49 percent. However an additional irrigation dam increases weight-for-age in downstream 6-digit river basins by .10 to .20 points on average but this has no effect on the incidence of malnourishment. Lastly we analyze the impact of all types of dams taken together on child nutrition. We find that the impact of dams on weight-for-age are small and not precisely estimated. However we find that while an additional dam has no effect on height-for-age, on average, it increases the proportion of children who are stunted by .08 points or 17 percent. Moreover we find that an additional dam increases height-for-age and reduces stunting in downstream 6- digit river basins by .15 and .08 points respectively . It is important to know the overall net effect of dam construction on the nutritional status of children. We focus on the results for irrigation dams to carry out this analysis, results for all dams show a similar pattern.These impacts are small compared to standard errors of the point estimates or sample standard errors suggesting that dam construction has little aggregate effect on height-for-age and and weight-for-age z-score.
Turning to stunting and malnutrition, we find that dam construction has increased stunting in the average river basin by .025 points and malnutrition by .012, effects that are small compared to standard errors of the point estimates and sample standard errors. Taken together our findings show that while dam construction in Sub-Saharan Africa has had little aggregate effects, it clearly generates losers and winners. This suggests the scope for more effective policy making in order to capture the benefits from dam construction while compensating those who may lose. As most poor regions, rural African economies are characterized by limited access to insurance against risk. The inability to cope against shock may interact with the increased variance of agricultural production from dam construction to reduce household’s income and food security. We investigate this in Table 1.5 by examining how dam construction reduces or exacerbates the impact of rainfall shocks on child nutrition. We use rainfall deviation from its mean between 1970 and 2002 as a measure of rainfall shock, and we consider two types of intensity of rainfall shocks: the number of instances during the survey year and the two years prior to the survey year when rainfall was at least.3 point .6 point of standard deviation below the mean. So these rainfall shocks can take any value between 0 and 3. In columns to rainfall shock is the number of times rainfall was at least .3 point of standard deviation below the mean, while columns to show results where the rainfall shock is the number of times during which rainfall was at least .6 point of standard deviation below the mean. We find that dams amplify the impact of rainfall shocks both in the basin where they are built and in downstream river basins . We also find that dams amplify the effect of rainfall shocks more in the 6-digit river basin where they are built than in downstream basins, but these differences are significant only for larger shocks. In this section we analyze whether the effect of dams are more pronounced along certain demographic characteristics.
It is particularly important to know whether the construction of dams affect more significantly vulnerable children and poor households. Because the DHS do not collect measures of income or consumption we use demographic characteristics to proxy for a household’s likelihood to be poor or vulnerable: a dummy for whether the household is female headed and mother’s years of education. We also investigate whether the impact of dams is different for boys and girls. These regressions have 31037 observations each. All models control for child gender, a dummy for twin child, mother’s age, mother’s education, a dummy for female headed household, the number of household members and children under 5 years, rainfall during the survey year and the two years prior to the survey year. The results reported in Table 1.6 consistently show that girls, female headed households and children of more educated mothers benefit more from a dam in an upstream river basin. For children living in river basins where a dam is located we don’t find, across different measures of child nutritional status, any systematic difference in the effect of the dam by demographic characteristics: for height-for-age the “own dam” effect are lower for girls, male headed households,bucket flower and less educated mothers; while for weight-for-age these effects are larger. The framework presented in Section 2 highlights improved access to food as an important channel in the causal link between dam construction and the nutritional channel of children. The relevance of this channel is confirmed by Strobl and Strobl who find that large dams in Africa increased crop productivity in downstream regions while cropland within the vicinity of the dam tends to experience productivity losses. However the construction of a new dam may be accompanied or substituted for the provision of other public goods that may affect children’s nutritional status. We examine this possibility in Table 1.7. Table 1.7 shows estimation of model where the dependent variables are measures of access to health services, tap water and electricity. The model is estimated following an IV strategy using the same instruments described above. Columns to show that dam construction improves access to health services and electricity in the basin where the dam is built. However columns to show that access to health services, tap water and electricity are affected little by the construction of an irrigation dam.
Put together with the findings in Strobl and Strobl, the results in Table 1.7 also point to the importance of improved access to food as the main mechanism between dam construction and the nutritional status of children in Africa. Rural-urban and rural-rural population movements are central mechanisms in the process of structural transformation and economic development. As the Agricultural sector shrinks, workers leave rural areas for manufacturing and services jobs in cities. Moreover differences in agricultural development and economic activity between rural areas may prompt a process of labor reallocation between these regions. What is the effect of these population flows on rural economic activity and welfare? Despite a large empirical literature on the effects of rural out-migration on individuals and households, little empirical evidence exists on their consequences at a macro or meso level. Exceptions include papers that focus on a specific aspect related to migration such as remittances and investigate how they affect village-level investment and economic growth. An under-investigated question in this literature is how rural out-migration influences rural labor market outcomes. This analysis is related to papers that examine the effect of international emigration on domestic labor markets. Lucas shows that migration of mine workers to South Africa increased wages in both Malawi and Mozambique – the two largest source of foreign mine workers in South Africa – suggesting that out-migration may contract the labor market at origin. However Mishra is the first paper to provide evidence of a causal link between emigration and wages in source countries. The analysis presented in this paper is the first attempt to measure the effect of internal migration on wages in sending regions. I focus on a specific internal migration process, rural out-migration, because it accounts for a large share of a country’s internal migration and is particularly relevant for understanding rural economic development. Moreover given the magnitude of internal labor flows relative to international labor flows, it is surprising that the literature overlooked the effects of rural out-migration on rural labor markets. By way of comparison Mishra estimates that, in 2000, the number of Mexican emigrants living in the US was about 16%. In other words, in 2000, for every 100 Mexicans living in Mexico, 16 were residing in the US. I find much larger rural emigration shocks in Brazil. My analysis shows that between 1980 and 2000, for every 100 individuals residing in rural areas, 95 had already migrated out to cities or other rural areas outside their state of original residence. The empirical framework developed in this paper builds on Card and Lemieux, Borjas and Mishra. As in Mishra, this paper investigates how emigration affects the wages of those who stay behind. The empirical strategy in Mishra follows the framework in Borjas to investigate how an emigration shock of Mexican workers to the US in a specific skill group defined by education and experience affects the wages of workers of that skill group who remained in Mexico. In this paper, I use a cohort analysis similar to the approach in Card and Lemieux to examine how rural out-migration of a given cohort affects the wages of rural workers in that cohort. This cohort analysis is motivated by several considerations. First, rural workers in Brazil present little differentiation by educational attainment. In 2000, less than 13% of individuals included in my sample – individuals residing in rural areas and aged 20 to 54 years in 1991 – had completed secondary education with the majority having less than a complete primary education. Second, occupations in rural areas are characterized by many routine activities with a high level of learning-by-doing which may reward more experience than education.