PANEL DATA APPROACHES TO ESTIMATING IMPACTS 29
have access to irrigation.1 Second, for those households that do have access to irrigation, the fact that the growth equations are defined at the household level rather than at the plot level, as in the hedonic specification, may also render parameter estimates less precise because not all of a household’s plots may be irrigated, but the household is identified as having access to irrigation if any plot is irrigated. Third, the extension service was largely disrupted by civil strife later in the survey period. Therefore, the measure of extension visits in the initial 1995/96 survey period may not represent the true degree of extension service received by farmers in 2003/04. Because of the sensitivity of GMM estimates to changes in specification, we re-estimate a series of alternative specifications in Table 5.2. These changes include reestimating Equation 5.1 using a limited-information maximum- likelihood estimator as opposed to the GMM estimator. We also vary the set of covariates included in the specification, dropping various combinations of shock variables, including the rainfall and price shocks, then the conflict variables,2 and finally all three sets of shocks. The coefficient estimates com- pared with the original estimates from Table 5.1 are presented in Table 5.2. These robustness checks suggest that the coefficient estimates are insensitive to these changes in specification or estimator, with most coefficients rela- tively stable. All results with respect to travel time remain statistically sig- nificant at the 1 percent level, while our results for irrigation and extension remain statistically insignificant.
One should also be aware of the disadvantage of using this panel data approach. By the second wave of the household surveys, the panel households were older than the average households in the cross-section. This means that analysis based on panel households may underestimate or overestimate the effect of infrastructure on older households, because they may benefit more (or less) than younger households. In addition, the infrastructure and exten- sion variables were initial values in 1996 in the growth regressions and do not reflect any changes in these variables and the resultant impact on the out- come variables. Both hedonic and panel data approaches provide a robustness check that enables us to derive the key findings that rural road investments have a strong effect on household welfare.
1 When we restrict the regressions in Table 5.1 to the set of landholders, we find no difference in the statistical significance of the irrigation variable. This suggests that the unit of analysis may be driving the nonresult rather than the specific subsample on which the specification is
estimated. 2 In Table 5.1, we also note the positive coefficient of the conflict variable on agricultural income growth. This relationship seems inconsistent, which could be due to the endogeneity of the conflict variable or to measurement. Our results on the impact of publicly funded rural infrastructure and services are robust to these concerns, as we demonstrate in Table 5.2.
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50