METHODOLOGICAL FRAMEWORK 25
In contrast, participants tend to be younger and have more land but reside farther from an all-weather road compared with nonparticipants outside Fadama II communities. These results suggest that Fadama II targets vulner- able groups, such as female household heads, larger households, and people in more remote locations, although apparently the project also targets com- munities with relatively large farms. It does not select for other factors, such as education, ownership of productive assets or livestock, and agroecological zone. We also observe that younger people and those in remote areas were more likely to participate in EIGs. Education also increased the propensity to participate in EIGs.
These probit model results were used to compute the propensity scores that determined the PSM estimate of ATT. Several methods are possible for selecting matching observations (Smith and Todd 2001). We used the kernel matching method (using the normal density kernel), which uses a weighted average of neighbors (those observations within a given range in terms of the propensity score) of a particular observation to compute matching observa- tions. Unlike the nearest-neighbor method, using a weighted average improves the efficiency of the estimator (Smith and Todd 2001). Observations outside the common range of propensity scores for both groups (meaning those lack- ing common support) were dropped from the analysis. This requirement of common support eliminated about half of the total number of observations, indicating that many of the observations from the various strata were not comparable.
Further testing of the comparability of the selected groups was done using a balancing test (Dehejia and Wahba 2002), which tests for statisti- cally significant differences in the means of the explanatory variables used in the probit models between the matched groups of Fadama II participants and nonparticipants. In all cases, that test showed statistically insignificant differences in observable characteristics between the matched groups (but not between the unmatched samples), supporting the contention that PSM ensures the comparability of the comparison groups (at least in terms of observable characteristics).
We used bootstrapping to compute the standard errors of the estimated ATT, generating robust standard errors, because the matching procedure matched control households to treatment households “with replacement” (see Abadie and Imbens [2002, 2006] on the use of bootstrapping for infer- ence in matching estimators).11 Given that FUGs were managed by FCAs located
11 Sampling with replacement means that the sampled observation is replaced such that with repeated sampling the probability of it being randomly picked for each new draw remains the same.
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 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62 |
Page 63 |
Page 64 |
Page 65 |
Page 66 |
Page 67 |
Page 68 |
Page 69 |
Page 70 |
Page 71 |
Page 72 |
Page 73 |
Page 74 |
Page 75 |
Page 76 |
Page 77 |
Page 78 |
Page 79 |
Page 80 |
Page 81 |
Page 82 |
Page 83 |
Page 84 |
Page 85 |
Page 86 |
Page 87 |
Page 88 |
Page 89 |
Page 90 |
Page 91 |
Page 92 |
Page 93