The share of titled land positively influenced households’ adoption decisions,
and an increase in this share by 1 percentage point increased the probability of
adopting agroforestry by 0.26 percentage points. Whether the farmer or neighbor
experienced a reallocation of upland plots did not appear to influence adoption,
neither did the household’s own expectations regarding such a reallocation. The
villagers’ expectations, however, were found to negatively affect farmers’ propensity to adopt. An increase by 1 percentage point of the share of villagers believing
that a reallocation was likely to occur reduced the adoption probability by 0.61
percentage points, an indication that, because of the tight social organization of the
villages, the general opinion prevailed over individual expectations regarding this
decision. More importantly, it showed that the reallocation threats, as perceived by
the villagers, may have discouraged the adoption of agroforestry.
7.6.3 Determinants of the Adoption of Soil Conservation
Techniques When Considering Plot Characteristics
Apart from investigating which factors influence a household’s decision to adopt
agroforestry practices, we are also interested to know where within a farm agroforestry is practiced. Thirty-two percent of our sample households cultivated both titled
and untitled land (households cultivated on average four upland plots). A householdlevel model is unable to capture the effects of land tenure, soil characteristics and
other plot-specific variables that may impact adoption; hence, we also developed a
plot-level model for the adoption decision. This approach has been widely applied in
the literature to estimate the impact of land tenure on adoption incentives (Besley
1995; Hagos and Holden 2006; Hayes et al. 1997; Pender and Fafchamps 2001).
In the plot-level model we were not able to correct for exposure bias, as the
selection regarding knowledge occurred at the household level. However, a test of
independence of equations (Wald test) applied to the household-level model did not
reveal any selection bias. As a result, we based our plot-level model on the
households that were aware of agroforestry as an SCT only (see Technical Note
2 at the end of this chapter). The plot-specific explanatory variables X 3ij included in
the model are described in Table 7.9.
The regression results of the plot-level probit model are shown in Table 7.10.
The predictive power of the model is rather limited, which can be explained both by
the presence of numerous household-level regressors and the low plot-level adoption rate. However, apart from two regression coefficients (on wealth level and
relative upland size), the results are very similar to the ones produced by the
household-level model, indicating their robustness.
As in the household-level model, we found soil characteristics to be very
important determinants of farmers’ adoption decisions. Agroforestry was used on
7 Soil Conservation on Sloping Land: Technical Options and Adoption Constraints
267
and an increase in this share by 1 percentage point increased the probability of
adopting agroforestry by 0.26 percentage points. Whether the farmer or neighbor
experienced a reallocation of upland plots did not appear to influence adoption,
neither did the household’s own expectations regarding such a reallocation. The
villagers’ expectations, however, were found to negatively affect farmers’ propensity to adopt. An increase by 1 percentage point of the share of villagers believing
that a reallocation was likely to occur reduced the adoption probability by 0.61
percentage points, an indication that, because of the tight social organization of the
villages, the general opinion prevailed over individual expectations regarding this
decision. More importantly, it showed that the reallocation threats, as perceived by
the villagers, may have discouraged the adoption of agroforestry.
7.6.3 Determinants of the Adoption of Soil Conservation
Techniques When Considering Plot Characteristics
Apart from investigating which factors influence a household’s decision to adopt
agroforestry practices, we are also interested to know where within a farm agroforestry is practiced. Thirty-two percent of our sample households cultivated both titled
and untitled land (households cultivated on average four upland plots). A householdlevel model is unable to capture the effects of land tenure, soil characteristics and
other plot-specific variables that may impact adoption; hence, we also developed a
plot-level model for the adoption decision. This approach has been widely applied in
the literature to estimate the impact of land tenure on adoption incentives (Besley
1995; Hagos and Holden 2006; Hayes et al. 1997; Pender and Fafchamps 2001).
In the plot-level model we were not able to correct for exposure bias, as the
selection regarding knowledge occurred at the household level. However, a test of
independence of equations (Wald test) applied to the household-level model did not
reveal any selection bias. As a result, we based our plot-level model on the
households that were aware of agroforestry as an SCT only (see Technical Note
2 at the end of this chapter). The plot-specific explanatory variables X 3ij included in
the model are described in Table 7.9.
The regression results of the plot-level probit model are shown in Table 7.10.
The predictive power of the model is rather limited, which can be explained both by
the presence of numerous household-level regressors and the low plot-level adoption rate. However, apart from two regression coefficients (on wealth level and
relative upland size), the results are very similar to the ones produced by the
household-level model, indicating their robustness.
As in the household-level model, we found soil characteristics to be very
important determinants of farmers’ adoption decisions. Agroforestry was used on
7 Soil Conservation on Sloping Land: Technical Options and Adoption Constraints
267