land is much more likely to have other crops grown on it, once soybean and paddy
rice have been harvested.
I now discuss the effect of variables that can be affected by policy on agricultural
land and intensity of cultivation. Results are presented in column 1, Table 8.8
30,31
Results in Table 8.8 show that a 1 % decrease in travel time to market increases the
percentage of agricultural land cultivated by 2.9 % points. Population has no effect
on the intensity of cultivation for either group of villages. Short run crop water
availability increases the percentage of area cultivated by almost 6 percentage
points. This may be occurring if short run crops such as soybean and mung bean
are grown on intra-marginal lands.
Results show that the effects of explanatory variables are different for villages
that have no secure property rights (NPR villages). On average NPR villages
cultivate land less intensively than APR villages by 71 percentage points. Additionally in NPR villages, there is almost no effect of a change in travel time to
market (travel time estimate for NPR villages ¼ 0.343 (which is coefficient for log
(travel time estimate) ¼ À2.868 þ coefficient (NPR ¼ 1*Log(travel time
estimate)) ¼ 3.212) ¼ 0.343; z ¼ 0.42; Prob > Chi-square ¼0.67). Short run water
availability also has no effect on intensity of cultivation in NPR villages (the short
run water coefficient in NPR villages ¼ 5.716–4.11 ¼ 1.6; Z-statistic ¼ 1.04;
Prob > Z ¼ 0.30).
To investigate land expansion as measured by village agricultural land, the same
variables are used to explain the equation as used for agricultural intensity. This is
because variables that affect intensity of cultivation should also affect land expansion. Results are presented in column 2, Table 8.8.
32
Results in column (2) show that a 1 % increase in village population leads to a
0.4 % increase in area devoted to agricultural land in the villages in the estimation
sample. BAAC credit use increases agricultural land by 1.1 % in these villages. A
1 % increase in travel time to the market increases the area under cultivation in APR
30 Since the intensity of cultivation is measured as a categorical variable, with each value
representing an interval, I estimate the equations for intensity of cultivation using a random effects
interval regression model. Similar to the procedure followed for the crop area equations, I estimate
a reduced form equation where BAAC credit use is endogenous. The results I discuss here use a
two-step variant of the interval regression model in which the first step estimates a reduced form
model for BAAC credit use, using a random effects probit model. Column (5) is a two-step variant
of the random effects interval regression, where the first stage uses a random effects probit
equation to estimate the model for BAAC credit use. Results from the first stage are reported in
Table 5.17.
31 The different specifications and sensitivity analyses are presented in Puri 2006.
32 I estimate a random effects equation via generalized two stage least squares to estimate the
model for agricultural land. The dependent variable is in logs. In Table 5.16 I present only one
specification. BAAC credit use instrumented for, by using three identifying instruments. These are
proportion of population with compulsory education, travel time to the district and HYV rice
dummy. The results from the first stage random effects equation for BAAC credit use are not
shown here.
146
J. Puri
rice have been harvested.
I now discuss the effect of variables that can be affected by policy on agricultural
land and intensity of cultivation. Results are presented in column 1, Table 8.8
30,31
Results in Table 8.8 show that a 1 % decrease in travel time to market increases the
percentage of agricultural land cultivated by 2.9 % points. Population has no effect
on the intensity of cultivation for either group of villages. Short run crop water
availability increases the percentage of area cultivated by almost 6 percentage
points. This may be occurring if short run crops such as soybean and mung bean
are grown on intra-marginal lands.
Results show that the effects of explanatory variables are different for villages
that have no secure property rights (NPR villages). On average NPR villages
cultivate land less intensively than APR villages by 71 percentage points. Additionally in NPR villages, there is almost no effect of a change in travel time to
market (travel time estimate for NPR villages ¼ 0.343 (which is coefficient for log
(travel time estimate) ¼ À2.868 þ coefficient (NPR ¼ 1*Log(travel time
estimate)) ¼ 3.212) ¼ 0.343; z ¼ 0.42; Prob > Chi-square ¼0.67). Short run water
availability also has no effect on intensity of cultivation in NPR villages (the short
run water coefficient in NPR villages ¼ 5.716–4.11 ¼ 1.6; Z-statistic ¼ 1.04;
Prob > Z ¼ 0.30).
To investigate land expansion as measured by village agricultural land, the same
variables are used to explain the equation as used for agricultural intensity. This is
because variables that affect intensity of cultivation should also affect land expansion. Results are presented in column 2, Table 8.8.
32
Results in column (2) show that a 1 % increase in village population leads to a
0.4 % increase in area devoted to agricultural land in the villages in the estimation
sample. BAAC credit use increases agricultural land by 1.1 % in these villages. A
1 % increase in travel time to the market increases the area under cultivation in APR
30 Since the intensity of cultivation is measured as a categorical variable, with each value
representing an interval, I estimate the equations for intensity of cultivation using a random effects
interval regression model. Similar to the procedure followed for the crop area equations, I estimate
a reduced form equation where BAAC credit use is endogenous. The results I discuss here use a
two-step variant of the interval regression model in which the first step estimates a reduced form
model for BAAC credit use, using a random effects probit model. Column (5) is a two-step variant
of the random effects interval regression, where the first stage uses a random effects probit
equation to estimate the model for BAAC credit use. Results from the first stage are reported in
Table 5.17.
31 The different specifications and sensitivity analyses are presented in Puri 2006.
32 I estimate a random effects equation via generalized two stage least squares to estimate the
model for agricultural land. The dependent variable is in logs. In Table 5.16 I present only one
specification. BAAC credit use instrumented for, by using three identifying instruments. These are
proportion of population with compulsory education, travel time to the district and HYV rice
dummy. The results from the first stage random effects equation for BAAC credit use are not
shown here.
146
J. Puri