column reports the results when each equation is separately estimated. It is clear from the standard errors
that the maximum likelihood estimates of the predicted efficiency equation are more efficient. The fourth
column reports linear production function where farmer characteristics are directly included as arguments
in a OLS regression. Here again, the standard errors are considerably larger suggesting that the non-linear
specification provides a better fit of the data. The linear and non-linear skill indices have a positive partial
correlation coefficient of 0.74.
The linear index will be used as a test of the normality assumptions of the frontier-based index. In
addition, we also use a fourth skill index where asset ownership variables are excluded as determinants of
skill levels. This allows us to address the potential criticism that our measure may in fact measure risk-
aversion because asset ownership is really a proxy for income and therefore the degree of risk-aversion.
Figure 7 represents the estimated frontier based skill indices (with and without asset ownership) for all
types of farmers. The low skill level of share tenants stands out in both indices, especially compared to
fixed-rent tenants who appear to be more skilled than even owner farmers and landlords. The
extraordinarily high skill level of fixed-rent tenants decreases when asset ownership is excluded, but the
large gap between fixed and share tenants remains.
Probit and Tobit Models
The determinants of share and fixed-rent tenancy can be estimated with separate probit or tobit
equations for each type of tenancy. The age of the household head, the type of non-agricultural activity
conditional on the degree of non-agricultural activity, the proportion of family members with a primary
education and ownership of non-farm input assets (cattle and vehicles) are excluded as identifying
variables of the skill index because they do not have a direct effect on land leasing. The results of these
estimates are given in Table 3. The probit estimates, given in the first two columns, show that farming
skills have a positive effect on choosing fixed-rent contracts but not on sharecropping. In addition,
35