Skills, Partnerships and Tenancy in Sri Lankan Rice Farms

Fortunately, the potential endogeneity of land in the production function does not affect the power
of our tests in the leasing model because the skill variable is unbiased by definition under the null
hypothesis that farming skills do not affect land leasing decisions. It must be emphasized however that, if
the null hypothesis is rejected, the coefficient of the skill variable in the leasing equation cannot be reliably
interpreted as the magnitude of the effect of skill on leasing since the land input is in fact correlated with
skills.

Specification Issues II: Endogeneity in the Leasing Equation

Another potential problem in our specification is the possible correlation of farming skills and the
error term in the leasing equation. The stochastic two-sided error term removes some of these effects,
such as weather shocks, from the technical efficiency estimates. More factors are removed in the second
stage when the predicted values are obtained. But some unobserved factors that affect the leasing choice
may remain in the skill index if they are correlated with observed farmer characteristics. If, for instance,
the skill index includes some elements of unobserved land quality, the coefficient on technical efficiency in
the leasing model will be biased. We include irrigation and district dummy variables to control for some of
the land quality differences. Any remaining unobserved land quality effects may be picked up by the skill
variables and reduce the power of our tests.

We resort to theoretical arguments to assert that potential endogeneity problems in the skill
variable do not affect our conclusions on contract choice. The null hypothesis that skills do not affect the
choice of contract will be incorrectly rejected if the effect of both skill and unobserved land quality have
the same sign. Fortunately, the theoretical literature on the relation between land quality and contract
choice comes to a different conclusion, that sharecropping is used as a mechanism to minimize “land
mining” or “land quality abuse” by tenants [Murrell 1983, Datta et al 1986, Ray 1998, DeWeaver and
Roumasset 1999]. In these land mining models, fixed-rent contracts are offered only under conditions

31

<<   30   31   32   33   34   35   36   >>

More intriguing information