and φ and Φ are the standard normal PDF and CDF18. If the bivariate probit model is used for selection,
the selectivity term is calculated in the same way using the appropriate predicted probability from [43].
The Index of Farming Skills
The biggest obstacle to understanding the role of management or entrepreneurial skills in land
leasing is the absence of an appropriate measure of skill. Management skill or capacity has several
dimensions including “(1) drives and motivations, e.g. farmer’s goals and risk attitude; (2) abilities and
capabilities, e.g. cognitive and intellectual skills; and (3) biography, e.g. background and experience”
[Rougoor et. al. 1998]. While some aspects of skill such as education and age are easily measured, many
others such as drive, motivation, entrepreneurship and ability are neither directly observable nor
quantifiable.
Farm studies have used two distinct approaches to measure management skills. The first, based
on Mundlak [1961], treats management inputs as the time invariant portion of the production function
residual. Skoufias [1991] uses this method to estimate a linear land leasing equation using panel data from
India. However, the panel data fixed effects estimation is clearly unsuitable in our context because many
explanatory variables such as asset ownership have negligible time (within) variation compared to the
cross-sectional (between) variation.19 More fundamentally, by focusing solely on the residual, the panel
data methods fail to make use of the observed dimensions of skill such as education and experience.
The second approach is the inclusion of observed variables that are assumed highly correlated
with managerial skill. Bliss and Stern, Nabi and Skoufias [1991] include age, education and caste as
18 This model is proposed in Lee [1983]. This paper also describes the corrected asymptotic covariance matrix for the
two-step estimator.
19 Skoufias himself points out, for example, his land ownership coefficient is biased as a result. A solution to the fixed
effects problem is proposed by Lanjouw [1999] who models the error structure explicitly with in a joint maximum
likelihood estimation of leasing and production functions.
24