likelihood estimates of the frontier and the efficiency equations.29 The production function shows slight
increasing returns to scale with the Cobb Douglas coefficients adding to 1.06. As expected, land and labor
have the largest elasticities in the production function. The coefficient for bullocks is negative and
insignificant, probably because the functional forms does not take into account the fact that bullocks are an
inferior substitute for tractors. The maximum likelihood estimates of the predicted efficiency equation
show that the productive efficiency of farmers is significantly influenced by several characteristics of the
households. For example, households that own more paddy land and assets (bullocks, cattle, tractors,
vehicles and farm implements) are more efficient. Households whose primary occupation is agriculture,
have younger and male heads and have a large proportion of educated members also have greater
productive efficiency. While engaging in non-agricultural activity as a primary activity has a negative
efficiency effect, the type of non-agricultural activity has a positive efficiency effect if household members
work for the government or are self-employed. Government employment probably reflects a selection
effect or gains from better access to information, credit, markets and extension services. Self-employment
may indicate entrepreneurship. Two effects that are somewhat difficult to interpret are the negative
effects of the education of the head (over most sample values) and the proportion of male adults in the
households.30 The education effect may reflect a lesser commitment to agricultural activity due to greater
outside options.
The second column presents the results of the three equation system where fertilizer is treated as
an endogenous input. These estimates are also qualitatively similar to those when all inputs are treated as
exogenous. The skill indices are also very similar with a partial correlation coefficient of 0.94. The third
29 17 district dummy variables are included in both the production frontier and the efficiency equation. These dummy
variables are jointly significant at the 5% level.
30 Since the quadratic term for education is positive, the overall education effect becomes positive for households
with a head who has more than 10.87 years of schooling (using the one step ML estimates). It must be noted,
however, that most households face a negative education effect because 92.2% of sample education values are less
than 10.87 years and the sample mean is only 5.4 years.
34