Quarterly Journal of International Agriculture No. 2/06
A comparison of parametric and semi-parametric estimators of farm profit-accounting for sample design in household surveys
Dean Jolliffe
Economic Research Service, USDA, Washington, DC, USA, and the
Institute for the Study of Labor (IZA), Bonn, Germany
Abstract
Much empirical research assumes that data are selected from a sample
frame using a simple random draw, when in practice farm-household
survey data most typically results from complex, multi-stage sample
designs. This paper shows that understanding sample design is important
to determining an appropriate estimation strategy, and argues that the
least absolute deviations (LAD) and censored least absolute deviations
(CLAD) regression are useful tools when using survey data which comes
from a multi-stage draw. LAD and CLAD estimators are used to predict
the effect of education on farm and non-farm income and labor supply.
CLAD standard errors, which are robust to violations of
homoscedasticity and independence, are generated by a boot-strap method
which replicates the two-stage sample design.
Keywords: farm households, off-farm labor, human capital, least absolute deviations, censored least absolute deviations, sample design, Ghana
JEL: C14, C42, I2
Vol. 45 (2006), No. 2: 107-125