ROBUST ESTIMATION AND OUTLIER DETECTION IN FARM AND NONFARM INCOME - AN EMPIRICAL STUDY
Loading...
Date
41618
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Agricultural Sciences GKVK, Bangalore
Abstract
The present study is an effort to detect outliers in farm and
nonfarm income and use them in estimation for the data obtained
from the Cauvery command area of Karnataka. The study area
divided into three reaches as upper, mid and lower reach has data
on three hundred households for each reach and this is used to
detect outliers and parameters are estimated for each reach and
entire study area. Outliers detected by using robust procedure
based on Mahalanobis distance, resulted in 14, 16, 16 and 51
outliers for farm income considered as dependent, whereas 8, 6, 7
and 25 outliers are detected when nonfarm income is considered as
dependent variable for three reaches and for entire study area,
respectively. Conventional MLR method and robust regression
methods like MM and LTS methods are used for estimation and
comparison is made among three methods. Dependent variable
farm income has significant contribution from age, education,
working adults and persons employed in upper and mid reach. In
lower reach only age of persons is significant. For entire study area
education, family size and working adults are significantly
contributing to farm income. Age, family size and farm income are
significantly contributing for nonfarm income as dependent
variable in upper and mid reach. In lower reach only farm income
is significant. For entire study area education, persons employed
and farm income significantly contribute to nonfarm income in
both the methods. Finally estimates by LTS method have lesser
standard errors compared to MM and MLR method.