I am using a panel dataset, where my unit of observation is a village. The
dataset
spans 10 years, with data collected every two years. I have an unbalanced
dataset with the maximum no. of time periods being 6, and minimum being 1.
I am examining land devoted to agriculture and have - a la Woolridge - many
'corner solutions' i.e. many zeros.
I use xttobit on the data, after setting id variable as i(village). Quadchk
tells
me that my specifications are unstable. Indeed my rho (percentage of
variance
explained by intra-panel variance) is quite high, varying from 0.6 to 0.75.
While thinking about this problem, I realize that the data is probably
collected
in two stages - to wit, the district is selected first and then the village.
Thus, I am
led to believe that my observations are not independent within a district
but
are independent across districts.
In response, instead of setting the id variable
as i(village) I set it as i(district).
Is my thinking erroneous?
I am also unclear about how STATA's estimation is changing. I realize that
in
a perfect world I would want to use the cluster() world for this, but
xttobit
does not have that option. If changing the id() option is not changing
making
the standard errors robust (and ofcourse my coefficient estimates also
change,
since my 'unit of observation' is now the district), what is it doing?
Many thanks for clues.
-Jyotsna
Jyotsna (Jo) Puri
Cell and Voicemail: 201-805-1690 [email protected]
Doctoral Candidate, Department of Agriculture and Resource Economics,
University of Maryland, College Park, MD
