Jayesh Kumar wrote:
---------------------begin excerpt from posting---------------------------------
As I want to test whether the Fixed Effect, is at all needed verses OLS?
I am using firm level data to analyze firm's performance.
My equation is something like:
Y_it= (X1_it + X2_it + X3_it + ...+ X9_it) + (Z1_it + Z2_it + Z3_it +Z4_it) +
u_it
where X's and Z's both are firm specific variables.
-----------------------end excerpt from posting---------------------------------
Use the same approach as for the ealier illustration: -test- X's Z's together.
You can also use likelihood-ratio tests, dropping both sets of the predictors
at once from the model that contains them. I've illustrated both below by
slightly modifying the previous illustration do-file to recast the between-
subjects predictors, ind1 and ind2, to within-subjects predictors, x1d through
x9d and z1d through z4d.
For the illustration, I'm assuming that Jayesh is referring to a panel dataset
(modeled with two error terms, u_i and e_it) of 2000 companies evaluated at
each of 15 intervals with two sets (X1 through X9 and Z1 through Z4) of time-
varying covariates. For brevity here, the initial model isn't the default one
that is typical for a blind or naive model-building exercise or hypothesis-
testing exercise, that is, fully saturated with two- and three-way interactions
of covariates and time.
Joseph Coveney
------------------------begin illustration do-file------------------------------
clear
set seed 20030720
set obs 15
set more off
forvalues i = 1/15 {
generate float a`i' = 0.7
quietly replace a`i' = 1 in `i'
}
mkmat a*, matrix(A)
local means "100"
local sd "15"
local dep "dep1"
forvalues i = 2/15 {
local means = "`means'" + ", 100"
local sd = "`sd'" + ", 15"
local dep = "`dep'" + " dep`i'"
}
drawnorm "`dep'", n(2000) means("`means'") /*
*/ sd("`sd'") corr(A) clear
matrix drop A
generate int rid = _n
reshape long dep, i(rid) j(tim)
forvalues i = 1/9 {
generate byte x`i'd = uniform() >= 0.5
}
forvalues i = 1/4 {
generate byte z`i'd = uniform() >= 0.5
}
*
xi: xtreg dep x*d z*d i.tim, i(rid) fe
*
* Testing each alone
*
test x1d x2d x3d x4d x5d x6d x7d x8d x9d
test z1d z2d z3d z4d
*
* Testing both together
*
test x1d x2d x3d x4d x5d x6d x7d x8d x9d, notest
test z1d z2d z3d z4d, accumulate
*
* Likelihood-ratio tests (using -xtreg , mle-)
*
xi: xtreg dep x*d z*d i.tim, i(rid) mle nolog
estimates store A
xi: xtreg dep x*d i.tim, i(rid) mle nolog
estimates store B
xi: xtreg dep z*d i.tim, i(rid) mle nolog
estimates store C
xi: xtreg dep i.tim, i(rid) mle nolog
estimates store D
* Testing Z's
lrtest A B, stats
* Testing X's
lrtest A C, stats
* Testing both
lrtest A D, stats
exit
--------------------------end illustration do-file------------------------------
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
