I run a random effect model and would like to make a robustness test with monte carlo. I am pretty new in this field, and am unsure about the results.
my model
-----------
use "C:\STATA\20190118 fbdata.dta", replace
xtset ID year
xtreg MATH football age sex parentaleducation parentalincome, re
-----------
my monte carlo simulation:
-----------
capture program drop regprog
program regprog, rclass
drop _all
set obs 100
use "C:\STATA\20190118 fbdata.dta", replace
gen u=invnorm(uniform())
xtset ID year
xtreg MATH football+u age sex parentaleducation year
return scalar b0 = _coef[_cons]
return scalar b1 = _coef[football]
return scalar b2 = _coef[age]
return scalar b3 = _coef[sex]
return scalar b4 = _coef[parentaleducation]
return scalar b5 = _coef[year]
return scalar b6 = u
end
simulate b0 = r(b0) b1 = r(b1) b2 = r(b2) b3 = r(b3) b4 = r(b4) b5 = r(b5) b6 = r(b6), reps(100): regprog
sum
histogram b1, kdensity plot(function stdnorm = normalden(x,0,1), lpattern(dash))
----------------
the results seems to look weird.
for the random effect model I get
MATH | |
football | 0.838 |
age | 4.935 |
sex | -0.586 |
parentaleducation | 0.882 |
year | 0.183 |
b0 | -173 |
b1 | -0.041 |
b2 | 4.986 |
b3 | -0.543 |
b4 | 0.878 |
b5 | 0.181 |
b6 | 0.164 |
On my interpretation, the values look all very good. also u is with 0.164 close to 0. but why is b1 so different from the coefficient of football? When I change the random variable to i.e. sex, I have the same phenomenon. All coefficients are almost identical to the monte carlo simulation but not sex, even if the error term has a standard deviation.
can somebody help me to get the point? thank you very much for your appreciate help!
0 Response to Monte carlo simulation to test RE model
Post a Comment