M. Haider Hussain wrote:
Errors were non-normally distributed, that's why I'm using -qreg-. In
other words, I'm not prepared to enforce CLRM assumptions.
[redacted]
> > . . . If I want to compute the joint significance
> > of the regressors, can I still use the F-test given by
> >
> > F=[r2/(k-1)]/[(1-r2)/(n-k)]
> >
> > If this isn't the case, what's the measure of joint significance of
> > the regressors after -qreg- / -bsqreg-?
--------------------------------------------------------------------------------
It seems that the answer to your first question is "no." You can always
check that by a Monte Carlo experiment to see just what the test size is for
the formula after -bsqreg- under a joint null hypothesis.
As to your second question, what's the problem with using -test-, as Maarten
Buis recommended?
Joseph Coveney
Using the example in the online helpfile:
. sysuse auto
(1978 Automobile Data)
. bsqreg price weight length foreign, reps(400)
[output redacted]
. test weight length foreign
( 1) weight = 0
( 2) length = 0
( 3) foreign = 0
F( 3, 70) = 4.65
Prob > F = 0.0051
. local r2 = 1 - e(sum_adev) / e(sum_rdev)
. local k = e(N) - e(df_r)
. display (`r2' / (`k' - 1) ) / ( ( 1 - `r2') / e(df_r) )
7.1577324
*
* 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/
