<>
sdm,
you could try -stepwise- with the -lockterm1- option. This option forces
-stepwise- to hold on to the first term no matter what. You can specify this
first term with parantheses so it can include several covariates, in your
case x1- x200...
*************
sysuse auto, clear
generate weight2 = weight^2
regress price mpg weight2 gear turn headroom foreign disp weight
*gear and mpg are dropped
sw, pr(0.1): regress price mpg weight2 gear turn headroom foreign disp
weight
*keep gear and mpg in there
sw, pr(0.1) lock: regress price (mpg weight2 gear) turn headroom foreign
disp weight
*************
HTH
Martin
-----Ursprüngliche Nachricht-----
Von: [email protected]
[mailto:[email protected]] Im Auftrag von sdm1
Gesendet: Freitag, 30. Januar 2009 02:08
An: [email protected]
Betreff: st: estimation of series of OLS regressions based on t-values from
previous regression
I would like to 'automate' the estimation of a series of OLS regressions.
The regressors can be divided into two sets of variables called, say, x1 -
x200 and x201 - x400 (it's a large data set).
I would like x1 - x200 to be included in every regression. However, I would
like to reduce the number of regressors from the set x201 - x400 so that,
after a number of re-estimations, only those with an absolute t-ratio
greater than 2.00 remain.
I would like to start with all regressors (x1 - x400) present and then
re-estimate having excluded those variables from x201 - x400 whose
(absolute) t-value in the first regression was less than, say, 0.25. On the
basis of the second regression result I would then like to re-estimate again
but excluding those variables from x2001 - x400 whose t-value in the third
regression was less than, say, 0.50. I'd like this process -- of
estimation, dropping variables from x201 - x400 on the basis of their
t-value, re-estimation, and dropping further variables from x201 - x400
using a gradually increasing t-value threshold -- to continue until, after
eight iterations, I am left with x1 - x200 and only those variables from
x201 - x400 whose t-value in the most recent regression is greater than
2.00.
Is it possible to program this?
Thanks.
Steve
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
