Re: st: RE: Regressions with combinatorial of variables

[Nick Cox <njcoxstata@gmail.com>]

The procedure here assumes that you know that it makes best sense to
try the variables in precisely that order, and only that order.
Perhaps you know that, but either way that would be a very special
problem.

Nick

2011/2/1 José Luis Chávez Calva <josechc@gmail.com>:

> Thanks Keith, I'm going to implement it to see if it works for my model.

> 2011/2/1 Keith Dear <Keith.Dear@anu.edu.au>:

>> I think this does what you want:
>>
>>
>> * set up some data with weak relationships
>> glo N 131                       // number of x-variables
>> set obs 256
>> forval i=1/$N {
>>        gen var`i'=runiform()
>> }
>> egen Y = rsum(var*)             // outcome variable
>> replace Y = 0.2*Y + rnormal(0,1)
>>
>> * now the work. $X holds a growing list of significant vars
>> glo X
>> forval i=1/$N {
>>        di "`i': $X"
>>        qui reg Y $X var`i'
>>        qui test var`i'
>>        if r(p)<.05 glo X $X var`i'
>> }

*
*   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/