I am trying to run a simple regression in sas and the results seem to change depending on what order i put the variables in my model. Why?

For specifics, here is my code:

proc reg data=dataset;
model lsalary = x1 x2 x3 x4 x5 x6 x7;
run;

proc reg data=dataset;
model lsalary = x3 x4 x5 x6 x7 x1 x2 ;
run;

Why does the first code give different results from the second code? Does it matter if x1 and x2 are fitted values from a previous regression?

4/16/2009 2:38:42 AM

does it give you the same results when you use proc glm instead?

4/16/2009 1:03:26 PM

because sas sucks?

jk jk

4/16/2009 1:04:32 PM

them's fightin' words, son.

4/16/2009 1:07:01 PM

4/16/2009 1:13:40 PM

Have you tried support.sas.com? They should get you an answer within a few business hours.

4/17/2009 12:16:38 AM

I don't use SAS, but SPSS has something about block and stepwise entry. I can't remember offhand.

4/17/2009 1:19:56 AM

^^lawl, or just that.

i forgot to ask around today at work - i'll do so tomorrow, someone in my building must be able to tell me the answer

4/17/2009 1:57:00 AM

I dunno about our building particular

4/18/2009 11:39:17 PM

(Damn, it's been awhile since I took ST430 and ST435)

The order shouldn't matter in this case. Can you clarify how the results are different?
http://www.tau.ac.il/cc/pages/docs/sas8/stat/chap55/sect24.htm

And you can feed in previous values, done frequently with Quality Control Statistics, where the current batch's quality may be affected by the previous batch's quality. Just make sure you state as such whenever detailing your methods / models.

4/19/2009 8:18:40 AM

Sorry for the question, but, are you sure that youre reading your results correctly? If x1 is 'first' and your 'first' variable in the next equation are not the same, its because theyre listed in the order you put them in.

If not, and Im assuming its not. Try another method?
use proc autoreg or something else?

4/19/2009 6:11:59 PM