Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: centering explanatory variables around zero


From   Nikolaos Pandis <npandis@yahoo.com>
To   statalist@hsphsun2.harvard.edu
Subject   st: centering explanatory variables around zero
Date   Sun, 9 Aug 2009 10:50:20 -0700 (PDT)

Hi to all.

I was reading in the text "Stat methods in the SS" by Agresti, and he was explaining with an example how in a linear regression model, for example with 2 predicors, the p value of the predictors may change dramatically between the interaction and the non-interaction model.

y=a + b1x1 + b2x2
vs.
y=a + b1x1 + b2x2 + b3x1x2

He was explaining that you may get similar results for the interaction and no-interaction model by centering the explanatory variables around zero and then running the analyses using the centered explanatory variables.


I have the following model

y=a + b1x1 + b2x2

y and x2 are continuous
x1=categorical with 3 levels

I find a large difference in the p-values for the explanatory variables between the interaction and no-interaction model.

I centered for the continuous variable x2 ((sum x2, gen cx2=x2-r(mean)).

I get similar p values for interaction and no-interaction models.

Question: How about the categorical explanatory variable? Is it appropriate to center this variable also? Does it make any sense?

Many thanks,

Nick 



      
*
*   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index