```Date: Mon, 22 Jan 2007 22:36:15 +0100 Reply-To: "adel F." Sender: "SAS(r) Discussion" From: "adel F." Subject: RE : Re: RE : Re: contribution of independent variables in a random intercept model (mixed NLmixed gllmmix ) Comments: To: David L Cassell In-Reply-To: Content-Type: text/plain; charset=iso-8859-1 Thanks David, My question was about the "contribution of each indepedent vaiable in the variance of the dependent variable". I take your point , that it is not a correct question in the first time, I had some references from othe members of the list Thanks for the suggestions Adel David L Cassell a écrit : adel_tangi@YAHOO.FR wrote back: > > >David L Cassell a écrit : > adel_tangi@YAHOO.FR wrote: > > > >Hi, > > > > I have a continuous depedent variable y and 3 independent variable X1 > >(binary), X2 continuous and X3 (categorical with 3 categories) > > > > I have students nested within schools > > > > I am considering a random intercept model > > > > y = a + X1+X2+X3+ u+ e > > > > u the school residual > > e the student residual > > > > My question how I can compute the contribution of each of X1,X2 and X3 > >in the variation of y, using mixed, NLmixed or gllmmix > > > > > > Thanks a lot for your suggestions > > > > Adel > >This is going to sound like a post I jsut wrote a few seconds ago, but >isn't this a survey sample problem? You have a sample of schools, with >a sample of students pulled from each. That sounds like a survey sample >analysis problem to me. > >This means that the random intercept model may be the wrong approach. >What are you trying to estimate, and why are you using a random >intercept model, and what do you want to achieve (in a larger sense) >with your model? > >HTH, >David >-- >David L. Cassell >mathematical statistician >Design Pathways >3115 NW Norwood Pl. >Corvallis OR 97330 > >Thanks David, > The example of students nested within schools, is the classical example >of multilevel models. > Of course, radom intercept model is just an example. > > My question was about contribution of each indepedent vaiable in the >variance of the dependent variable in a multilevel model. > > Adel First, the 'classic' example of multi-level models is often *wrong*. All too often, the data are from a sample survey, and should be analyzed as such. So I need to know about your data and your meta-data and your data sources before I can even decide on the proc to use, much less what appropriate hypothesis tests would be. Second, typically there is no way on earth to tease out the "contribution of each indepedent vaiable in the variance of the dependent variable". If you have a carefully-designed experimental design with orthogonal factors, then it might be possible. For anything else, it is not reasonable to attempt, because the variability in your Y is not an additive function of the properties of the regressors. I see people all the time, trying to do this, and to figure out the "most important" regressor. These are not logical things to look for. So I *still* do not know what you are trying to get, or what your real problem is, or what your overal goal is. I'm stymied. Sorry, David -- David L. Cassell mathematical statistician Design Pathways 3115 NW Norwood Pl. Corvallis OR 97330 _________________________________________________________________ Turn searches into helpful donations. Make your search count. http://click4thecause.live.com/search/charity/default.aspx?source=hmemtagline_donation&FORM=WLMTAG --------------------------------- Découvrez une nouvelle façon d'obtenir des réponses à toutes vos questions ! Profitez des connaissances, des opinions et des expériences des internautes sur Yahoo! Questions/Réponses. ```

Back to: Top of message | Previous page | Main SAS-L page