mixed model residuals computation and plotting

[Maguin, Eugene]

I know that mixed provides fixed predicted values,  predicted (fitted = fixed+random) values  and residuals (observed-predicted). Is it possible to compute the level 2 residuals from these predicted and residuals data? How would it be done computationally? Singer and Willet mention this in their book and may have offered directions for computing these on their supplementary website but that seems to be gone. UCLA/ats-idre has examples but not those details, so far as I can tell.

Thanks, Gene Maguin

[Ryan]

http://listserv.uga.edu/cgi-bin/wa?A2=ind1403&L=spssx-l&P=R35311

On Jun 17, 2014, at 5:17 PM, "Maguin, Eugene" <[hidden email]> wrote:

I know that mixed provides fixed predicted values,  predicted (fitted = fixed+random) values  and residuals (observed-predicted). Is it possible to compute the level 2 residuals from these predicted and residuals data? How would it be done computationally? Singer and Willet mention this in their book and may have offered directions for computing these on their supplementary website but that seems to be gone. UCLA/ats-idre has examples but not those details, so far as I can tell.

Thanks, Gene Maguin

[Maguin, Eugene]

Thanks, Ryan.

 

I looked at the two posts but they don’t seem to answer what I was looking for.

Your reply to Carol is the relevant email but it looks as she was asking about a single random term model, maybe an intercept only model.

Let’s take a specific example.

 

mixed y with xc w z/fixed=xc w z xc*w xc*z/

   method=ml/print solution/random intercept xc | subjectid(clus) covtype(un)/

   save=fixpred(fpy) pred(tpy) resid(eresid).

 

I write the composite equation as

y = (g00 + g01*w + g02*z + u0) + xc*(g10 + g11*w + g12*z + u1) + e

 

perhaps I’m wrong but I think resid (y – pred) is u0 + xc*u1 + e.

I assume that ‘e’ in the composite equation is y – fixpred.

If this assumption correct, can I extract/compute u0 and u1?

And if it’s not correct, how do I get all three?

 

Gene Maguin

 

 

From: GMAIL [mailto:[hidden email]]
Sent: Tuesday, June 17, 2014 5:38 PM
To: Maguin, Eugene
Cc: [hidden email]
Subject: Re: mixed model residuals computation and plotting

 

http://listserv.uga.edu/cgi-bin/wa?A2=ind1403&L=spssx-l&P=R35311

 

On Jun 17, 2014, at 5:17 PM, "Maguin, Eugene" <[hidden email]> wrote:

I know that mixed provides fixed predicted values,  predicted (fitted = fixed+random) values  and residuals (observed-predicted). Is it possible to compute the level 2 residuals from these predicted and residuals data? How would it be done computationally? Singer and Willet mention this in their book and may have offered directions for computing these on their supplementary website but that seems to be gone. UCLA/ats-idre has examples but not those details, so far as I can tell.

Thanks, Gene Maguin