Date:         Tue, 30 Nov 2010 11:42:50 -0800
Reply-To:     Bruce Weaver <bruce.weaver@hotmail.com>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         Bruce Weaver <bruce.weaver@hotmail.com>
Subject:      Re: Multinomial Logistic Regression SPSS 16.02
In-Reply-To:  <15D48D8485B6A6429B3F85DBD5AE0FD054845339D7@COM-EXCH.OUCOM.OHIOU.EDU>
Content-Type: text/plain; charset=us-ascii

Bianco, Joseph wrote:
>
> Hello Listers,
>
> Apologies for the remedial nature of this question, but here goes:
>
> I'm conducting a multinomial logistic regression (3 levels of the DV) and,
> based on significant chi-squares/ANOVAs, would like to enter predictors
> hierarchically (i.e., in blocks, as in binary logistic regression and
> regular linear regression). On SPSS, that's not an option for multinomial;
> instead, you can "force entry" certain predictors and then add others in a
> stepwise fashion.  What I'm not clear about is how to conduct this such
> that I force entry of specific predictors in three iterations. If this
> were a binary logistic regression, I would do this:
>
> Block 1:  demographic predictors (age, education)
> Block 2:  treatment factors (treatment condition [2 levels], therapy
> attendance)
> Block 3:  psychotherapy process variables (alliance, group cohesion, group
> similarities)
>
> In other words, I'd like to see if/how the psychotherapy process variables
> predict group membership when predictors from Blocks 1-2 are controlled
> for. I hope this makes sense. Any advice will be appreciated.
>
> Best,
> Joe
>
>

Hi Joe.  It's been a while since I used NOMREG, but I just took a quick look
at it, and it would appear that you have to run a separate NOMREG command
for each step in your hierarchical model building.  (This is the case for
MIXED too, by the way).

One issue you have to be aware of is that when there are more variables in
the later models, you could lose cases that appeared in the earlier models,
due to missing data on later added variables.  That's a problem, because the
comparison of nested models via the change in -2LL is only valid when all
models are computed using exactly the same cases.  A very useful  trick you
can use to get around that problem is to list ALL of the variables that will
appear in the final model each time, but limit which ones are used in a
particular model on the /MODEL sub-command.  E.g.,

* Step 1 .
NOMREG Y (BASE=FIRST ORDER=ASCENDING)
            BY { all categorical variables in the final model }
            WITH { all continuous variables in the final model }
  /MODEL = { all variables forced in on step 1 }
  /INTERCEPT=INCLUDE
  /PRINT=PARAMETER SUMMARY LRT CPS STEP MFI IC.

* Step 2 .
NOMREG Y (BASE=FIRST ORDER=ASCENDING)
            BY { all categorical variables in the final model }
            WITH { all continuous variables in the final model }
  /MODEL = { all step 1 variables plus all step 2 variables }
  /INTERCEPT=INCLUDE
  /PRINT=PARAMETER SUMMARY LRT CPS STEP MFI IC.

* Step 3 .
NOMREG Y (BASE=FIRST ORDER=ASCENDING)
            BY { all categorical variables in the final model }
            WITH { all continuous variables in the final model }
  /MODEL = { all step 1 & 2 variables plus all step 3 variables }
  /INTERCEPT=INCLUDE
  /PRINT=PARAMETER SUMMARY LRT CPS STEP MFI IC.

You'll have to compute your own tests on the change in -2LL from one step to
the next.  Remember that it is distributed (approximately) as chi-squared
with df = the change in the number of model parameters from one step to the
next.

HTH.


-----
--
Bruce Weaver
bweaver@lakeheadu.ca
http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.

--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Re-Guidelines-or-advice-for-data-management-correction-tp3286194p3286617.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@LISTSERV.UGA.EDU (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD