multinomial log. reg

[Scott Roesch]

Hi all:

I would like to conduct a hierarchical multinomial logistic regression.
Using the NOMREG procedure I do see an option under the Model Tab that
allows one to Customize the order of entry. However, this does not appear
to allow one to enter predictors on different steps (as does the Binary
Logistic Regression procedure in SPSS). Can someone shed a little light on
how one can enter predictors at different steps for a multinomial
regression (either via the dialog box or syntax)? Thanks in advance!

Scott

******************************************
Scott C. Roesch, Ph.D.
Department of Psychology
San Diego State University
5500 Campanile Drive
San Diego, CA 92182-4611

email: [hidden email]
office phone: [phone]
fax: [phone]
******************************************

[Dale Glaser]

Hi Scott..I'm not sure if this is an appropriate strategy but what I've done in the past with hierarchical multinomial logistic regression (since as you correctly stated there isn't an option to run a forced entry model), is to run the models separately as follows:

  Model1:

  NOMREG
  dv (BASE=LAST ORDER=ASCENDING) BY iv1
  /CRITERIA CIN(95) DELTA(0) MXITER(100) MXSTEP(5) CHKSEP(20) LCONVERGE(0) PCONVERGE(0.000001) SINGULAR(0.00000001)
  /INTERCEPT =INCLUDE
  /model = iv1
  /PRINT = PARAMETER SUMMARY LRT CPS STEP MFI .


  Model2:

  NOMREG
  dv (BASE=LAST ORDER=ASCENDING) BY iv1 iv2
  /CRITERIA CIN(95) DELTA(0) MXITER(100) MXSTEP(5) CHKSEP(20) LCONVERGE(0) PCONVERGE(0.000001) SINGULAR(0.00000001)
  /INTERCEPT =INCLUDE
  /model = iv1 iv2
  /PRINT = PARAMETER SUMMARY LRT CPS STEP MFI .

  and compare parameter estimates, the pseudo r-squares and -2LL; unfortunately information-theoretric indices such as the AIC/BIC are not offered to compare non-nested models, but if one argues for M1and M2 being more and less restrictive models, one can argue that the differences between -2LL are chi-square distributed, and akin to SEM, conduct a chi-square difference test.....

  I will look forward to others on this listserv if my strategy is problematic!

  Dale

"Scott C. Roesch" <[hidden email]> wrote:
  Hi all:

I would like to conduct a hierarchical multinomial logistic regression.
Using the NOMREG procedure I do see an option under the Model Tab that
allows one to Customize the order of entry. However, this does not appear
to allow one to enter predictors on different steps (as does the Binary
Logistic Regression procedure in SPSS). Can someone shed a little light on
how one can enter predictors at different steps for a multinomial
regression (either via the dialog box or syntax)? Thanks in advance!

Scott

******************************************
Scott C. Roesch, Ph.D.
Department of Psychology
San Diego State University
5500 Campanile Drive
San Diego, CA 92182-4611

email: [hidden email]
office phone: [phone]
fax: [phone]
******************************************



Dale Glaser, Ph.D.
Principal--Glaser Consulting
Lecturer--SDSU/USD/CSUSM/AIU
3115 4th Avenue
San Diego, CA 92103
phone: [phone]
fax: [phone]
email: [hidden email]
website: www.glaserconsult.com