interaction in multinomial logistic

[Zdaniuk, Bozena-2]

Hello, I am running a multinomial logistic with factor predictor A (three categories), continuous predictor B and an interaction term A*B. (syntax below).
For some reason, the results do not show the main effect for B ("this parameter is set to zero because it is redundant").
I have hard time understanding why it is redundant. When running similar model in binary logistic regression, I do get all the main effects. Anybody willing to share some wisdom on that?
thanks so much.
Bozena

NOMREG
  grouping  (BASE=FIRST ORDER=ASCENDING) BY A WITH B C D
  /CRITERIA CIN(95) DELTA(0) MXITER(100) MXSTEP(5) CHKSEP(20) LCONVERGE(0)
  PCONVERGE(0.000001) SINGULAR(0.00000001)
  /MODEL= A*B A B C D
  /STEPWISE = PIN(.05) POUT(0.1) MINEFFECT(0) RULE(SINGLE) ENTRYMETHOD(LR)
  REMOVALMETHOD(LR)
  /INTERCEPT =INCLUDE
  /PRINT = PARAMETER SUMMARY LRT CPS STEP MFI .


________________________________________

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (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

[Anthony Babinec]

On the face of it, A should have one redundant parameter
and the A*B interaction should have one redundant parameter.
This comes about because of NOMREG's "set to 0" parameterization
of categorical variables. Double-check the messages to see
whether this is not in fact the case. Just speculating - Instead
of a stepwise method, try enter. If the A*B interaction is in
the model, you usually want the respective main effects in,
and this is not guaranteed with stepwise. Also, re-write the MODEL
statement to list A*B last.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Zdaniuk, Bozena
Sent: Tuesday, May 13, 2008 12:16 PM
To: [hidden email]
Subject: interaction in multinomial logistic

Hello, I am running a multinomial logistic with factor predictor A (three
categories), continuous predictor B and an interaction term A*B. (syntax
below).
For some reason, the results do not show the main effect for B ("this
parameter is set to zero because it is redundant").
I have hard time understanding why it is redundant. When running similar
model in binary logistic regression, I do get all the main effects. Anybody
willing to share some wisdom on that?
thanks so much.
Bozena

NOMREG
  grouping  (BASE=FIRST ORDER=ASCENDING) BY A WITH B C D
  /CRITERIA CIN(95) DELTA(0) MXITER(100) MXSTEP(5) CHKSEP(20) LCONVERGE(0)
  PCONVERGE(0.000001) SINGULAR(0.00000001)
  /MODEL= A*B A B C D
  /STEPWISE = PIN(.05) POUT(0.1) MINEFFECT(0) RULE(SINGLE) ENTRYMETHOD(LR)
  REMOVALMETHOD(LR)
  /INTERCEPT =INCLUDE
  /PRINT = PARAMETER SUMMARY LRT CPS STEP MFI .


________________________________________

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (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

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (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

[Zdaniuk, Bozena-2]

Putting the A*B last did the trick. Thanks!!

Would someone also be willing to have a look at my attempt to interpret  categorical*continuous interaction in Multinom Logistic Regr (predicting no event vs event1; vs event2):
I am running a multinomial logistic with factor predictor Race (three
categories), continuous predictor depression and an interaction term Race*Depr predicting two separate events.
I find that interaction term for category 2 of Factor Race (Hispanics*depr)  significantly predicts event 1 (B=.08).
Here is my interpretation:
Higher depression is associated with higher probability of event 1 for Hispanics more so than for non-Hispanics.
Am I correct?
Bozena

________________________________________
From: SPSSX(r) Discussion [[hidden email]] On Behalf Of Anthony Babinec [[hidden email]]
Sent: Tuesday, May 13, 2008 4:43 PM
To: [hidden email]
Subject: Re: interaction in multinomial logistic

On the face of it, A should have one redundant parameter
and the A*B interaction should have one redundant parameter.
This comes about because of NOMREG's "set to 0" parameterization
of categorical variables. Double-check the messages to see
whether this is not in fact the case. Just speculating - Instead
of a stepwise method, try enter. If the A*B interaction is in
the model, you usually want the respective main effects in,
and this is not guaranteed with stepwise. Also, re-write the MODEL
statement to list A*B last.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Zdaniuk, Bozena
Sent: Tuesday, May 13, 2008 12:16 PM
To: [hidden email]
Subject: interaction in multinomial logistic

Hello, I am running a multinomial logistic with factor predictor A (three
categories), continuous predictor B and an interaction term A*B. (syntax
below).
For some reason, the results do not show the main effect for B ("this
parameter is set to zero because it is redundant").
I have hard time understanding why it is redundant. When running similar
model in binary logistic regression, I do get all the main effects. Anybody
willing to share some wisdom on that?
thanks so much.
Bozena

NOMREG
  grouping  (BASE=FIRST ORDER=ASCENDING) BY A WITH B C D
  /CRITERIA CIN(95) DELTA(0) MXITER(100) MXSTEP(5) CHKSEP(20) LCONVERGE(0)
  PCONVERGE(0.000001) SINGULAR(0.00000001)
  /MODEL= A*B A B C D
  /STEPWISE = PIN(.05) POUT(0.1) MINEFFECT(0) RULE(SINGLE) ENTRYMETHOD(LR)
  REMOVALMETHOD(LR)
  /INTERCEPT =INCLUDE
  /PRINT = PARAMETER SUMMARY LRT CPS STEP MFI .


________________________________________

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (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

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (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

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (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