multinomial logistic - predicting group membership

[yusof]

I have a multinomial logistic model where the dependent has 3 categories and the independent variables are 5 scale variables. I want to use the model to predict group membership manually based on the indeopendents and the coefficients for new data for which group membership is not available. I have seen examples of doing this while holding other variables constant and varying one independent. How do I do this by letting all the independents to vary freely - can anyone share the algorithm which I can implement in Excel?

Thanks ain advance

[Andy W]

1) This is an SPSS list - so you will not get answers to how to do this in Excel.

2) If you paste the syntax you use to estimate the model you will be able to get better answers to your problem.

If you used a model that can save a parameter file (e.g. NOMREG) then you can estimate the model, save the parameter file, and then score the new data. If you used a model that can not save the parameter file (e.g. PLUM), one trick is to add the cases into the current file with the dependent variable missing and estimate the model. SPSS will still provide predicted estimates from whatever regression model, but the model will only be fit on the non-missing data.

You can of course just calculate the predictions yourself by using COMPUTE statements, but these options are typically much more convenient.

[Art Kendall]

To start a discussion on the list What would be the pros and cons of using discriminant function
vs multinomial logistic in an instance like this where there are 3 groups and 5 scale variables?
On 5/22/2014 8:10 AM, Andy W [via SPSSX Discussion] wrote:

1) This is an SPSS list - so you will not get answers to how to do this in Excel.

2) If you paste the syntax you use to estimate the model you will be able to get better answers to your problem.

If you used a model that can save a parameter file (e.g. NOMREG) then you can estimate the model, save the parameter file, and then score the new data. If you used a model that can not save the parameter file (e.g. PLUM), one trick is to add the cases into the current file with the dependent variable missing and estimate the model. SPSS will still provide predicted estimates from whatever regression model, but the model will only be fit on the non-missing data.

You can of course just calculate the predictions yourself by using COMPUTE statements, but these options are typically much more convenient.

---

If you reply to this email, your message will be added to the discussion below:

http://spssx-discussion.1045642.n5.nabble.com/multinomial-logistic-predicting-group-membership-tp5726180p5726181.html

To start a new topic under SPSSX Discussion, email [hidden email]
To unsubscribe from SPSSX Discussion, click here.
NAML

[Maguin, Eugene]

As Ryan said in another context last week or the week before, it's good to understand how the computations work. If you search the archives for a message Ryan posted on Mon 2/24/2014 1:05 PM, subject line Genlin, you'll see how the computations work for both ordinal and nominal regressions. Then make use of what Andy posted. Gene Maguin

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of yusof
Sent: Thursday, May 22, 2014 7:55 AM
To: [hidden email]
Subject: multinomial logistic - predicting group membership

I have a multinomial logistic model where the dependent has 3 categories and the independent variables are 5 scale variables. I want to use the model to predict group membership manually based on the indeopendents and the coefficients for new data for which group membership is not available. I have seen examples of doing this while holding other variables constant and varying one independent. How do I do this by letting all the independents to vary freely - can anyone share the algorithm which I can implement in Excel?

Thanks ain advance



--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/multinomial-logistic-predicting-group-membership-tp5726180.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
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

[yusof]

this is my syntax, I am using nomreg. What i want to know is how to score new data?

NOMREG QCL_1 (BASE=LAST ORDER=ASCENDING) BY Q66 Q69 Q76 Q77 Q132 Q133 Q139
  /CRITERIA CIN(95) DELTA(0) MXITER(100) MXSTEP(5) CHKSEP(20) LCONVERGE(0) PCONVERGE(0.000001)
    SINGULAR(0.00000001)
  /MODEL
  /STEPWISE=PIN(.05) POUT(0.1) MINEFFECT(0) RULE(SINGLE) ENTRYMETHOD(LR) REMOVALMETHOD(LR)
  /INTERCEPT=INCLUDE
  /PRINT=CELLPROB CLASSTABLE FIT PARAMETER SUMMARY LRT CPS STEP MFI.

[yusof]

also, I do not want to know "how to do it in excel". I want to know the SPSS computations so that I can implement them in Excel if needed. So this is still an spss question

[David Marso]

Click Help>Algorithms!
Presumably you have sufficient grasp of the math to read them and apply them?

yusof wrote

also, I do not want to know "how to do it in excel". I want to know the SPSS computations so that I can implement them in Excel if needed. So this is still an spss question

[Ryan]

Use Example 3 from this website:

http://www.ats.ucla.edu/stat/spss/dae/mlogit.htm

Data for Example 3:

http://www.ats.ucla.edu/stat/data/hsbdemo.sav

Demonstration of how to calculate predicted probabilities:

http://listserv.uga.edu/cgi-bin/wa?A2=ind1302&L=spssx-l&P=R40278

Ryan

Sent from my iPhone

On May 23, 2014, at 5:59 AM, yusof <[hidden email]> wrote:

also, I do not want to know "how to do it in excel". I want to know the SPSS
computations so that I can implement them in Excel if needed. So this is
still an spss question



--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/multinomial-logistic-predicting-group-membership-tp5726180p5726200.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
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

[Jon K Peck]

Save the model as xml, and then use the scoring wizard (under Utilities) to generate appropriate scoring syntax.


Jon Peck (no "h") aka Kim
Senior Software Engineer, IBM
[hidden email]
phone: [phone]




From:        yusof <[hidden email]>
To:        [hidden email],
Date:        05/23/2014 03:44 AM
Subject:        Re: [SPSSX-L] multinomial logistic - predicting group membership
Sent by:        "SPSSX(r) Discussion" <[hidden email]>

---

this is my syntax, I am using nomreg. What i want to know is how to score new
data?

NOMREG QCL_1 (BASE=LAST ORDER=ASCENDING) BY Q66 Q69 Q76 Q77 Q132 Q133 Q139
 /CRITERIA CIN(95) DELTA(0) MXITER(100) MXSTEP(5) CHKSEP(20) LCONVERGE(0)
PCONVERGE(0.000001)
   SINGULAR(0.00000001)
 /MODEL
 /STEPWISE=PIN(.05) POUT(0.1) MINEFFECT(0) RULE(SINGLE) ENTRYMETHOD(LR)
REMOVALMETHOD(LR)
 /INTERCEPT=INCLUDE
 /PRINT=CELLPROB CLASSTABLE FIT PARAMETER SUMMARY LRT CPS STEP MFI.





--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/multinomial-logistic-predicting-group-membership-tp5726180p5726199.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
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