Subscribe Bookmark RSS Feed

Multinomial logistic regression in JMP

abba0046

Community Member

Joined:

Jun 5, 2014

I am trying to perform multinomial logistic regression in JMP.  I am using example from IDRE-UCLA (http://www.ats.ucla.edu/stat/sas/dae/mlogit.htm).  In this this example (hsbdemo data set) how can I specify the baseline category for prog using (ref = "2") and the reference group for ses using (ref = "1") while working in JMP?

1 ACCEPTED SOLUTION

Accepted Solutions
Solution

Hello,

I think you can specify the reference level by using the Value Ordering column property for prog.

For the ses variable, JMP and SAS use different parameterizations, so to get JMP to match, you would need to create indicator columns yourself.

For the prog Value Ordering property, make sure that the column is Nominal or Ordinal. Then make sure that the value '2' is listed last in the Value Ordering property (right click on the column heading, select Column Info, Column Properties, Value Ordering to add/edit the property).

For the ses indicator columns, create a column ses2 that equals 1 if ses=2 and 0 otherwise. Create an analogous column ses3 set equal to 1 if ses=3 and 0 otherwise. Then run Fit Model with Y=prog and Effects ses2, ses3, and write.  The resulting logistic regression should match what is on the IDRE-UCLA page.

A colleague at JMP that helped me with this also pointed out that you can get SAS to match JMP's default results by using the param=effect option in the class statement in PROC LOGISTIC, like this:

proc logistic data = hsbdemo;

class prog ses / param=effect;

model prog = ses write / link = glogit;

run;

Hope this helps!

Michael

Michael Crotty
Sr Statistical Writer
JMP Development
1 REPLY
Solution

Hello,

I think you can specify the reference level by using the Value Ordering column property for prog.

For the ses variable, JMP and SAS use different parameterizations, so to get JMP to match, you would need to create indicator columns yourself.

For the prog Value Ordering property, make sure that the column is Nominal or Ordinal. Then make sure that the value '2' is listed last in the Value Ordering property (right click on the column heading, select Column Info, Column Properties, Value Ordering to add/edit the property).

For the ses indicator columns, create a column ses2 that equals 1 if ses=2 and 0 otherwise. Create an analogous column ses3 set equal to 1 if ses=3 and 0 otherwise. Then run Fit Model with Y=prog and Effects ses2, ses3, and write.  The resulting logistic regression should match what is on the IDRE-UCLA page.

A colleague at JMP that helped me with this also pointed out that you can get SAS to match JMP's default results by using the param=effect option in the class statement in PROC LOGISTIC, like this:

proc logistic data = hsbdemo;

class prog ses / param=effect;

model prog = ses write / link = glogit;

run;

Hope this helps!

Michael

Michael Crotty
Sr Statistical Writer
JMP Development