Discussion: View Thread

Has anyone built any multinomial logistic regression models? I have been building logistic regression models in the past years but I haven't built any multinomial logistic regression model. While I am looking into it and getting some ideas, I would like to ask here if anyone can share some insight and/or resources about it.

Thanks.


-------------------------------------------
Mei Najim
Project Manager - Predictive Modeling Lead
Sedgwick
-------------------------------------------

Mei:

It is very common in medical research for the response to be ordered multinomial and in these cases, the proportional odds model is the most popular approach. Agresti's Categorical Analysis book is a good intro text for both ordered and non-ordered logistic regression.

Roy Tamura

-------------------------------------------
Roy Tamura
Associate Professor
University of South Florida
-------------------------------------------

If you are using SAS, I wrote a presentation on this:

http://www.statisticalanalysisconsulting.com/wp-content/uploads/2010/10/articleNESUG2010.pdf

-------------------------------------------
Peter Flom
-------------------------------------------

http://www.amazon.com/R-Companion-Applied-Regression-ebook/dp/B008P5BJIO/ref=sr_1_1?s=digital-text&ie=UTF8&qid=1399147564&sr=1-1&keywords=r+companion+to+applied+regression has a nice discussion 
of both unordered (baseline-category logit) and ordered (proportional odds) models and how to fit them in R.

-------------------------------------------
Vadim Bichutskiy
Data Science Consultant
-------------------------------------------







Original Message:
Sent: 05-03-2014 15:51
From: Peter Flom
Subject: Multinomial Logistic Regression

If you are using SAS, I wrote a presentation on this:

http://www.statisticalanalysisconsulting.com/wp-content/uploads/2010/10/articleNESUG2010.pdf

-------------------------------------------
Peter Flom
-------------------------------------------

Some more details about the analysis, may be helpful.

Econometricians like to use multinomial models for "discrete choice" modelling

http://en.wikipedia.org/wiki/Discrete_choice

http://pages.stern.nyu.edu/~wgreene/Curtin2013.htm

-------------------------------------------
Chris Barker, Ph.D.
Consultant and
Adjunct Associate Professor of Biostatistics
www,barkerstats.com

---
"In composition you have all the time you want to decide what to say in 15 seconds, in improvisation you have 15 seconds."
-Steve Lacy
-------------------------------------------

SPSS has extensive documentation for generalized linear models for situations where the values are strictly nominal and where they are ordered. (General liner models are available if interval level of measurement  is plausible.

In addition, the CATREG procedure has built-in capabilities to compare fits etc under all 3 assumptions, strictly nominal, ordinal, and interval level of measurement.

-------------------------------------------
Arthur Kendall
Social Research Consultants
-------------------------------------------

Also, the reasoning  the model is used in depends of whether the underlying construct actually has few values or whether the operational measurement is a coarsening of a continuous construct.

-------------------------------------------
Arthur Kendall
Social Research Consultants
-------------------------------------------

Thanks for all the valuable replys so far. I feel strong support from our group - I apprecaite it!

For now, I am using SAS but I am open to use R as well.

Mei
-------------------------------------------
Mei Najim
Project Manager - Predictive Modeling Lead
Sedgwick
-------------------------------------------







Original Message:
Sent: 05-04-2014 08:28
From: Arthur Kendall
Subject: Multinomial Logistic Regression

Also, the reasoning  the model is used in depends of whether the underlying construct actually has few values or whether the operational measurement is a coarsening of a continuous construct.

-------------------------------------------
Arthur Kendall
Social Research Consultants
-------------------------------------------