Email discussion lists for the UK Education and Research communities

allstat@JISCMAIL.AC.UK

View:

 Message: [ First | Previous | Next | Last ] By Topic: [ First | Previous | Next | Last ] By Author: [ First | Previous | Next | Last ] Font: Proportional Font

Options

Subject:

Re: Discriminant Analysis Vs Multinomial Logistic Regression

From:

Date:

Fri, 16 Mar 2012 17:05:45 +0000

Content-Type:

text/plain

Parts/Attachments:

 text/plain (84 lines)
 ```Dear Nigel Caveat emptor, but ..... My understanding is that their aims are similar, but Discrim Anal is simply is simply a data analysis tool which does not require a probability model, while MLR is probabilistic in essence. In the 2-group case they may lead to the same 'solution' if you use suitable assumptions for MLR. However, in k-group analysis, the partitions in MLR are all "parallel" to each other (i.e. they produce 'slices' of n-space), but the (k-1) partitions in discrim anal each 'halve' the space and the cuts are not necessarily parallel. (They depend on successive eigenvectors of the W^(-1)T matrix.) I seem to recall there is a variant of discim analysis which produces parallel slices, but the dominant variant is that developed by Fisher which I have described above. However, I may be wrong. I'd like to say it's all in Mardia, Kent and Bibby, and if it is I am sure you can rely upon it. But MLR is a post-MKB development so you will have to wait for the next edition (and wait, and wait ......!) JOHN BIBBY On 16 March 2012 13:28, Nigel Marriott <[log in to unmask]> wrote: > Dear Allstatters, > > > > Can anyone clarify for me what is the fundamental difference between > Discriminant Analysis & Multinomial Logistic Regression?  When I look these > up on Wikipedia, neither article makes any reference to each other yet to my > mind both have the same goal in mind i.e. to classify objects into one of > multiple categories using a number of predictor/explanatory variables.  I > ask because my version of R doesn’t have Discriminant Analysis installed but > it does have Multinomial Logistic Regression and I wondering if there is a > need to install an R package to do so. > > > > Regards > > > > Nigel Marriott > > Chartered Statistician > > www.marriott-stats.com > > > > Ground Floor, 21 Marlborough Buildings, Bath BA1 2LY, United Kingdom > > Tel (mobile)  +44 (0)773 4069997 > > Tel (office)    +44 (0)1225 489033 > > Fax                  +44 (0)870 6221969 > > > > Marriott Statistical Consulting Limited, Company No. 5577275, VAT No. > 883304029 > > Registered in England,   Registered Office – Equity House, 4-6 School Road, > Tilehurst, Reading, RG31 5AL > > > > You may leave the list at any time by sending the command > > SIGNOFF allstat > > to [log in to unmask], leaving the subject line blank. You may leave the list at any time by sending the command SIGNOFF allstat to [log in to unmask], leaving the subject line blank. ```