discriminant analysis

Dear forum's members
I conducted discriminant analysis. I interpreted covariances when I compare
differences of between the groups at report.  The juries of the journal
that I sent my manuscript suggest that I must interpret the correlations
instead of covariances. I get confused in the face of this suggestion. As
the Cor = Cov/sdx*sdy, correlation is dependent the group atributes and
therefore it should be used the covariances
thanks in advance

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

[SR Millis-3]

In the 2-group situation, there is little reason to use discriminant function analysis (DCA).  Logistic regreession has many more advantages:

--less restrictive assumptions. DCA assume multivariate normality and homogeneity of variance/cov matrices.

--Interpretation of the results of logistic regression are much more straight forward, eg, the exponentiated coefficients are odds ratios.

Scott Millis




--- On Tue, 7/28/09, [hidden email] <[hidden email]> wrote:

> From: [hidden email] <[hidden email]>
> Subject: discriminant analysis
> To: [hidden email]
> Date: Tuesday, July 28, 2009, 5:25 PM
> Dear forum's members
> I conducted discriminant analysis. I interpreted
> covariances when I compare
> differences of between the groups at report.  The
> juries of the journal
> that I sent my manuscript suggest that I must interpret the
> correlations
> instead of covariances. I get confused in the face of this
> suggestion. As
> the Cor = Cov/sdx*sdy, correlation is dependent the group
> atributes and
> therefore it should be used the covariances
> thanks in advance
>
> =====================
> 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

[SR Millis-3]

Dear Dr Ozer,

I'm confused by your reviewer's comments.  Generally, the way that one interprets discriminant functions is as follows:

--Identify the variable that have the highest and lowest weights on a function. The size of the weight/coefficient tells you  how much a variable contributes to group discimination and the sign tells you the direction of the relationship.

--Examine the structure matrix cofficients, which shows the correlations of variables with the function. Similar to a factor  analysis, you can use the correlations to indentify the underlying construct of a function.  Structure matrix correlations <.30 are typically not interpreted.

--Examine the group centroids for each function and identify the groups having the highest and lowest scores. Those groups arre differentiated the bests by the function.

--You also want to make sure that you have sufficient sample size.  In your case, identify the smallest group of 4 groups.  TThat group should have at least 10 subjects per variable.  In your case, you have 9 variables, so your smallest group needs to have at least 90 subjects.

Scott R Millis, PhD, ABPP (CN,CL,RP), CStat, CSci
Professor & Director of Research
Dept of Physical Medicine & Rehabilitation
Dept of Emergency Medicine
Wayne State University School of Medicine
261 Mack Blvd
Detroit, MI 48201
Email:  [hidden email]
Tel: [phone]
Fax: [phone]


--- On Wed, 7/29/09, arif ozer <[hidden email]> wrote:

> From: arif ozer <[hidden email]>
> Subject: RE: discriminant analysis
> To: [hidden email]
> Date: Wednesday, July 29, 2009, 3:47 PM
>
>
>
> #yiv1795973779 .hmmessage P
> {
> margin:0px;padding:0px;}
> #yiv1795973779 {
> font-size:10pt;font-family:Verdana;}
>
>
>
> Dear Millis
>
> � I have four groups.and nine independent variables. my
> question whether or not I must use correlations or
> covariances?
>
>
>
> Yrd. Doç. Dr. Arif OZER
> Gazi Universitesi
> Meslek eğitim Fakültesi
> Eğitim Bilimleri Bölümü
> Rehberlik ve Psikoljik Danışmanlık Anabilim Dalı
> Beşevler / Ankara
> Cep: 0506 287 72 65
> iş: 0312 226 28 29
> [hidden email]; [hidden email]; [hidden email]
>
>
>
> �
> > Date: Wed, 29 Jul 2009 12:09:32 -0700
> > From: [hidden email]
> > Subject: Re: discriminant analysis
> > To: [hidden email]; [hidden email]
> >
> > In the 2-group situation, there is little reason to
> use discriminant function analysis (DCA). Logistic
> regreession has many more advantages:
> >
> > --less restrictive assumptions. DCA assume
> multivariate normality and homogeneity of variance/cov
> matrices.
> >
> > --Interpretation of the results of logistic regression
> are much more straight forward, eg, the exponentiated
> coefficients are odds ratios.
> >
> > Scott Millis
> >
> >
> >
> >
> > --- On Tue, 7/28/09, [hidden email]
> <[hidden email]> wrote:
> >
> > > From: [hidden email] <[hidden email]>
> > > Subject: discriminant analysis
> > > To: [hidden email]
> > > Date: Tuesday, July 28, 2009, 5:25 PM
> > > Dear forum's members
> > > I conducted discriminant analysis. I interpreted
> > > covariances when I compare
> > > differences of between the groups at
> report.�  The
> > > juries of the journal
> > > that I sent my manuscript suggest that I must
> interpret the
> > > correlations
> > > instead of covariances. I get confused in the
> face of this
> > > suggestion. As
> > > the Cor = Cov/sdx*sdy, correlation is dependent
> the group
> > > atributes and
> > > therefore it should be used the covariances
> > > thanks in advance
> > >
> > > =====================
> > > 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