Date:         Thu, 4 Feb 1999 15:29:29 NZST-12NZDT
Reply-To:     Andrew Gray <agray@COMMERCE.OTAGO.AC.NZ>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@UGA.CC.UGA.EDU>
From:         Andrew Gray <agray@COMMERCE.OTAGO.AC.NZ>
Organization: University of Otago
Subject:      Re: Trying to find appropriate technique
In-Reply-To:  <Pine.BSF.3.91.990203153046.18304F-100000@user.xtdl.com>
Content-type: text/plain; charset=US-ASCII

On  3 Feb 99 at 15:36, Donald F. Burrill wrote:
> On Wed, 3 Feb 1999, Andrew Gray wrote:
> > I'm trying to find an appropriate technique to model and/or test for
> > relationships with an ordinal dependent variable and two interval and
> > one nominal independent variables.  Any suggestions would be most
> > appreciated.
>         Is there any compelling reason (apart from the usual
> unreasonable scales-of-measurement prejudices) for not treating the
> dependent variable as though it were interval?  If it behaves
> nicely, use it so;  if not, perhaps a nonlinear transformation would
> help. If the nominal IV has only two levels, it too can be treated
> WLOG as interval;  if it has more than 2 levels, I'd be inclined to
> entertain what used to be called analysis of covariance (with 2
> covariates, and explicitly modelling all definable interactions).
 
I have tried treating it as interval and would be happy to do so
except that on past data sets (the current one is still forthcoming)
none of the standard transformations have resulted in equal variance.
The problem is that there was remarkably high consistency in the
lowest category (1) and very high consistency in the highest category
(7).  The spread-versus-level plot remains the same 'n' shape for all
standard transformations.
 
Thanks very much for your help.
 
Cheers,
Andrew
 
 
Software Metrics Research Laboratory, University of Otago
Phone: +64 3 479 5282 Fax: +64 3 479 8311
email: agray@infoscience.otago.ac.nz
http://divcom.otago.ac.nz:800/COM/INFOSCI/SMRL/home.htm