On 7 Jun 2002 06:58:54 -0700, [EMAIL PROTECTED] (Mobile Survey)
wrote:

> Rich Ulrich <[EMAIL PROTECTED]> wrote in message 
>news:<[EMAIL PROTECTED]>...
> > I do most of my scale validating with Pearson correlations, 
> > which are the ones that people are most apt to understand.

> The problem with using the Pearson is not with the semantic
> differential etc. My problem is with a dichotomous variable and one
> ordinal variable.The rest of the variables I would treat as
> continuous.

So, what is the problem?

You can apologize that the correlations are smaller
because a dichotomy loses information compared to the
original continuous variable - if there was one.

A lot of so-called ordinal variables  fare very well when
you treat them as decently-distributed and continuous.
And just do the regular tests and measures.

If you *need*  to keep the ordinal variable as 'ordinal'
(you won't trust its metric, nor apply a simple transformation),
then you could rank-transform that variable.  
What else do you think of?  Dichotomize it?  

Document  your work.  

-- 
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to