🚀 go-pugleaf

In article <STAT-L%92012714295564@VM1.MCGILL.CA>, EDFJHN@GSUVM1.BITNET writes:
>Regarding R.I. Beekun's request.  There is a multiserial correlation in which o
>ne variable is assumed to be continuous while the other exists only in k >= 2 c
>ategories.  These categories are assumed to be formed from an underlying normal
> distribution.  When k=2 the multiserial is the same as the biserial.  The bis
>erial should not be appropriate for the given case as the biserial only applies
> to the dichotomous case.  I can probably find a reference, if anyone is intere
>sted.

What I've seen people do is take an average of the possible biserials.  Does
that necessarily provide an equivalent result, or does the multiserial
coefficient "get at" something else?

___________________________________________________________________________
Mike Kattan                       | Internet: kattan@jetson.uh.edu         
Decision and Information Sciences | Bitnet:   kattan@uhupvm1  kattan@jetson
University of Houston             | Phone:    (713) 749-6789  Fax: 749-6765
===========================================================================