🚀 go-pugleaf

On Wed, 5 Feb 1992 22:06:46 GMT <DDS@LEICESTER.AC.UK> said:
   [stuff deleted]
>I have even had one experince where someone has consulted me about the
>output from some analysis they had done.  I pointed out that the
>particular analysis they had done was not really answering what they wanted
>answered as they had done a lot of two-way chi-square tests using the
>CROSSTABS procedure in SPSS, whereas a log-linear modelling approach
>might give them exactly the kind of analysis they wanted and more. The
>reply was that since similar papers had appeared in which CROSSTABS had
>been used to answer very similar questions, they would prefer to stick
>to the "tried and tested methods rather than these new things". After
>some fruitless attempts to change their mind, I ended up by saying that
>I could help them no more and sent them on their way.
   [stuff deleted]
    I agree with most of the sentiments expressed in this posting, but
don't care much for the illustrative example above. Without telling us
what the research objective was, the chi-squares are said to be
inappropriate, as if the superiority of log-linear modeling was
self-evident to all.  I work as a consultant in an educational
environment, where most of my clients are graduate students with varying
degrees of statistical sophistication.  In that context, one of the most
important considerations is whether the student understands the method
they are using.  Much can be accomplished with simple methods if they are
well understood.  Why is it that I see students in the social sciences who
want to do multi-mode maximimum likelihood hierarchial factor analysis,
whereas the engineering types want to know why the slope of their simple
regression line seems to be determined by only a few outlying points?
Okay, I realize this sort of sweeping generalization is unwarranted, but
the main point is the contrast between the complex and the simple, not the
differences among fields of study.  Too often the methodology can become
a more overwhelming concern then the substantive content of the field.

Tim Dorcey                        BITNET:   TCD@CORNELLC
Statistical Software Consultant   Internet: TCD@CORNELLC.CIT.CORNELL.EDU
Cornell Information Technologies  Phone:    (607) 255-5715
Cornell University
Ithaca, NY  14853