This is a good one to liven up a Friday (with apologies
to any readers who are not currently experiencing Friday).
My suggestions are fourfold.
1. You would be hard pushed to find clear and consistent
definitions of "parametric" and "nonparametric" within
many literatures, let alone between literatures. They
are terms that often confuse rather than clarify.
2. I doubt that normality of response distribution is
nearly as important for most modern modelling methods
as many introductory courses or texts state or imply.
At most normality of conditional distributions is ideal
for some methods but not even essential for many purposes.
3. Likert scales in the sense you use the term
appear to be discrete ordered scales. The normal
distribution is for continuous variables. No
transformation really can bridge that gap.
A set of spikes remain a set of spikes no matter
how much you stretch or shrink the supporting space.
4. Several excellent methods are available
for ordered (ordinal) scales. I would start by looking
at the manual entry for -ologit- and then follow
links.
Nick
[email protected]
Jen McCormick
> I have a question regarding using parametric versus non-parametric
> analyses.
>
> A colleague and I conducted a national survey (n=845). Much
> of our data
> are from Likert-scale based questions. We have been exploring the
> options for analysis and have found some differences of
> opinion in the
> literature on whether parametric methods should be used with
> these kind
> of data (versus nonparametric). My question is whether
> anyone has had
> experience analyzing Likert scale data (5 point, anchored
> scale) using
> parametric methods and how you transformed the data to correct for
> non-normal distribution.
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
