Re: st: Dependent continuous variable with bounded range

["Joseph Coveney" <jcoveney@bigplanet.com>]

Pavlos C. Symeou wrote:

Stata does not allow any other "family" than "binomial" to be used with
the "logit"  link function. [snip] Considering my problem with bounded
values, would you suggest the use of a different link function that
allows the "Gaussian" or "gamma" distributions (these would be
"identity", "log", "power", and "reciprocal")? Otherwise, should I
continue with my OLS model given that the predicted values stay well
within the possible range?

--------------------------------------------------------------------------------

I thought that one of the features of a sumscore, such as a Likert scale, is
that it tends to to yield normal-like values.  The central moments that you
show seem reasonably normal-like:  not really very skew or leptokurtic.  If
your linear model avoids your main concern, i.e., out-of-bounds predictions,
then you seem to be home-free with it.  Take a look at the residuals, e.g.,
with -qnorm-, -pnorm-, -rvp-.  If they seem acceptable, then aren't you 
done?

If not, then you can look into ordered logistic or ordered probit
regression.  Stata has a few, e.g., -ologit-, -oprobit-, and there are more
user-written extensions, e.g., -gologit-.  If you have panel data, then
there are -gllamm-, -reoprob-, or -cluster()- options in the others.  As Jay
mentions, none of these specifically handles autoregressive covariance.  I
don't know whether that's an overwhelming concern for you.

Joseph Coveney

P.S. What's fl.C20?


*
*   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/