Re: st: Dependent ordered variable

[David Hoaglin <dchoaglin@gmail.com>]

Hi, Tobias.

To me, the credit ratings look like the ones given by rating agencies
to the debt of entities of government.

For some dependent variables, I might expect the influence (or effect)
of credit rating to be monotonic --- the better the credit rating, the
higher (or lower) the average value of the dependent variable --- but
not linear.  Thus, as others have suggested, I would use an indicator
variable (I'm with Nick!) for each category of credit rating except
the category that you choose as the reference category (T has given
the coding for using the first category or the last category).  Then I
would look at the relation of the fitted effects (with 0 for the
reference category, which will be part of the constant term) to the
categories.

Those effects are adjusted for the other predictors in the regression.
 You refer to them as "control variables," but your data are likely to
be observational.  If so, "control variables" is a misnomer, because
nothing is under control (as in a designed experiment).

David Hoaglin

On Mon, May 28, 2012 at 3:52 AM, Tobias Bertschinger
<tobias.bertschinger@uzh.ch> wrote:
> Hi
> I'm currently analyzing data which have an approximately normal distributed dependent variable Y (dependent_var) and I also have a couple of dependent variables.
> I want to test the influence of credit ratings (ordered, e.g. AAA, AA+, AA, AA-...) as the main independent variable and a couple of control variables.
>
> What kind of test should i use?
> Are there other tests than:
> (1) include the credit rating as a dummy variable
> (2) use xi: regress dependent_var i.credit_rating control_vars
> (3) use xi3: regress dependent_var b.credit_rating control_vars
> ?
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/