On Thu, Jan 24, 2013 at 9:53 PM, Doruk Ilgaz <drdoruk@gmail.com> wrote:
> Linearity control when treating a categorical dependent variable as continous
>
> I am estimating firm's rating changes (each category is equally
> spaced) and distribution is symmetrical, there are about 10
> categories.
>
> Instead of ordered-probit I would like to use OLS. I would like to
> check and see if the deviations from the linearity are statistically
> significant in the dependent variable.
There's nothing inherently wrong with using OLS on such a dependent
variable as long as it reasonably satisfies OLS assumptions such as
symmetry. You can check linearity assumptions graphically or using
methods such as constructed variables, for which I recommend seeing A.
C. Atkinson's excellent Plots, Transformations and Regression: An
Introduction to Graphical Methods of Diagnostic Regression Analysis,
Oxford, 1986. Stata has a number of tools for that such as the added
variable plot.
How large is your sample size? If it's pretty large chance are good
that an ordinal model will work OK.
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/
