On survival models, others can help much better than I.
When R-squared is mentioned, I reach for my prejudices.
The first thing to do is
. search rsquared, faq
for all kinds of (indirectly) relevant discussion.
In a nutshell, there is no such thing as "the R-squared".
Some commands calculate measures that are, or are
analogous to, a squared correlation between observed and
fitted (predicted). Some commands calculate measures
that are based on a comparison of likelihoods. Other
commands calculate neither. Some of these measures
are honestly labelled "pseudo", although why such
measures should be taken seriously is often unclear.
The desire to encapsulate all the merits and demerits
of a model in a single measure really is rather primitive,
although most of us cannot resist most of the time.
Depending on what the problem is, two sets of estimates
can be compared directly, via some simple plot or by
looking at differences, ratios, etc.
Nick
n.j.cox@durham.ac.uk
Nishant Dass
> I want to estimate a hazard-rate model and I would like to
> compare estimates from -streg- with those from -stcox-,
> however, I am puzzled as to why -streg- and -stcox- report
> different statistics for the overall fit of the model (I
> used -ereturn list to check this).
>
> More specifically:
> streg stcox
> R-squared NO YES
> Wald chi-2 YES YES
> p-value for Wald YES NO
>
> So, my question is: is there a specific reason why the
> R-squared is reported in -streg-? (While I can calculate
> the p-value of Wald test for the dozens of Cox
> proportional-hazard regressions I have, it would have been
> easier if I could just -outreg- it so I also wonder why is
> the p-value not reported for -stcox-.)
>
> If anyone has a clue, could you please tell me? Thank you.
*
* 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/