R help - Jul 2007 - p-value from survreg

The question was how to get the p-value from the fit below, as an S object
 
sr<-survreg(s~groups, dist="gaussian")
Coefficients:
(Intercept)      groups
-0.02138485  0.03868351

Scale= 0.01789372

Loglik(model)= 31.1   Loglik(intercept only)= 25.4
        Chisq= 11.39 on 1 degrees of freedom, p= 0.00074
n= 16
 

----
  In general, good places to start are 
    > names(sr)
    > help(survreg.object)  
    > ssr <- summary(sr)
    > names(ssr)
As someone else pointed out, it's also easy to look at the print.survreg
function and see how the value was created -- one of the things I love
about S.
    
Unfortunately, doing the above myself showed that I have let the documentation
page for survreg.object get seriously out of date -- quite embarassing as
that is logically the first place to start.

As to the print function creating things "on the fly": there is an
area where
there is no good answer.  Does one make the return object from a fit such
that it contains only minimal data, or add in all of the other computations
that can be derived from these?  The Chambers and Hastie book "Statistical
Models in S", which was the starting point for model objects, leaned
towards
the former, and this still influences many functions.  Often the summary
function will "fill in" these derived values, the std and t-tests for
the individual coefficients for instance.

	Terry T.

On 7/12/07, Terry Therneau <therneau at mayo.edu>
wrote:
> The question was how to get the p-value from the fit below, as an S object
>
> sr<-survreg(s~groups, dist="gaussian")
> Coefficients:
> (Intercept)      groups
> -0.02138485  0.03868351
>
> Scale= 0.01789372
>
> Loglik(model)= 31.1   Loglik(intercept only)= 25.4
>         Chisq= 11.39 on 1 degrees of freedom, p= 0.00074
> n= 16
>
>
> ----
>   In general, good places to start are
>     > names(sr)
>     > help(survreg.object)
>     > ssr <- summary(sr)
>     > names(ssr)
> As someone else pointed out, it's also easy to look at the
print.survreg
> function and see how the value was created -- one of the things I love
> about S.
>
> Unfortunately, doing the above myself showed that I have let the
documentation
> page for survreg.object get seriously out of date -- quite embarassing as
> that is logically the first place to start.
>
> As to the print function creating things "on the fly": there is
an area where
> there is no good answer.  Does one make the return object from a fit such
> that it contains only minimal data, or add in all of the other computations
> that can be derived from these?  The Chambers and Hastie book
"Statistical
> Models in S", which was the starting point for model objects, leaned
towards
> the former, and this still influences many functions.  Often the summary
> function will "fill in" these derived values, the std and t-tests
for
> the individual coefficients for instance.
I think this is where it's nice to have a separate function that does
the filling in - then you can have the best of both worlds.  That's
the role that summary often plays.

Hadley