[R] [Q] Goodness-of-fit test of a logistic regression model using rms package

[Bert Gunter]

... and, furthermore, in most real world situations there are several
-- or even "lots" -- of quite different, incomparable models that give
essentially equivalent fits. Distinguishing among the alternatives
typically requires focused studies designed for the task.

Indeed, as Brian Joiner remarked a long time ago (in a galaxy far
away), "Often, even the data aren't sufficient." (This a cryptic
statistical "in" joke; example for the in-crowd: is it an outlier or
an indication of curvature?).

Cheers,
Bert

Bert Gunter
Genentech Nonclinical Statistics

On Wed, Sep 1, 2010 at 12:22 PM, Frank Harrell <f.harrell at vanderbilt.edu> wrote:
>
>
>
> On Wed, 1 Sep 2010, GMail (KU) wrote:
>
>> Hello,
>>
>> I was looking for a way to evaluate the goodness-of-fit of a logistic
>> regression model. After googling, I found that I could use "resid(fit,
>> 'gof')" method implemented in the rms package. However, since I am not used
>> to the "le Cessie-van Houwelingen normal test statistic," I do not know
>> which statistic from the returned from the "resid(fit, 'gof')" call that I
>> could use to evaluate the goodness of fit.
>>
>> When I ran the "resid(fit, 'gof')", I got the following results:
>> ##############################################
>> Sum of squared errors     Expected value|H0                    SD
>>         6844.684594           6805.672315              2.790969
>>                   Z                     P
>>           13.978043              0.000000
>> ##############################################
>>
>> I tried to read the le Cessie and van Houwelingen's original paper, but I
>> found that it required prerequisite knowledge I don't current have.
>> Could someone explain how to interpret the results from "resid(fit, 'gof')
>> call?
>>
>> Any help would be much appreciated.
>>
>> Young-Jin Lee
>
> Young-Jin,
>
> I think everyone has trouble interpreting omnibus tests of lack of fit, so
> don't feel bad.  You just know that something somewhere is probably wrong
> with the model.  I focus on directed tests such as allowing all continuous
> variables to have nonlinear effects or allowing selected interactions, and
> finding out how important the complex model terms are.
>
> Frank Harrell
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>