The H-L test is now considered to be obsolete by many. One replacement is
the Hosmer - le Cessie test as implemented in the rms package residuals.lrm
function. This is a one degree of freedom test.
One problem with H-L is its use of arbitrary binning and suboptimal power.
The new test requires
The warning is commented out or it would have been printed:
# warning("Some expected counts are less than 5. Use smaller number of
groups")
For 23 data points, the default of 10 bins is too many since one of the bins
is 0. Eight bins gives the warning, but prints results. You didn't indicate
Any method that requires binning is problematic. Instead, take a look at the
calibrate function in the rms package. There is a new option for continuous
calibration curves for survival models.
Frank
jane.wong wrote
>
> Dear list,
>
> Usually we use Hosmer-Lemeshow test to test the goodness of
- On *Tue, 16/9/08, Frank E Harrell Jr /<[EMAIL PROTECTED]>/*
wrote:
From: Frank E Harrell Jr <[EMAIL PROTECTED]>
Subject: Re: [R] Hosmer- Lemeshow test
To: [EMAIL PROTECTED]
Cc: "R list" <[EMAIL PROTECTED]>
Date: Tuesday, 16 September, 2008,
g you in adavance
Â
With regards
Â
Saggak
Â
--- On Tue, 16/9/08, Frank E Harrell Jr <[EMAIL PROTECTED]> wrote:
From: Frank E Harrell Jr <[EMAIL PROTECTED]>
Subject: Re: [R] Hosmer- Lemeshow test
To: [EMAIL PROTECTED]
Cc: "R list" <[EMAIL PROTECTED]>
Date: Tue
5 matches
Mail list logo
