Interpreting "validate.lrm"
Matías Guzmán
Hello everyone,
I have a very basic question. I have a logistic regression with lrm but I'm having trouble interpreting the bootstrap validation. I'm not sure if it's overfitted or not, and I also don't really understand what each statistic mean (I have to report them in a paper). What I get from the anova analysis is:
Wald Statistics Response: Type
Factor Chi-Square d.f. P
Context (Factor+Higher Order Factors) 17.30 2 0.0002
All Interactions 6.43 1 0.0112
Pronominal (Factor+Higher Order Factors) 14.88 2 0.0006
All Interactions 6.43 1 0.0112
VerbTyp 11.94 3 0.0076
Context * Pronominal (Factor+Higher Order Factors) 6.43 1 0.0112
TOTAL 23.90 6 0.0005
And from the bootstrap validation, B=200:
index.orig training test optimism index.corrected n
Dxy 0.6259 0.6543 0.5913 0.0630 0.5630 200
R2 0.3842 0.4324 0.3462 0.0862 0.2980 200
Intercept 0.0000 0.0000 0.0132 -0.0132 0.0132 200
Slope 1.0000 1.0000 0.8049 0.1951 0.8049 200
Emax 0.0000 0.0000 0.0506 0.0506 0.0506 200
D 0.3310 0.3873 0.2921 0.0953 0.2357 200
U -0.0177 -0.0177 0.0271 -0.0448 0.0271 200
Q 0.3487 0.4050 0.2649 0.1401 0.2086 200
B 0.1747 0.1633 0.1858 -0.0225 0.1972 200
g 1.6948 2.0311 1.5322 0.4989 1.1959 200
gp 0.3146 0.3273 0.2942 0.0331 0.2814 200
From what I can understand the model is slightly overfitted, but I don't really understand what each statistic mean or how to interpret the values for optimism.
Thanks
Stefan Th. Gries
index.orig= values from the non-bootstrapped regression
training= average values from bootstrapped samples
test= average values when bootstrapped models are applied to full data
optimism= training-test
index.corrected=index.orig-optimism
n: number of bootstrap samples
STG
--
Stefan Th. Gries
-----------------------------------------------
University of California, Santa Barbara
http://www.linguistics.ucsb.edu/faculty/stgries
-----------------------------------------------