1
$\begingroup$

Recently I was trying to do logistic regression using the rms::lrm() function. But I had some trouble understanding the model objects from the function. Here is the example from the package:

#dataset
n            <- 1000    # define sample size
set.seed(17)            # so can reproduce the results
treat        <- factor(sample(c('a','b','c'), n,TRUE))
num.diseases <- sample(0:4, n,TRUE)
age          <- rnorm(n, 50, 10)
cholesterol  <- rnorm(n, 200, 25)
weight       <- rnorm(n, 150, 20)
sex          <- factor(sample(c('female','male'), n,TRUE))
L            <- .1*(num.diseases-2) + .045*(age-50) +
                (log(cholesterol - 10)-5.2)*(-2*(treat=='a') +
                3.5*(treat=='b')+2*(treat=='c'))
y            <- ifelse(runif(n) < plogis(L), 1, 0)
#fit model
g            <- lrm(y ~ treat*rcs(age))

> g

Logistic Regression Model

lrm(formula = y ~ treat * rcs(age))

                      Model Likelihood     Discrimination    Rank Discrim.    
                         Ratio Test            Indexes          Indexes       
Obs          1000    LR chi2      76.77    R2       0.099    C       0.656    
 0            478    d.f.            14    g        0.665    Dxy     0.312    
 1            522    Pr(> chi2) <0.0001    gr       1.945    gamma   0.314    
max |deriv| 3e-06                          gp       0.156    tau-a   0.156    
                                           Brier    0.231    
> anova(g)
                Wald Statistics          Response: y 

 Factor                                     Chi-Square d.f. P     
 treat  (Factor+Higher Order Factors)        5.62      10   0.8462
  All Interactions                           1.30       8   0.9956
 age  (Factor+Higher Order Factors)         65.99      12   <.0001
  All Interactions                           1.30       8   0.9956
  Nonlinear (Factor+Higher Order Factors)    2.23       9   0.9872
 treat * age  (Factor+Higher Order Factors)  1.30       8   0.9956
  Nonlinear                                  0.99       6   0.9858
  Nonlinear Interaction : f(A,B) vs. AB      0.99       6   0.9858
 TOTAL NONLINEAR                             2.23       9   0.9872
 TOTAL NONLINEAR + INTERACTION               2.57      11   0.9953
 TOTAL                                      69.06      14   <.0001

Here are my questions:
For the object g,

  • What does the max |deriv| 3e-06 mean?
  • What do the Discrimination and Rand Discrim. Indexes suggest?

For the anova(g) object,

  • What's the Factor +Higher Order Factors for the treat?
  • Why there are two all interactions? How to explain the nonlinear parts?
$\endgroup$
3
$\begingroup$

Max |deriv| is the maximum (over $\beta$s) of the absolute value of the first derivative of the log-likelihood function at the apparent maximum likelihood estimates. Being close to zero is a good indication that convergence happened. There are a lot of background sources explaining the various indexes, especially my course notes at http://biostat.mc.vanderbilt.edu/CourseBios330. To interpret anova, type ?print.anova.rms to see the various printing options for anova objects. These options include printing subscripts of $\beta$s that are tested in a given row, or printing a matrix of asterisks to the right of the statistics with blank spaces for $\beta$s not included in the current test.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.