Skip to content

pR2 stumped

2 messages · ivo welch, Achim Zeileis

Original
ivo welch
# 
Dear R experts:  I am stumped.    I am trying to pick off the mcfadden
R^2 for a probit.  simple, me thinks---except my code works only in my
main program, but not in my sub!?    I am probably doing something
obviously wrong, but I have stared at my code for a while now and I
feel even more stupid than usually.  am I doing something obviously
wrong here?  advice, as always, appreciated.  /iaw


library(MASS)
library(pscl)

summarize <- function(p) {
  pp <- polr( factor(r1) ~ factor(r2), Hess=TRUE, data=p, method="probit" )
  print(pp)
  cat("---------------- (sub) now we do the pseudo R^2\n")
  print(pR2(pp))  ## it fails in the summarize sub program, wanting 'p'  huh?
  cat("---------------- ok\n")
}

################################################################
d <- data.frame(r1=factor( c("a", "a", "b", "b", "c", "c") ),
r2=factor( c("a", "b", "c", "b", "c", "a") ))
################################################################

### it works in the main program
pp <- (polr( factor(r1) ~ factor(r2), Hess=TRUE, data=d, method="probit" ))
print(pp)
cat("---------------- (main) now we do the pseudo R^2\n")
print(pR2(pp))
cat("---------------- ok\n")
summarize(d)

 Error in eval(expr, envir, enclos) : object 'p' not found.

----
Ivo Welch (ivo.welch at gmail.com)
Achim Zeileis
#
On Mon, 6 May 2013, ivo welch wrote:

            

      
      
      
    

        
pscl:::pR2.polr calls update(object, ~ 1) to obtain the fitted 
log-likelihood of the null model (intercept only). If the calls are nested 
deeply enough, re-evaluating the model call does not work because it 
happens in the wrong environment.

A workaround could be to do the updating within your summarize() function 
directly and then call pscl:::pR2Work(llh, llhNull, n) directly. Or if you 
really just need the McFadden pseudo R-squared, it is then probably 
simplest to compute 1 - llh/llhNull yourself...

Best,
Z