1

I have fitted a conditional logistic regression model using the clogit() function from the survival package in R, and I would like to identify the influential observations in my dataset. I have tried several functions, including influence(), influence.measures(), and dfbeta(), but none of them seem to work for clogit objects.

The influence() function returns the error message "lm object does not have a proper 'qr' component. Rank zero or should not have used lm(.., qr=FALSE)." The influence.measures() function returns the error message "Error in UseMethod("influence.measures") : no applicable method for 'influence.measures' applied to an object of class "c('clogit', 'survmod')". The dfbeta() function also does not seem to work.

Can anyone suggest a method for identifying influential observations in a clogit model? I would greatly appreciate any help or advice. Thank you in advance!

H.Dave
  • 37
  • 5
  • 1
    How are you defining "influential observations" in your case? What formula do you want to calculate that would be appropriate for your model? This is probably a better question for [stats.se] since it's more about finding the right method that will work with your model rather than a specific programming question. To make it a programming question you should include a simple [reproducible example](https://stackoverflow.com/questions/5963269/how-to-make-a-great-r-reproducible-example) with sample input and desired output that can be used to test and verify possible solutions. – MrFlick Feb 22 '23 at 15:06
  • Thanks for the comment. I actually want exactly what other influence measures do, ie, how the coefficient estimates change by deleting a certain observation. I think it should be a programming question, because clogit objects provide outputs that do not fit the influence measures above. So either there should be a specific function for clogit objects or a way to transform clogit output to feed into these usual functions. – H.Dave Feb 22 '23 at 15:12
  • I suspect that the error message means you are using functions designed for ordinary linear regression, i.e. for class="lm", on output that has a different link function. When I was learning logistic regression methods we would use a leave-one-out approach to calculating a delta for coefficients. I don't remember seeing that in R packages but you could certainly search on that phrase to see if someone has implemented a "canned" approach. The other approach might be to examine residuals. – IRTFM Feb 26 '23 at 02:09

0 Answers0