fitting an inverse gaussian distribution to data in R

Question

Im trying to use the fitdist function in R to fit data to three different distributions by maximum likelihood to compare them. Lognormal and Weibull work fine, but I am struggling with Inverse Gaussian. I need to specify starting values, however when I do I get an error message.

fw<-fitdist(claims,"weibull") WORKS
fln<-fitdist(claims,"lnorm") WORKS
fig<-fitdist(claims,"invgauss",start=list(mu=0,lambda=1)) DOES NOT WORK

Error: 'The pinvgauss function should return a zero-length vector when input has length zero and not raise an error'

What is wrong with my code?

Have you tried other starting values? I don't believe mu=0 makes sense. — Dason, Oct 01 '18 at 05:43
Both parameters must be greater than zero in invgauss. Wiki link: https://en.wikipedia.org/wiki/Inverse_Gaussian_distribution — JacobJacox, Oct 01 '18 at 06:21
I see. But even when I put higher values than 0, I get the same error. — sam, Oct 02 '18 at 23:38

score 3 · Answer 1 · answered May 10 '19 at 17:10

I was working with a similar issue and found the issue was with how I labeled my start values. The actuar library I was working with required the labels "mean" and "shape" on the values. The following code provided me a solution:

library(actuar)
library(fitdistrplus)
fig <- fitdist(claims, "invgauss", start = list(mean = 5, shape = 1))

fitting an inverse gaussian distribution to data in R

1 Answers1