Fitting a GEV Distribution: dataset and results

Question

I'm trying to adjust a GEV Distribution to a dataset, using the "fitdistr" function from "fitdistrplus" package. I'm also having trouble in the interpretation of the p-value result for another dataset. The "chisqpvalue" from "gofstat" function is returning a null vector.

Thank you very much in advance.

I tried the following script:

library(fitdistrplus)
library(evd)

x <- c(11, 6, 3, 3, 4, 4, 5, 9, 4, 16, 8, 9, 7, 8, 16, 11, 5, 8, 9, 4, 17, 6, 7, 7, 6, 6) 
par <- list(loc = 0.0, scale = 1.0, shape = 0)

fitX <- fitdist(abs(x), "gev",start=par)
resultX <- gofstat(fitX)

It's returning the following message: the function mle failed to estimate the parameters, with the error code 100

When i remove some values from the dataset, it runs withour error:

x <- c(11, 6, 3, 3, 5, 9, 4, 16, 8, 9, 7, 8, 16, 11, 5, 8, 9, 4, 17, 6, 7, 7, 6, 6) 
par <- list(loc = 0.0, scale = 1.0, shape = 0)

fitX <- fitdist(abs(x), "gev",start=par)
resultX <- gofstat(fitX)

I think the problem could be the starting parameters that i've choose. But, there is a correct way to estimate it?

Another problem of mine is about the p-value resuits. In the example below (another dataset), the code is running, fitting the GEV distribution, but the p-value(chisqpvalue) result is Null

x <- c(10, 21,  7, 17, 18, 16, 17, 12, 22, 19, 12, 49, 11,  9) 
par <- list(loc = 0.0, scale = 1.0, shape = 0)

fitX <- fitdist(abs(x), "gev",start=par)
resultX <- gofstat(fitX)

Answer 1

Using bbmle::mle2, which is a little more flexible and lets us fit the scale parameter on the log scale so we can stay out of trouble:

library(bbmle)
m1 <- mle2(x ~ dgev(loc, exp(logscale), shape), data = data.frame(x), 
    start = list(loc = 0, logscale = 0, shape = 0), method = "Nelder-Mead")

Using these starting values is good enough to let fitdist succeed:

fitX <- fitdist(x, "gev", 
    start= with(as.list(coef(m1)), list(loc = loc, scale = exp(logscale), shape = shape)))