Dear all, I have been trying to fit my data (only right censored) with gumbel distribution using fitdistrplus. I am getting very high standard error. I have been wondering why. The followings are the outputs: fit1=fitdistcens(dr0, "gumbel", start=list(a=99, b=0.6), optim.method= "L-BFGS-B", lower = 0.0, upper = Inf)> summary(fit1)FITTING OF THE DISTRIBUTION ' gumbel ' BY MAXIMUM LIKELIHOOD ON CENSORED DATA PARAMETERS estimate Std. Error a 97.8260371 3115.09 b 0.1115094 173.79 Loglikelihood: -9.749883e-10 AIC: 4 BIC: 21.21178 Correlation matrix: a b a 1.0000000 -0.6693773 b -0.6693773 1.0000000 [[alternative HTML version deleted]]
Ben Bolker
2013-Feb-12 16:13 UTC
[R] standard error very high in maximum liklihood fitting
Abu Naser <likhonnaser <at> hotmail.com> writes:> I have been trying to fit my data (only right censored) > with gumbel distribution using fitdistrplus. I am > getting very high standard error. I have been wondering why. > The followings are the outputs: > > fit1=fitdistcens(dr0, "gumbel", start=list(a=99, b=0.6), > optim.method= "L-BFGS-B", lower = 0.0, > upper = Inf) > > summary(fit1)[snip]> estimate Std. Error > a 97.8260371 3115.09 > b 0.1115094 173.79 > Loglikelihood: -9.749883e-10 AIC: 4 BIC: 21.21178[snip] Impossible to say without a reproducible example. It seems very suspicious that your log-likelihood is almost exactly zero. Do you have a very small data set?
Abu Naser
2013-Feb-14 01:51 UTC
[R] FW: standard error very high in maximum liklihood fitting
I thought it was going to the r-list.> Date: Wed, 13 Feb 2013 11:33:13 -0500 > From: bbolker@gmail.com > To: likhonnaser@hotmail.com > Subject: Re: [R] standard error very high in maximum liklihood fitting > > I can't respond to e-mails off-list. Please follow up on r-help. > > cheers > Ben Bolker > > On 13-02-13 09:46 AM, Abu Naser wrote: > > Thank you Ben for responding my email. I tried with the small data set. > > It was the same: > > > > fit1=fitdistcens(dr8, "gumbel", start=list(a=170, b=1), optim.method> > "L-BFGS-B", lower = 0.0, upper = Inf) > > > > estimate Std. Error > > a 169.9770615 43631.070 > > b 0.4807869 2053.354 > > Loglikelihood: -1.340523e-09 AIC: 4 BIC: 9.08659 > > Correlation matrix: > > a b > > a 1.0000000 -0.9574057 > > b -0.9574057 1.0000000 > > > > I think censoring causing the problem. The following was my data: > > > > left right > > 1 NA 200 > > 2 NA 184 > > 3 NA 185 > > 4 NA 212 > > 5 NA 181 > > 6 NA 188 > > 7 NA 184 > > 8 NA 195 > > 9 NA 181 > > 10 NA 183 > > 11 NA 191 > > 12 NA 214 > > 13 NA 183 > > 14 NA 193 > > 15 NA 182 > > 16 NA 194 > > 17 NA 188 > > 18 NA 202 > > 19 NA 186 > > 20 NA 185 > > 21 NA 215 > > 22 NA 183 > > 23 NA 184 > > 24 NA 184 > > 25 NA 186 > > 26 NA 183 > > 27 NA 196 > > 28 NA 195 > > 29 NA 197 > > 30 NA 181 > > 31 NA 193 > > 32 NA 183 > > 33 NA 187 > > 34 NA 192 > > 35 NA 186 > > 36 NA 186 > > 37 NA 187 > > 38 NA 182 > > 39 NA 203 > > 40 NA 182 > > 41 NA 181 > > 42 NA 183 > > 43 NA 189 > > 44 NA 185 > > 45 NA 183 > > 46 NA 193 > > 47 NA 186 > > 48 NA 182 > > 49 NA 198 > > 50 NA 189 > > 51 NA 187 > > 52 NA 190 > > 53 NA 181 > > 54 NA 192 > > 55 NA 185 > > 56 NA 190 > > 57 NA 196 > > 58 NA 181 > > 59 NA 186 > > 60 NA 182 > > 61 NA 213 > > 62 NA 181 > > 63 NA 184 > > 64 NA 189 > > 65 NA 187 > > 66 NA 182 > > 67 NA 223 > > 68 NA 184 > > 69 NA 185 > > 70 NA 185 > > 71 NA 222 > > 72 NA 188 > > 73 NA 187 > > 74 NA 196 > > 75 NA 194 > > 76 NA 181 > > 77 NA 193 > > 78 NA 185 > > 79 NA 181 > > 80 NA 181 > > 81 NA 182 > > 82 NA 184 > > 83 NA 193 > > 84 NA 209 > > 85 NA 181 > > 86 NA 183 > > 87 NA 184 > > 88 NA 186 > > 89 NA 196 > > 90 NA 198 > > 91 NA 182 > > 92 NA 202 > > 93 NA 197 > > 94 NA 217 > > > > > > > > > >> To: r-help@stat.math.ethz.ch > >> From: bbolker@gmail.com > >> Date: Tue, 12 Feb 2013 16:13:27 +0000 > >> Subject: Re: [R] standard error very high in maximum liklihood fitting > >> > >> Abu Naser <likhonnaser <at> hotmail.com> writes: > >> > >> > I have been trying to fit my data (only right censored) > >> > with gumbel distribution using fitdistrplus. I am > >> > getting very high standard error. I have been wondering why. > >> > The followings are the outputs: > >> > > >> > fit1=fitdistcens(dr0, "gumbel", start=list(a=99, b=0.6), > >> > optim.method= "L-BFGS-B", lower = 0.0, > >> > upper = Inf) > >> > > summary(fit1) > >> > >> [snip] > >> > >> > estimate Std. Error > >> > a 97.8260371 3115.09 > >> > b 0.1115094 173.79 > >> > Loglikelihood: -9.749883e-10 AIC: 4 BIC: 21.21178 > >> > >> [snip] > >> > >> Impossible to say without a reproducible example. It seems > >> very suspicious that your log-likelihood is almost exactly zero. > >> Do you have a very small data set? > >> > >> ______________________________________________ > >> R-help@r-project.org mailing list > >> https://stat.ethz.ch/mailman/listinfo/r-help > >> PLEASE do read the posting guide > > http://www.R-project.org/posting-guide.html > >> and provide commented, minimal, self-contained, reproducible code. >[[alternative HTML version deleted]]