[R] Fit continuous distribution to truncated empirical values

[Michele Mazzucco]

Hello David,

thanks for your answer.
I have done as you told me, however the fit is very poor, much worse than that obtained from using the whole dataset (without upper bound).
Any idea?

Thanks,
Michele

On Nov 4, 2011, at 8:56 PM, David Winsemius wrote:

> 
> On Nov 3, 2011, at 7:54 AM, Michele Mazzucco wrote:
> 
>> Hi all,
>> 
>> I am trying to fit a distribution to some data about survival times.
>> I am interested only in a specific interval, e.g., while the data lies in the interval (0,...., 600), I want the best for the interval (0,..., 24).
>> 
>> I have tried both fitdistr (MASS package) and fitdist (from the fitdistrplus package), but I could not get them working, e.g.
>> 
>> fitdistr(left, "weibull", upper=24)
>> Error in optim(x = c(529L, 528L, 527L, 526L, 525L, 524L, 523L, 522L, 521L,  :
>> L-BFGS-B needs finite values of 'fn'
>> In addition: Warning message:
>> In dweibull(x, shape, scale, log) : NaNs produced
>> 
>> Am I doing something wrong?
> 
> You didn't supply data to test,  but shouldn't you supply a lower bound if you want to fit "weibull"? It is, after all, bounded at 0.
> 
> > left <- c(529L, 528L, 527L, 526L, 525L, 524L, 523L, 522L, 521L, 50*runif(100))
> > fitdistr(left, "weibull", upper=24)
> Error in optim(x = c(529, 528, 527, 526, 525, 524, 523, 522, 521, 18.3964251773432,  :
>  L-BFGS-B needs finite values of 'fn'
> In addition: Warning message:
> In dweibull(x, shape, scale, log) : NaNs produced
> 
> > fitdistr(left, "weibull", upper=24, lower=0.5)
>      shape         scale
>   0.58195013   24.00000000
> ( 0.04046087) ( 3.38621367)
> 
>> 
>> 
>> Thanks,
>> Michele
>> 
>> 
>> p.s. I have seen similar posts, e.g., http://tolstoy.newcastle.edu.au/R/help/05/02/11558.html, but I am not sure whether I can apply the same approach here.
>> ______________________________________________
>> R-help at 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.
> 
> David Winsemius, MD
> Heritage Laboratories
> West Hartford, CT
>