Re: [R] three par. fitting with fitdistr

From: Martin Maechler <maechler_at_stat.math.ethz.ch>
Date: Wed 13 Jul 2005 - 02:13:33 EST

>>>>> "Carsten" == Carsten Steinhoff <carsten.steinhoff@stud.uni-goettingen.de>
>>>>> on Tue, 12 Jul 2005 17:49:34 +0200 writes:

Carsten> Hello,

```    Carsten> I want to fit a tree parameter distribution to
Carsten> given data. I tried it with sample data using the
Carsten> "fitdistr" function.

```

Carsten> Here my workflow that didn't had any result:

Carsten> I started with the generalized gamma distr, which is:     Carsten> r*dgamma(x^r,shape,rate)

Carsten> The R-function is:

Carsten> ggamma = function (x,r,shape,rate) r*dgamma(x^r,shape,rate=rate)

Carsten> For the first step I assumed r = 1 and I generated     Carsten> random numbers with the "standard" Gamma distr.

Carsten> rn=rgamma(1000,10,5)

Carsten> In the last step I want to reconstruct the parameters from the dataset:

Carsten> library(MASS)
Carsten> fitdistr(rn, ggamma, list(r=1,shape=10,rate=5))

```    Carsten> But there is an error: Error in fitdistr(rn, ggamma, list(r = 1, shape = 10,
Carsten> rate = 5)) :  optimization failed
Carsten> although I should have a nearly model-made dataset.

```

Carsten> Where is the problem and how could it be solved?

ggamma() is not a density.
You need to provide the correct multiplication factor (a function of 'r' at least) in the definition of ggamma() such that it *does* integrate to 1.

Martin

Carsten> Carsten

```    Carsten> [[alternative HTML version deleted]]
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

```

please read the posting guide and learn how to not produce     the above !

R-help@stat.math.ethz.ch mailing list