R help - Mar 2007 - fitting a mixed exponential distribution

Hi all,

I am attempting to fit, and test the goodness of fit of, a mixed
exponential distribution to my dataset which consists of 15minute
rainfall intensity data. FYI, the dataset spanning approx.2 years and
7 rainfall stations consists of some three hundred thousand 15min data
records, of which some 30 thousand are non-zero rainfall amounts.

Could anyone please tell me how i could do this/which package i could
use in R? I am familiar with MASS functions fitdistr which i've used
for the single exponential distributions, but have not yet
successfully found out how to fit mixed distributions. The single
gamma and exponential distributions did not fit the data well at all,
hence my interest in fitting a mixed exponential distribution.

Also, i thought of formulating the likelihood function and using mle()
in package stats4-- but had trouble writing down the likelihood
function for the case of a mixed exponential distribution...

thanks,
vishal

Vishal Mehta
PhD Candidate
Crop and Soil Science
Cornell University

Vishal Mehta <vishalm1975 <at> gmail.com> writes:
> Also, i thought of formulating the likelihood function and using mle()
> in package stats4-- but had trouble writing down the likelihood
> function for the case of a mixed exponential distribution...

  I don't know offhand if there are CRAN packages etc. that
will do this -- but the canonical example is the model in MASS
for Old Faithful eruptions, which is a mixture model coded by
hand -- I believe it's a normal mixture model but it should
be a good starting point for rolling your own.

   The mixture model log-likelihood would look something like this:

mixexplik <- function(p,lambda1,lambda2) {
  log(p*dexp(x,lambda1) + (1-p)*dexp(x,lambda2))
}

  good luck,
    Ben Bolker