[R] Compare two Power law or Exponential distributions

[Ben Bolker]

Emmanuel Levy <emmanuel.levy <at> gmail.com> writes:

> 
> Dear All,
> 
> I'd like to compare exponential or power-law distributions.
> 
> To do so, people are often referred to the ks.test. However,
> I imagine ks.test wouldn't be as powerful as a test specifically
> designed for a distribution type.
> 
> So my question is, is there a more specific test for each of
> these distribution? (exponential or power-law)
> 

  assuming that you've got a data set with two groups, estimate
the parameters and find the maximum log-likelihood (minimum
negative log-likelihood) for (1) pooled data and (2) each group
separately.  Add the negative log-likelihoods for the grouped
estimates.  Use the likelihood ratio test to decide whether
the reduced model (all parameters equal) is significantly worse
than the full model (all parameters different).



   Ben Bolker