Ashton, Gail <ashtong <at> si.edu> writes:
>
> I have 3 variables relating to the successful introductions of species
> to 95 different areas: introduction frequency; number of successes pre
> 1906; number of successes post 1906
>
> The data are not normal, nor homo-skedatic, so I am using non-parametric
> statistics.
>
> I have calculated Kendall's tau between both introduction &
successes
> pre 1906 (tau=0.3903) and introduction & successes post 1906
> (tau=0.3317)- pre + post values are independent.
>
> Is it possible to test whether there is a significant difference between
> the tau values, ie one correlation is 'significantly'
'better'? Is there
> a way to calculate confidence intervals for tau?
>
> Thanks,
>
> Gail
I think I would try bootstrapping/permutation tests in this case.
Just a warning: nonparametric tests do _not_ necessarily do what
you think in the case of heteroscedasticity -- they are often
constructed to test for a difference in location parameter, assuming
the same distribution around the location parameter (mean, median etc.) --
i.e. assuming homoscedasticity.
Any chance of using GLMs (i.e., treating success as a binary
variable)?
Ben Bolker