[R] Accuracy problem in dchisq for non-central chi-squared

[Peter Dalgaard BSA]

Peter Dalgaard BSA <p.dalgaard at biostat.ku.dk> writes:

> Uffe Høgsbro Thygesen <uht at dfu.min.dk> writes:
> 
> > Any comments on this one?
> 
> Ahem. I went to look at the code and it stared me right in the face:
> 
>     const static int maxiter = 100;
> 
> That needs to be keyed to the actual value of lambda! At lambda=225,
> we're not even at the mode of the weighting Poisson distibution which
> has mean lambda/2.
> 
> I think we can sneak a fix for that one in before the release of 1.2.0
> tomorrow.  (pnchisq.c has 
> 
>     const static int itrmax = 10000;
> 
> which I should probably clone for now, but there must be a better
> long-term solution for both of them. Anyone need a small exercise for
> their students?)

Hmm. I did this and it did help, but the function is still getting in
trouble already for mean=37. I believe this is due to underflow in the
calculation of dchisq used to start the recursion. This'll need some
redesign (including a little theoretical analysis) to function
properly I think.

-- 
   O__  ---- Peter Dalgaard             Blegdamsvej 3  
  c/ /'_ --- Dept. of Biostatistics     2200 Cph. N   
 (*) \(*) -- University of Copenhagen   Denmark      Ph: (+45) 35327918
~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk)             FAX: (+45) 35327907
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._