Roger Hui wrote:
>> John's version relies on ! which is limited to !170.  When the mean
>> exceeds
>> 170, as Mr. Hui suggested,  perhaps the function should simply return a
>> call
>> to the normal function.
>
> Alternatively, you could recode tpdf to use equation 5 in
> http://mathworld.wolfram.com/Studentst-Distribution.html
> and compute the ratio  (gamma 0.5*1+x)%gamma 0.5*x
> not directly but by  */vec0%vec1 .
>
> Separately, another possibility is to use the exact method
> up to its limit and apply integration on the remaining tail.

Some of this depends on what accuracy you are shooting for and over what
domain.  SPSS regards 3 decimal places as adequate for P-values; as a rule
of thumb, you can use the normal distribution if the number of degrees of
freedom is over 30; and you never need y large (apart from nu=1, all the
critical values for alpha=0.005 are less than 10).  If the aim is to
produce a function accurate over a larger domain, a more patchwork
approach may be needed.

Best wishes,

John


----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm

Reply via email to