On Sep 9, 8:26 pm, djh <halitsk...@att.net> wrote: > I've programmed and desk-checked the custom heteroscedastic test in > Perl. > > By checking each step of the calculation against the corresponding > step in Excel 2010, I have verified that this PERL version is working > correctly. > > However: > > 1) the PERL "dist" function only accepts integer "ft's" (degrees of > freedom); > > 2) it reports a one-tailed p out to 6 decimal places. > > So, for: > > t = 1.75392432472924 > ft = 1545.45548783244 > > I get the following p (after multiplying the PERL one-tailed p by 2): > > p (two-tailed) = 0.079642 > > which matches the Excel 2010 two-tailed p as far as it goes out. > > (I get this p whether I submit with 1545 or 1546 ... I do both in the > code just to check for a difference, though there won't often be one, > I don't think.)
Use the lower value (e.g., 1545, not 1546).
A related question: Does any of your software have an "Incomplete Beta" function,
were 0 <= x <= 1 and a,b > 0? Or (even better) a "regularized" or "normalized" incomplete beta, that gives I[x,a,b]/B[a,b], where B[a,b] = I[1,a,b] is the usual Beta function? If so then the t and F probs can be obtained from I[x,a,b].