[R] ncp > 37.62 in pt

[Deepayan Sarkar]

Hi,

?qt says:

     ncp: non-centrality parameter delta; currently for 'pt()' and
          'dt()', only for 'ncp <= 37.62'.

'pt' doesn't complain when I give ncp > 37.62 (pnt.c seems to have a
workaround), but there is a discontinuity:

> cbind(30:40, dt(10, df = 10, ncp = 30:40, log = T))
      [,1]      [,2]
 [1,]   30 -27.74986
 [2,]   31 -30.98710
 [3,]   32 -35.62105
 [4,]   33 -41.85932
 [5,]   34 -49.76884
 [6,]   35 -59.35573
 [7,]   36 -70.60410
 [8,]   37 -83.48987
 [9,]   38 -65.33569
[10,]   39 -69.99854
[11,]   40 -74.82708

In my usage there are many values of 'ncp', a small but positive
fraction of which are greater than 37.62. Any suggestions about how I
should handle it?

Deepayan
--
http://www.stat.wisc.edu/~deepayan/