HypoTestInverter and 0 computed error

Roofit and RooStats
1

Hi,
I am running cls limits with HypoTestInverter in 5.32.02. Often i get something like
computed upper limit is: 1.16974 +/- 0

a. is this result still valid in spite of the identically = 0 error? the
expected errors seen reasonable.

b. can one explain why this (identically = 0) occurs?

c. i’ve tried varying npoints and ntoys which sometimes results
in a more sensible error (or sometimes the dreaded ERROR, some proxies could not
be adjusted error), but no pairings that are consistent. is there advice how
to avoid this?

thanks…kj

2

Hi,

If you get the zero error is because you are using the asymptotic approximation. This error in the limit is obtained by propagating the statistics uncertainty in the p-value due to the limited number of toys (pseudo-experiments). In case of the asymptotic calculators, no toys are used, therefore there you have an error = 0.

In reality an error could be computed from the fact that the points are interpolated. I am planning to update the class to compute this.

Thank you for your post

Lorenzo