33 / 65
Then for the other people, just assign them a page or 2 to do, and have them build the sheet
from scratch in their favorite software – WITHOUT LOOKING AT THE CODES IN THE EXCAL SHEET
UNDER TEST. It is very critical that they do it from the Appendix. Have them call me or email if
they get stuck. The sheets were:
1.
QUAL SLV – No Ref method comparison
2.
QUAL SLV UNPAIRED
3.
QUAL SLV PAIRED
4.
QUAL COLLAB UNPAIRED
5.
QUAL COLLAB PAIRED
The calculations for confidence intervals are the tricky parts, although we also need to check all
the calcs for mean PODs and dPODs. In addition, people may have difficulty finding the correct
formulae for these confidence intervals, due to the fact that there are several, and also due to
the fact that we have added some since publishing the Appendix. Here’s my best shot at
identifying where the formula are:
1.
QUAL SLV – No Ref method comparison
CI calcs are given in Appx J, Annex C, very first section on the left of the page.
2.
QUAL SLV UNPAIRED
Calcs are Appx J, Annex C, immediately following the above section, starting at lower left, it
gives calcs for dPODs then on the upper right, gives the calcs for the CIs for dPODs. For this
sheet, you have 2 methods, so you calc POD values (and CIs) for both methods as done in sheet
1, then calculate dPOD and CIs on the dPOD as given here in this section of Annex C.
3.
QUAL SLV PAIRED
This one is not in the annex and is discussed in the ppt file we presented at the meeting in
Dallas. The section of Annex C called “dPOD for Paired Studies” will need to be changed to
incorporate the info in the ppt file.
4.
QUAL COLLAB UNPAIRED
This one is in the Annex F – we have modified it slightly though in 2 areas.
We have found in our work in the ISO group that the very complicated Welch-Satterthwaite
estimation for df for the t-based interval is not necessary and the formula df = L-1 works just as
well for the collab sizes we are using. So p. 17 of the Appendix J, Annex F, that df calculation can
be simplified to just df = L-1, or degrees of freedom is number of labs minus one.
The second thing we found was better was to change how we switch from the t-based CI to the
binomial-based CI calculator. Originally, we decided to do this based on the POD value, but in
simulations, it seems to provide a more seamless transition if it transitions not on POD, but on x,
the number of positive hits. For this new version, the Appendix will need to be changed. See
ISO Draft standard or call Paul on this.
Then in addition, you will need to calculate s(R) and s(r) and s(L) for the data. So this will be
Qualitative Micro Validation Proposed Plan