Hi Concezio, indeed, S and P are anti-correlated which increases the errors. What puzzles me a bit is the fact that the chi^2 is too good. In other words: At large mX the S/P value should fluctuate more than it is observed. If we believe in a constant the systematics on the individual mX bin for the double ratio would (e.g. for IP>0.2) be 0.1491/1.328: 11%. This should result in an interesting systematic error on DeltaBRBR when only taking S/P from MC-enriched and by randomizing S/P by 11% per bin. Of course, one should add still to this error also the error on S/P from MC-enriched. Cheers, Heiko On Thu, 5 Oct 2006, Concezio Bozzi wrote: > Heiko Lacker wrote: > > * Concezio will produce (S/P)-double-ratios for depleted sample > > (data/MC) taking into account S/P-correlations in the error > > calculations in the data-depleted mES fit > > > plots which include the correlations are here: > > http://www.slac.stanford.edu/~bozzi/SoverP/SoverP_mcdata_depl_020.eps > http://www.slac.stanford.edu/~bozzi/SoverP/SoverP_mcdata_depl_030.eps > http://www.slac.stanford.edu/~bozzi/SoverP/SoverP_mcdata_depl_040.eps > http://www.slac.stanford.edu/~bozzi/SoverP/SoverP_mcdata_depl_050.eps > > the spreadsheet used to produce the plots is > http://www.slac.stanford.edu/~bozzi/SoverP/SoverPFullRew.sxc > > Please note that the correlations are not yet taken into account for > intpur>0.5, as I could not find the appropriate logfile. The effect of > the correlations is in any case small. Since the S and P are > ANTIcorrelated (as expected), the error on S/P increase. > > Here are the double S/P ratios (MC/data) for different intpur cuts, as > resulting from a simple p0 fit. Chisquares are also shown. > > intpur cut S/P chisquare/nDOF > 0.20 1.33 +/- 0.15 3.5/9 > 0.30 1.41 +/- 0.16 2.9/9 > 0.40 1.19 +/- 0.17 8.7/9 > 0.50 1.25 +/- 0.17 8.4/9 (no correlations yet) > > I would say that data and MC depleted samples agree better and their > ratio is weakly dependent on mX at higher integrated purity cuts. > Ciao, Concezio. >