From @helios.physics.utoronto.ca:LISS@FNALD.FNAL.GOV  Wed Oct  6 11:29:30 1993
Received: from helios.physics.utoronto.ca by cepheid.physics.utoronto.ca with SMTP id AA01535; Wed, 6 Oct 93 11:29:30 -0400
Received: from FNALJ.FNAL.GOV ([131.225.108.4]) by helios.physics.utoronto.ca with SMTP id <1222>; Wed, 6 Oct 1993 11:29:23 -0400
Date: 	Wed, 6 Oct 1993 11:29:02 -0400
From: LISS@FNALD.FNAL.GOV
To: shochet@FNALD.FNAL.GOV, carithers@FNALD.FNAL.GOV, claudioc@FNALD.FNAL.GOV,
        liss@FNALD.FNAL.GOV, gpyeh@FNALD.FNAL.GOV, contreras@FNALD.FNAL.GOV,
        yagil@FNALD.FNAL.GOV, dan@FNALD.FNAL.GOV, bed@FNALD.FNAL.GOV,
        tipton@FNALD.FNAL.GOV, huth@FNALD.FNAL.GOV, buckley@FNALD.FNAL.GOV,
        kuhlmann@FNALD.FNAL.GOV, mlm@FNALD.FNAL.GOV, alvin@FNALD.FNAL.GOV,
        cath@FNALD.FNAL.GOV, UNAL@FNALD.FNAL.GOV, WILLIAMS@FNALD.FNAL.GOV,
        BRANDENBURG@FNALD.FNAL.GOV, pekka@physics.utoronto.ca
Message-Id: <931006102902.23a1543f@FNALD.FNAL.GOV>
Subject: Significance w/ correlations
Status: RO

Hi,
	Last week Brig commented that background correlations between SVX and
SLT due to heavy flavor content need to be included in the significance
calculation.  This is true.  I've now done this, and so has Guillaume .
I think that our methods are virtually identical - but I use slightly different
input values (for reasons I explain below).  Nevertheless, the results are
very close to the same (two differences almost cancel).

	The technique is a monte carlo, one starts from the total number of
Wbbbar, Wccbar, and Wc *before* tagging.  I started from the numbers in
CDF 2252 and scaled up by 2.6, which is [(+DL)-(-DL)]/(Wbbbar+Wccbar from MC).
This is the first difference between Guillaume and me.  Guillaume scales up
by half the difference and quotes an uncertainty of half the difference.
This is fine, but it is not consistent with using "Method 1" for the
background - i.e. if there were no correlation we'd just be adding a background
of 2.0 tags from SVX. For the number of Wc before tagging I start with the
MC Wc number, with no scaling. To get the total number in each category before
tagging I divide by SVX tagging efficiencies provided to me by Guillaume:

	eps(SVX Wbbbar)=0.20		eps(SLT Wbbbar)=0.15
	eps(SVX Wccbar)=0.04		eps(SLT Wccbar)=0.052
	eps(SVX Wc)=0.035		eps(SLT Wc)    =0.020

This gives a total before tagging for Wbbbar of 3.90, Wccbar of 5.75,
Wc of 4.29.  For the uncertainties on these I use 30% from the SVX
tagging efficiency uncertainty.  Guillaume uses 2.63+/-1.25 , 4.29+/-2.03,
4.16+/-1.24 .  His larger uncertainties offset my larger means.  The total
number before tagging is used as a mean to throw a Poisson and events are
counted with the above efficienicies (+/- 30%).  These are added to uncorrelated
throws of the remaining backgrounds (1.16 events for SVX, 2.12 for SLT) with
uncertainties as quoted in the papers. The dileptons are treated as completely 
uncorrelated with the above.

	I find the probability of getting 15 or more tags from the background
alone to be 1.07E-3,  Guillaume gets 1.19E-3 .  I used to have 8E-4 without
the correlations.

					-Tony