Meeting to discuss BSA systematic uncertainties calculation
                               24-Apr-02
                             (Notes by GH)

Present
-------
Regina - Garth Huber, Nacer Hamdi, Ali Usman, Alicia Postuma, Muhammad Junaid,
   Nathan Heinrich, Vijay Kumar
JLab - Dave Gaskell
Ohio - Julie Roche
CUA - Tanja Horn
York - Stephen Kay

Alicia has prepared some slides to guide our discussion.

At present, Alicia calculates the BSA 10 different ways:
                    Full-Fit   Sine-Fit
     Nominal Cuts      A          A'
     Narrow CT         CN         CN'
     Wide CT           CW         CW'
     Narrow MM         MN         MN'
     Wide MM           MW         MW'

For each one, obtain a fitting uncertainty for Sin(phi) term, yielding 10
different values of the BSA and 10 corresponding uncertainties.

The question is how should we calculate A +/-stat +/-syst ?
- right now, taking the error-weighted average of A and A' as the final BSA
  result

Q1: at what stage do we take the weighted average of the Full and Sine fits?

Option 1: calculate error weighted average for every set of cuts, relative to
  the (A and A' average), and use this to calculate delta(CT), delta(MM)

Option 2: average the cut dependences first, not relative to (A and A' average)

Dave: likes Option1 (weighted averages first).  Want to find the systematic
  relative to the final results, not the intermediate results

Julie: you should take the full-fit A as the answer, and use the (A-A')
  difference from sine-fit as a systematic
  - Tanja prefers this too

Garth: since each of the cut study uncertainties include the statistical
  uncertainties resulting in the respective fits, we will likely have to remove
  the raw statistical uncertainty in quadrature from the systematic coming from
  the fits
  - stats do not enter in (A-A'), but it does in the others

Plotting a systematic error band, or double error bars (as Hall B does)?

Dave: prefers double error bars
  - a systematic error band implies that all points move togther, while they
    probably move independently

Plot points at mean -t or central -t?
  - agree that mean -t is better, since the BSA results are not bin centered

Q2: How to calculate error due to cut dependence?
  delta(C)= [ |A-CN|+|A-CW| ]/2
  delta(M)= [ |A-MN|+|A-MW| ]/2
  - Dave: this is how he does it
  - Nacer: many experiments give the RMS of 2,3 variations
  - Garth: likes delta(C), delta(M) separately for errors discussion, but we
    need to give the community an easily understandable way to use our BSA
    data.  For this, suggest to take RMS of (A-A', CN, ..., MW results) as the
    systematic.

Beam Polarization Error:
- all points should move together
- since this is not the dominant uncertainty (smaller than above systematic),
  just add it in quadrature with the others

Good Consensus:
- show A as result
- include RMS of all different variations as a systematic
- show as a double error bar