Summary of dicussion during chop-nod analysis telecon of Feb 6, 2008

Hiroko, Larry, Lero, Mike, John, Giles

_________________________________________________________________

M82 analysis - April 2007 runs:

file 36784: Larry explained that the error in I is calculated from the variance in the four measured values
of I (=H+V) for the four different hwp angles.  This means that the error in I is itself subject to error,
since it is being estimated from only four points.  We just got unlucky in that the four measurements were 
by statistical fluke very close.

Solutions are : (a) watch out for this and cut files when it happens (b) fix sharpinteg so that anomalously 
low sigma-I measurements are inflated.  Two ways to do this are set all sig-I values that are lower than the 
median sig-I value equal to the median, and set all sig-I values that are less than the median by more than 
three sigma equal to the median.  Larry will implement both (with a flag to choose which or neither) 
sometime in the next week or two.

Lero is trying to learn from Mike how to correct the pointing for M82.  In the process, they discovered that
when Mike and Lero run sharpinteg on the same file, they get different answers for what the EL is.  Lero
gets one answer or zamin, and Mike gets a different answer on his mac.  The next thing is to make sure they
are using the same flags.  Larry explained that the way he gets EL out of the raw file is to take some kind
of a median of the time-stream values of EL in the file.  Its hard to see how Lero and Mike could get
different answers.  This bears further investigation.  (The discrepancy in El is less than a couple of
degrees.)

(Mike uses fv to examine the header - this should work.)

_________________________________________________________________

Reduced-chi-squared (rcs) procedures:

Mike reported results from rcs of Feb 2007 M82 data.

mean rcs of Q and U are 2.3 and 1.7 provided that you discard three very high values for the U case.  (bad 
data ?)

But the spread of rcs values is high (the std is of order 10 for Q).  To see if this is consistent with
random statistics (remember the rcs is determined from just 4 bins = 3 degrees of freedom) one needs to plot
up the distributions and compare with the formula in Bevington.  This is the next step.

The rcs of I is low (~0.01).  This means that the sig-I are overestimates by a factor of ten.  I think this
may be expected.  When I looked at the DG Tau data I found that the I maps looked like a point source but
with a random DC offset superposed (a sky noise effect).  From this I would guess that the four I maps that
are used to get the sig-I have similar random DC offsets.  These will affect the sig-I adversely.  But then
John does background subtraction which to a large extent removes this problem.  So the errors are
overestimates or the true uncertainty in the final I-map produced by sharp-integ.

Sometimes the rcs is negative.  This is rare and is an artifact of the way that the error in Q is
reconstructed from the error in q and error in I from sharpcombine.  We discussed this reconstruction
process and its pitfalls in December.  Mike pointed out that in the individual error-estimates for the four
bins this also happens, and if you set these to NaN then a large part of your rcs map comes out to be NaN.

One important next step is to track down the few points that are giving very high rcs in the U map.

_________________________________________________________________
The following items were not discussed on the telecon:

Mike's plans to make a smoothed tau for November 2007 

listserve

alternate ideas for telecon - iChat (video?) on Mac

postscript problems on Lero's Mac and on kilauea, and Hiroko's
investigation of the kilauea problems.  In any case, I guess these are
solved.

_____________________________________________________________