Summary of chop-nod analysis telecon of Nov 26, 2008

Mike, Giles, Lero, Tristan, Hiroko, John

_____________________________________________________________

Mike's analysis of NGC 1333 IRAS 4:

Mike went through his recent posting (of earlier this week)

He is inflating errors using point-by-point reduced-chi-squared (rcs), 
not map-average rcs or cropped-region rcs.  (update flag accomplishes 
this)

The point-by-point method has the advantage of not having to specify 
cropped regions, which are to some extent arbitrary, but it has the 
disadvantage that one cannot use small bin numbers.  So Mike uses 14 
bins.

As a way to explore whether this misses long-time-scale errors, Mike did 
some experimentation with 7 bins: The map-average Q and U rcs show only 
modest increases (< a factor of 1.1) when moving from 14 to 7 bins.  
This is very encouraging.

The P-inflation-factor calculated and outputted by Mikes chi2 program is 
not currently being used by Mike for his 1333 data analysis.  Instead he 
calculates inflation factors directly from the rcs values, as we have 
done in the past.

Giles is worried that the P-inflation-factors seem too low to be 
correct. Note that the math is documented in a recent E-mail from Mike 
to the list-serve.

(After the telecon I realized the discrepancy may be related to the 
non-Gaussian nature of errors in P.  In any case, for now its best not 
to use these P-inflation-factors.  Just continue to inflate by the 
square-root of the rcs.)

_____________________________________________________________

New data analysis methods proposed Oct. 20:

NOTE: As a reminder, Giles' original suggestion is posted to the 
chop-nod e-mail archive.  Selecting "thread" is the easiest way to see 
Giles' e-mail, John's reply, and Giles' reply to that.  You log in with 
your e-mail address and the password is the 3-letter acronym for our 
favorite observatory (all lower case).  The URL is given in every 
sharp-software list-serve e-mail.

We discussed an alternative way to implement the super-bin/sub-bin idea.  
This was suggested by John in his reply to Giles' original e-mail.  The 
idea is to have sharpcombine calculate the rcs for every map point and 
inflate the errors accordingly.  This would allow full implementation of 
the super-bin/sub-bin idea: We would use sharpcombine to combine 
sub-bins (in this case individual files) into super-bins and then chi2 
to compare super-bins.

This sounds simple but in practice it may require modifications to 
several different subroutines in sharpcombine.  Topics for future 
discussion: Which of the two ways is easier? which is better? who has 
time to modify sharpcombine in the near future?

Note: Discussion between Tristan and John leads to the suggestion that 
we could implement Q and U background subtraction in sharpcombine in 
parallel with the super-bin/sub-bin changes, as they are independent 
improvements.

_____________________________________________________________