Re: Calculation of mag errors

[Doug Welch <welch@physics.mcmaster.ca>]

I guess these underestimates of the photometric errors help explain why
the W-S statistic wasn't performing quite as well as advertised!

Cheers,
Doug

On Sat, 2005-03-12 at 16:19, Andrew Bennett wrote:
> On Mon, 7 Mar 2005 18:49:12 -0500, Stupendous Man
> <richmond@stupendous.cis.rit.edu> wrote:
> 
> >
> >  As a first step to fixing the error in calculating uncertainty
> >in instrumental magnitude, I have added a new option to the
> >"phot" program within XVista.  This is the program which takes
> >a list of stellar position and a corrected image, extracts
> >instrumental magnitudes via aperture photometry, and prints out
> >the results.  
> >
> >  The old version calculated uncertainty by separately determining
> >the variance in the sky background, the readnoise, and the light
> >from the star itself, then adding them all up.  That method
> >underestimates the error if a mean sky level has been subtracted
> >from the image.
> >
> >  The new version takes the approach Mike G. suggested: it calculates
> >the scatter in pixel values within an annulus around each star,
> >and uses _this_ to estimate the variance in the background
> >(a combination of sky and readnoise and dark current).  
> >It adds in the variance due to the star, then compares to 
> >the signal from the star to estimate an uncertainty.
> 
> Great! I was always too lazy to attempt to program this ...
> 
> I noticed when I went through your code that you don't
> add a term for the uncertainty in estimating the
> background level that gets subtracted from each
> pixel. Years ago, I  went through your code, which was
> then quite different from the current version, and, as I
> remember it, it did add a term for this. I distinctly
> remember taking exception to the way you
> calculated the variance of the median. I also
> distinctly remember not posting to the list about it on the
> grounds that the difference was too small to make
> a fuss about.
> 
> Now that you have done all the dirty work of
> setting up the statistics for the annulus, is this
> a good time to revisit estimating this error?
> One might be able to estimate things like
> background curvature - if too big, go to a
> smaller annulus and accept the higher random
> error in exchange for the smaller systematic
> error. I never tried to do this: I couldn't see how
> to do it in a manner that was not horribly
> sensitive to nearby confusing stars.
> 
> Andrew Bennett, Avondale Vineyard, NS, Canada
-- 
==================================================================
 Douglas L Welch     | Res office/voicemail (905) 525-9140 x23186
 Physics & Astronomy | FAX                  (905) 546-1252
 McMaster University |
 Hamilton, Ontario   |
 Canada L8S 4M1      | E-mail           welch@physics.mcmaster.ca
==================================================================