Date: Thu, 16 Mar 2006 14:02:13 -0500
From: arne <arne
To: Michael Koppelman <lolife
References: <000c01c648a4$07a2bd20$6401a8c0

Michael Koppelman wrote:
> The problem is we end up with people doing more thorough error  analysis 
> getting uncertainties an order of magnitude greater than  those doing 
> 1/SNR with no way to tell the difference. This makes it  impossible to 
> meaningfully do weighted averages or weighted fits. I  always put in the 
> comments that my uncertainties include comp scatter  but I'm not sure 
> that info gets to researchers who download data  (yet). If we ask people 
> like me to submit 1/SNR uncertainties, we are  essentially asking them 
> to throw away part of their data.
> 
> If we are making any effort to get people to calculate uncertainties,  
> it makes sense to me that the effort should guide people to the most  
> useful technique and not the easiest, especially when the most useful  
> technique is pretty darn easy.
> 
> If we could flag which method we used, that would be the best of both  
> worlds.
> 
Problem is that what is "most useful" to you is also not "darn easy."
If you use standard analysis packages, they will report errors for
a given frame, but not standard deviations of a set of frames, nor
calculate midpoints, etc.  You have an ancillary program that does that.
But more pertinent, even such techniques do not address systematic
errors such as the lack of transformation or not applying extinction
and it takes some education to obtain the right number.

I'm planning on talking to software vendors later this year (after my
Travel Period From Hell) about requiring a common output format for
use by the AAVSO, and as part of that I'll request common error analysis.
Arne
_______________________________________________