Re: Analyzing Data

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

Tom,

I suspect that a very successful minor modification to the
WS statistic would be for it to calculate residuals from a
given night's mean magnitude instead of a global mean magnitude.
This would get rid of the effects of zero-point changes which
aren't currently calibrated out.

Cheers,
Doug

On Sun, 4 Aug 2002, Tom Droege wrote:

> Since I did not get data last night, I spent some time this morning looking 
> at data I took on August 2.  A first look had found just 3 variable 
> stars.  This applying WS and looking at those with large statistic.  > 10 I 
> recall.
> 
> This morning I reprocessed the data and looked at WS > 1.  I found 13 
> pretty likely variables.  One with a WS statistic of 1.2.
> 
> When I send out data sets, they contain a number of evenings of 
> observations for each star.  This causes WS to generate big numbers for 
> what amounts to day to day variations, or for long term variation.  This 
> swamps the short term variations, so if you look at the data the way I have 
> passed it out, it is hard to find the short term variables where the 
> variation is small.
> 
> I started thinking about this when I noticed that we were finding mostly 
> variables abound mag 10 and lower.  There should be more around mag 
> 11-13.  So why was I not finding them?
> 
> The answer is that they are there, one just has to look carefully with 
> something better than WS.  OK, Doug has done a good job getting us a start, 
> but something more is needed, I think.  We need a way to look at the data 
> with WS over 1 and see what I can see if I look at it by eye.
> 
> Sometimes there is a big WS statistic and one look shows it is due to 1 bad 
> point.  Well, it may be real, but not likely.  Mostly stars are not 
> expected to brighten .2 mag for one measurement point.  So one can throw 
> such things out.  But something has to look and see what is happening.
> 
> I think there is a lot of work to be done to develop better algorithms to 
> find the variables.
> 
> Tom Droege
> 
>