Re: Variability statistics

[Stupendous Man <richmond@stupendous.cis.rit.edu>]

> I have been working on improving the reliability of the W-S statistic
> (or getting rid of the spurious detections, take your pick).
> Then, I tried excluding the minimum and maximum observations for each
> star. This seems to have improved things greatly. There are far fewer of
> the "one high measurement" stars in the top W-S. Of the top 100 W-S
> values, 96 match a known GCVS/NSV variable. The other 4 look obviously
> variable.
> These new statistics are now available online.

> I have just recently tried using the modifications to the W-S statistic
> described in:
> http://cdsads.u-strasbg.fr/cgi-bin/nph-bib_query?bibcode=1996PASP..108..851S
...
> This procedure also seems to produce better results, though I just
> completed it yesterday, so I haven't had time to look at the data in
> great detail.

  For those who haven't read the paper yet, Stetson provides several
small modifications to the W-S statistic which make it more robust
to individual outliers, and also permit more easily the inclusion
of individual measurements instead of pairs of measurements.

> Would anyone care to comment on whether any of these would be
> significant in determining which method is best, or what metrics I could
> use to make that determination?

  I'd say this: look at the two "new" techniques, Stetson's and
yours.  Write a script to pick out the top N stars, rated by
each of the two methods, and make light curves for these candidates.
Scan through the light curves yourself, by eye, and count 
the number of candidates which are "obviously" bogus.  Yes,
it's a judgement call, but you probably have a _very_ good
idea of the common errors which creep into the Mark IV data.

  Whichever method yields fewer bogus candidates in the 
top N wins.  Try N=10, N=20, N=50.  If one method consistently
does better, go with it.  If the two methods are about the 
same, then you face a tough decision: pick the one you devised --
it's fast, it's easy, it's YOURS -- or pick Stetson's -- it's
described in great detail already, so you'll never have to 
explain it to anyone.  It won't be so bad -- you can't lose
either way :-)

                                        Michael