[vsnet-chat 3644] re Overobserving

[crawl@zoom.co.uk]

Taichi Kato writ:

>    I wonder whether analysts apply corrections for systematic differences
>between observers.
 
I should only speak for myself, but the general answer is NO.  Usually
observations are binned into means and the assumption taken that diparities
twixt observers cancel out cos they are effectively random [unless, of
course, these observers have been talking to each other, or reading others
results, which is a modern web worry we ought to think about I suppose
;^)].  This is usually freely admitted in the text of papers, whether
papers from amateur organisation or actual professional journals.

Around a decade ago Szatmary and others did some actual testing of this
against corrected data and decided the uncorrected stuff was just as
usable.  I can't remember which paper that was, but I vaguely remember it
was in response to a referee querying the use on non-corrected visual obs.

Lazlo Kiss used to be on chat, and he's been actively involved with much
Hungarian based LPV analysis, which has also included use of data from
vsolj, afoev, aavso etc, & been published professionally, and he may be
able to say what corrections they do or do not use for these "seriously"
refereed rags.

NOW, in a JBAA paper on BAAVSS data for mu Cephei some time ago, Chris
Lloyd, with others, noted that some observer correction method he'd
developed had been used on the data, I think it was a kind of bootstrap
arrangement.  When I asked him about it back in the Lower Lias he said the
methodology itself had never actually been written up for publication, and
he's not so active in this sort of thing as he used to be.

But basically, even if he'd published it, a lot of other folk would have
probably disagreed with his arguments, and then anybody using them would
have to worry if their entire analysis was invalidated by an assumption
made at the pre-processing stage!  You see, it's alright suggesting
corrections are used, but when nobody can agree on the validity of any
method, well... ...if you can fairly safely assume non-correlation of raw
data, then you don't need corrections cos that's what statistics is all
about [I bet I get some stick for that last bit].

But really the main reason corrections aren't made is because analysis is
now done mostly on computers.  You take an electronic archive of data, trim
it and clean it up, average it out or not, and then throw some software at
it.  Then you sometimes use some, computer software based, statistical
tests on the results, or you graph the living daylights out of them.

If you want to do some corrections you're going to have to get at the data
on an almost "by hand" level and pick through it observer by observer etc.
If you were doing the whole job by hand you may include a correction route
in the analysis, without much difference in workload.

Simple choice... ...analyse 10 or more LPVs in a day semi-automatically, or
maybe one LPV in about a week!

It's not just "idleness" though, I've analysed several hundred lpv
lightcurves of varying quality and length from many sources, and 99.5%
[ish] or more do bugger all interesting.  It can be a bit daunting
analysing a century worths of data for a Mira variable over several days,
only to come to the conclusion that the period is so many days, with no
other information gleaned, especially if this period is no different from
the one in GCVS4, GCVS3 or even GCVS2... ...feels like a real waste of
time!!!!

So, it is tempting not to get bogged down with irrelevant niceties.  Having
said that, these niceties become increasingly more relevant as the object's
amplitude approaches the limit of visual errors, ie around +/-0.2 mag.



Finally, I've got a 80% complete analysis from a long run of AF Cyg data
from VSOLJ which clearly shows that the _tertiary_ period was quite stable
at the current 920 day ish value between JD 2424000 and 2437000, but cored
about a slightly shorter value twixt from the last date up to JD 2444000
whence it returned to near 920 days.

Most folk quote about 915 days or less for this tertiary period, because
this above fact is hidden in an averaging out by normal DFT
lumping/superposition of all periodicities, and the low amplitude tertiary
period is definitely beyond O-C.  The chewy mass of long duration vsolj
observations are instrumental in showing this period change.

Is anybody in Japan interested in this, or can I leave it skulking
surreptitiously on the dark recesses of my poor defenseless hard disk, and
forget about it?

The point being that stochastic near-saltatory variation twixt two stable
periodicities is a well known [but not understood] pulsational
characteristic of LPV period changes, and in this case at least negates any
models for the handful of triple period LPVs where the tertiary period is
envisioned as being due to the expression of orbital/apsidal/rotational
motion.

Cheers

John

John Greaves
UK

VSNET Home Page

Return to Daisaku Nogami

---

vsnet-adm@kusastro.kyoto-u.ac.jp