Visual Observation Cadences

Hello Duane

You are thinking only of the errors and bias of the data.  What if the observer was interested in the short term variability of this star on a scale of minutes to hours--something like flickering.  Then their observations would be appropriate.  

Let the science dictate.

Gary

  Here's a possible related issue ... I've seen several examples where an observer has reported a visual estimate to two or more decimal places instead of the usual one decimal place. I've always been under the impression that only very experienced visual observers can achieve this kind of accuracy! I still remember my science teachers' lectures in high school about "significant figures" and how they would have frowned on this, when the comparison stars only are presented to one tenth of a magnitude. ☺

ZPA, I concur with your statement fully. Visual observations reported to an assumed accuracy beyond the nearest tenth of a magnitude is just so much wishful thinking with regard to true accuracy. In the first place, based on what more than 50 years with AAVSO has shown me, our comp stars' visual magnitudes are not more accurate than to the nearest tenth of a magnitude to begin with! Witness how often the brightness of many of our key comp stars in our sequences have changed values significantly over the years if there is any doubt concerning this. 

Observation cadence can likewise be a problem. Rapid observation cadence via CCD of known active, or flickering, variables is certainly valuable. However, reports of long runs on stars unlikely to show such activity makes little sense unless something unusual is, or has been, actually detected, or the observer is doing it in hopes of possibly finding something out of the ordinary and keeps these observations to himself.

Excessive visual observation of slowly varying stars is in my experience likewise a waste of the observer's time at the eyepiece. Too often I see nightly observations of stars the likes of MU Cep, or Gamma Cas, submitted by newcomers. Good for practice perhaps, but more likely just resulting in a skewing of our light curves.

In regard to all the various sorts of stars on the AAVSO's observing program our immediate past director, Arne, set out guidelines for the recommented frequency for observations by variables by type, or class, a few years back because of the over observation of objects. Janet had done the same sort of thing many years before that. As I noted already, excessive visual observation frequency can add at degree of unfavorable bias to our lightcurves, especially if done by new, or less experiencedI observers. I've recognized this in a number of our current light curves over the years and certainly currently.

My suggestion would be that the current director re-issue Arne's, or her own, observation frequency guidelines and post them prominently and permanently at several spots around our website for the benefit of all. 

J.Bortle   (BRJ)

Interesting discussion. I just found this link from 2010.

https://www.aavso.org/obs-frequency

Stephen [HSP]
New Zealand

Certainly the observing cadence should be tailored to the star.  There is a separate list of recommendations at https://www.aavso.org/suggested-observing-cadences-variable-star-types that agrees broadly but not precisely with the list at pukemaru's link, even though both lists are provided by the AAVSO.  The list here further notes that the suggested cadences are merely recommendations and can/should be modified for some stars based on the individual observer's experience with that star.  The Target Tool at https://filtergraph.com/aavso?orderby=obs_cadence provides yet a third set of recommendations for individual stars that is updated for special alerts and campaigns, etc.  I agree that it would be helpful to make such information more prominent. For example, it might warrant more emphasis even in the beginner's 10-star program at https://www.aavso.org/10-star-training. 

The question of bias is more complicated.  If (and this is a big if) we can correctly assume a positive correlation between too-frequent observation and average absolute noise, then excess observations by a novice will make the aggregate dataset noisier.  But if the noise is random, as noise often is, then there need not be any actual bias in the formal statistical sense.  Moreover, dedricksond reports that the excess observations that he noted seem to be of good quality (so, not necessarily noisier on average than those of a more experienced observer). 

If the argument instead is that visual observations are inherently noisier than CCD, DSLR, or other digital or "high-tech" observations (and therefore should not be allowed to dominate a dataset), I would note that this argument is not necessarily valid either:  I have worked extensively on data validation using Zapper at https://www.aavso.org/zapper and lately have seen more discrepant CCD observations than visual ones.  I recently mentioned this to a long-time PEPper who immediately agreed that many CCD observations are inaccurate. 

From a theoretical statistical perspective, as long as a particular source of data (e.g. a particular observer) is unbiased, the more observations the better, as RMS error tends (all else equal) to decrease systematically with larger numbers of observations.  Even for a single observer, it is easy to identify sources of noise that could render more frequent observations informative.  For instance, on any given night, there might be a thin haze of cirrus that may introduce noise into particular observations unnoticed by the observer, and it's possible that a more accurate estimate may be obtained on the previous night or the following night.  In such cases, reporting observations on three consecutive nights would provide a more accurate dataset even if the target star had not varied at all during those three nights.  That is indeed the nature of statistical noise and sampling theory. 

Even if it were reliably true that we face an inherent tradeoff between frequency and precision of estimates, researchers who use the AAVSO's database (or any database) can benefit from the addition of more numerous yet somewhat noisy datapoints to a baseline of a few exactly correct entries.  That's because numerous observations are needed to pin down a given star's exact light curve and how it may morph over long periods of time.  And we cannot be sure that there is always such a tradeoff. 

In a different application, audio engineers recognized decades ago that intentionally adding random noise (dither) to supposedly perfect digital recordings could actually improve the final result.  According to my recollection, that outcome was noted before the theoretical underpinnings were fully understood. 

Researchers using AAVSO data have the option of identifying each observer's individual datapoints, which can be used either to estimate and correct for an observer-specific bias or, presumably, to ignore the estimates from a particular observer.  Certainly we would like to make the job of researchers easier and avoid throwing sand in the gears, but that is one extant safeguard against bias, whether from excessive observations or otherwise. 

It is an interesting and valid point that BRJ makes about the imprecision of VSP's comp stars.  Some of that is simply due to the tradition of rounding to the nearest decimal, which one could make a case to revise.  VSX reports magnitudes with greater precision and it easy to verify that the rounding in VSP misprepresents some comp stars by nearly 0.05V.  But this is a shortcoming of VSP, not of visual observing or observers, and could be easily corrected by adopting a different convention (e.g., two decimal places).  Observers who worry about this (myself included) can look up their favorite comps in VSX and augment the VSP approximations where necessary. 

Some bias may also occur due to color mismatch between a comp star and the target, which is often necessary in order to fill in a workable comp sequence within a reasonable radius of the target.  I have often wished there were some way to calculate a "color-compensated visual equivalent magnitude" between a particular comp and a given target, but web searches have failed to turn up any reliable method for making such adjustments.   

I have read from two separate sources that experienced observers have been documented to make estimates with errors of 0.05V or less, so the supposed limit of 0.1V is not universal.  Similarly, I have consistently observed certain targets at times to fall distinctly between comps that differ by only 0.1V on the charts, or that even have identical VSP labeled magnitudes in a few cases.  In some cases the puzzle was resolved by reference to VSX, though I am less certain of the influence of color mismatch. 

At any rate, a fascinating and useful topic, well worth pondering and discussing. 

If you have a question about someones observing technique, ask them. This discussion happens frequently and goes for nothing if you don't ask the person concerned.

Alan.

Yes, I have noticed this too. I think we may well be talking about the same observer as well. However thinking back to when I first began observing (aged about 17) I dutifully observed gamma cas every clear night, because it was 'irregular'. Possibly the observer is looking at the period of (say) a SR star or an L-type - still I believe referred to as 'irregular', seeing it is about 100 days and thinking "best look at that one as often as possible because it's irregular!"