I get excellent images from MPO and use AAVSO sequence. SNR of target and comps lays in the range 50-400, check star estimations are close to real values. All is OK, except abnormal high errors about 0.08-0.09. All attempts to use shorter sequence increase check star deviations. I do not understand the reason of so high errors. Maybe longer exposures are needed or something else?
Hi,
most of your comp stars are dimmer than the target star (FW Boo) that increases your errors on the photometry. Try to remove comps 152, 140, 144 and rerun your analysis. SNR of 100 gives approximately an error of 0.01.
Josch
Thank you, Franz! Of course, I tried to use shorter sequences. Indeed, according to you recommendations errors decreased to normal level, but regrettably check star deviations increased dramatically:
5 comps: 0.002B, -0.007V
2 comps: 0.092B, 0.075V
Which way is more correct for photometry - high error and low check star deviation, or reasonable error and higher deviation?
Mikhail:
You have been observing for a while. What is your opinion/judgement? Can you think of circumstances where one or the other may be appropriate? Which puts more scatter in the lightcurves of āmultipleā observers on the LCG?
Ken
Hi Ken, Iām glad to see that you keep your method of questioning 
Truly speaking, I suppose good coincidence of check mags is more important than error. One can get extremely low error with single comp star, but longer sequense is more preferable.
But in this case Iām not so confidentā¦ Maybe two comps will be enough, in spite of check star deviations?.. Errors are so huge!
Mikhail,
Remember what the purposes of the check star are. The first is to ensure that the comp star is not itself variable. If it does exhibits short term variability, the check star light curve will not approximate a horizontal line. And of course it may be that the check star is variable! If you make enough observations over years, you will experience both situations. Experimenting with different sequences solves the problem if it arises.
The second purpose of the check star is just as it name suggests - to provide a reality check on your measurements and calculations. The measured magnitude of the check star should be close to its catalogue magnitude.
Provided that the precision of the check star measurements is acceptable, it does not matter if its measured magnitude differs from the catalogue value by a little more than you would like to see, particularly if at the same time the error for your variable star measurement is very small. In my opinion, this is the preferred situation.
Concerning comp star ensembles, the current guideline supports their use, and there are good reasons for this. Also, and particularly if you use VPhot, iteratively selecting different ensembles allows you to see easily how to optimise your errors.
However, I have personally never considered it mandatory to use ensembles. It is interesting to compare the calculated magnitude of the variable using the individual members of a comp star ensemble over a number of nights. For example, in my experience the check star magnitude will stay pretty much constant with each particular (individual) comp star, and there will be consistent differences between the magnitudes of the check depending on what comp star is use for the calculation. You might expect this, because it may be that the catalogue magnitudes of the comps will vary in their accuracy.
One other point about ensembles of comps is that, if one of them is in fact slightly variable, it may be that this will not be revealed in the check star light curve because the variability is āswampedā by the data from the other comps. You may argue that does not matter because one purpose of the ensemble is to average out (minimise) errors, but it has always bothered me that variability may be hidden in this way.
Considering all of the above, and because I obtain time series of non-transformed V magnitudes of short period eclipsing binaries using a camera with a 14 bit sensor, I prefer to have the comp B-V and V as close as possible to those of the variable, and therefore mostly do not use ensembles. This is a personal situation. It will not apply to many observers.
Roy
2 Likes
Thank you, Roy, for detailed explanations!
Well, probably compromise search between check star precision and target error is correct way, but it leaves too much for human factorā¦
Your note about check star LC is very important! Is it possible to see in VStar plot both check and target light curves???
Just a general hand-waving comment on photometry errorsā¦ 
Astronomical photometry (in most of our cases synthetic-aperture-photometry using data from 2-dimensional detectors) has many sources of errors. Some are random, some are systematic. Some errors are yours, some are intermediate such as the uncertainty that comes from the transform coefficient determination, some are comparison star ācatalogā errors which can be surprising large. Random errors improve with averaging, systematic errors (probably!) donāt.
Side note: The AAVSO VSP gives estimates of the uncertainty of the comparison star magnitudes.
I very rarely use ensembles. I prefer, at least with the current extended data format, to have users be able to know which comparison star I used for traceability purposes.
Below is a link to a thesis I found that might be fun to read, p. 58 and beyond perhaps. Maybe others have some good papers they can send a link to that discuss the full photometry error budget from photon collection to catalog errors.
Jim (DEY)
2 Likes
In Russian: CCD photometry of variable stars
1 Like
Thanks,
My transliteration/translation skills are nill these days. Iāll have to send it the AI 
to translate it for me!
Jim (DEY)
These two works, the Packard thesis especially, seem to deal largely with mathematical aspects of the problem, and certainly have no practical application within them. Neither presents results where the data for a target field are taken along with standards to determine transformations, dealing with non-linear transformations, radial gradients in the image that are not corrected by flats, and other subtle issues.
I tend to follow the lead of Peter Stetson, who invented DAOPhot, and whose precepts are laid out far more simply and various subtleties are discussed. A good place to start is his ā48 globularsā paper from 2019, which Iāve cited several times on this forum.
More specifically regarding photometric errors, work by Ben Taylor and Mike Joner are helpful in elucidating statistical issues. Two examples are:
notice the snapshot statistical exercises in the appendices
the procedures outlined here are of course broadly applicable
Although not trivial, none of these papers contains the forbidding mathematics of the two theses mentioned above. In addition, extensive on-sky results are presented, lacking in earlier ones.
\Brian
1 Like
The few Stetson papers I found with Google Scholar contained little on error budgets in general and propagation of errors, etc. These papers are a good start for sure in understanding errors and their sources.
Jim (DEY)
Jim wtote: I very rarely use ensembles. I prefer, at least with the current extended data format, to have users be able to know which comparison star I used for traceability purposes.
That is also my policy.
However, an ensemble of well-chosen comps allows an easy error estimate (SD of the target mags from the individual comps) that encompases all sources of error. I believe that is what VPhot does.
There is no corresponding simple error estimate for time series of short period variables when only one comp star is used. The SD of a set of check star measurements if the check and target are of similar brightness and colour is a surrogate, but of course cannot be used as a formal error estimate for the target itself.
For long period variables the SD of a set of measurement across a short time span encompasses all errors.
Otherwise, attempts to achieve ātrueā error estimates of target star measurements are complicated, as the thesis Jim referenced describes.
1 Like
ā.. an ensemble of well-chosen comps allows an easy error estimate (SD of the target mags from the individual comps) that encompases all sources of error.ā
In this situation magnitude measurements should be transformed otherwise the use of an emsemble may give a false sense of security that the target mag is accurate.
Google translate does a decent translation job on the Russian text. Any embedded text in the figures do not translate, however. Translation of those are left as an exercise for those who are interested in extra credit! 
I always wanted to say or type that!
Jim (DEY)
Roy:
I agree 100% with these quoted statements and that is the main reason why I regularly use an ensemble.
And yes, an ensemble may only ensure āconsistency not necessarily accuracyā BUT IMHO it is better than hoping that one comp provides any better accuracy than another. In fact, I believe the opposite is true when many observers each use different single comps.
It will continue to be our most contentious choice/decision! 
Ken
1 Like
Ken,
I didnāt explain in the above post (although Iāve mentioned it elsewhere previously) that the main reason I use single comps and not ensembles is that most of my targets are short period EBs observed for VSS and the purpose of the observations is to determine times of minimum as accurately as possible.
Therefore I (edit: almost always) use a single filter (V). Measurements are thus not transformed, and the single comp star is chosen to be as close as possible to the variable in colour and magnitude. I know my Tv_bv (itās about -0.05) and can therefore calculate a likely error (measured minus ātrueā magnitude) due to B-V target-comp differences, although that error is never used in reports.
To achieve a similar outcome with an ensemble of comps and non-transformed measurements the spread of comp B-V values would need to extend from blue-ward of the target to red-ward of the target. In practice, thatās not necessarily easy or even possible.
Photometry of MASTER:
The quotation is from a paper reporting the methods and results from a CCD survey in the R band of variable stars and exoplanets with a large aperture telescope, if I peruse the paper correctly. The numbers quoted need to be interpreted in context and not generalized.
2 Likes
Thank you, Roy, for the important note!
The link was given just for the information and comparison.