Dear observers, explain please why mag errors decrease with decreasing of compstars number? Which estimation is more correct - with great error and long sequence, or with lower error and lower number of active stars?
With ensembles the reported uncertainty estimate (the VPhot "Err") is based primarily on the standard deviation of the measurements of the active comps. If you cull the active comps list by removing the outliers the s.d., thus the reported Err, will decrease. If you keep removing active comps until you are down to a single comp star one might expect the reported uncertainty to be zero.
With ensembles VPhot also combines the s.d. of the comp measurements in quadrature with the inverse of the SNR of the target so you will never get down to zero. (I think it also somehow uses the uncertainty of the active comp[s] in the estimate.) When you cull the active comp list down to just a few stars you can force the reported uncertainty estimate down as well. The more you cull the active list, the better your Err appears.
The problem with this strategy is that the power of ensembles depends on the number of (good) comps used. The more the better. This is how the uncertainties and other errors in the active comps can be averaged out to give a more accurate estimate. The trick is to remove only the worst comps and keep the rest.
Thank you, Phil, I will try to keep golden middle way (Om Vajrasattva Hum!), even in the case of high error.
Truly speaking I do not believe in mags precision like 0.002, but probably somebody can achive one. For me 0.05 is fully enough, or even 0.15 in the worst case.
It seems to me that experienced observers using VPhot and ensembles of AAVSO comps typically report V uncertainties around 0.025 more or less. If you look at the V photometry for the AAVSO APASS comps I believe you will find those uncertainties are similar.