Re: Database

[Tom Droege <tdroege2@earthlink.net>]

Michael S,

Looking at the engineering run data, I realize that you have put an error 
bar on the data.  This is great, and I like to see error bars on data.  But 
the question is, where did you get that error bar since I don't know what 
it is?

I suspect that you are using the measurement error that is included in 
Michael R's measurement.  This is just the error due to the statistics of 
the number of photons detected.  (Michael please correct me if wrong.) 
While this is significant for large magnitudes, it is meaningless for small 
ones, I think.

Where to get an error to use?

One way is to take all the stars with a reasonable number of 
measurements.  For this data >10 might be a good number.  Now compute sigma 
from the measurements for each star.  Now compute the mean sigma for all 
the stars for the same magnitude measured in some bin width that makes 
sense.  This will give you data for error vs magnitude.  One can do this 
because there are just not enough variable stars to bias the real 
measurement error.  I hope.  In any case, the error including variables 
will be a conservative one.  Now you could attach this error to all the 
measurements at a particular magnitude.

For the engineering run data one needs to be particularly careful.  Most of 
this data was taken following a field for as long as possible.  This 
produces a small scatter for a single night.  If you combine two nights of 
data where the star is in different positions in the field, the error is 
much larger.  So I would compute error for stars with > 10 measurements 
taken on > 1 day.

Sigh!  But this is still not the error that I would want to quote.   This 
does not include a zero point shift between the Mark IV measurements and 
some absolute measurement.  This is yet to be determined.  I am taking the 
present data in a way that includes Landolt standards.  So it may be 
possible in the future to tie the new data to them.

The engineering data was taken in a way that included few (if any) Landolt 
standards.  So we have to reference it to the tycho2 photometry which I am 
told is not very good.  Possibly someone knows an error that can be tied to 
the tycho2 photometry that was used.  This could then possibly be added in 
quadrature to the internal error determined above.  Hopefully some 
statistical expert will step forward and tell us the right way to do this.

The message is:

Don't trust anyone's error bars until you understand exactly what went into 
their determination.  Data without error bars is just not a measurement.

Tom Droege