UCD question: what UCD word to use for percentiles?

[Bo Milvang-Jensen]

Hello again,

My other main UCD question regards what UCD word to use for percentiles.

Background: our photometry (fluxes and errors) is input to our two 
photo-z (SED fitting) codes called LePhare and EAZY. The codes compute 
the distribution of each physical parameter (e.g. photo-z, stellar mass, 
SFR). From this distribution is found the "best guess" value. Assigning 
a UCD to that is simple (e.g. src.redshift.phot for the photo-z). But 
instead of providing an error bar on e.g. the photo-z, the codes provide 
either 2 or 4 percentiles from the photo-z distribution.

As far as I can see, there is no UCD word for percentile (except for the 
median, which is the 50% percentile). One workaround is to use stat.min 
for the percentiles below the median and stat.max for the percentiles 
above the median. Here is the descriptions of the photo-z percentile 
columns and the UCDs:

LePhare photo-z lower limit, 68% conf. level | src.redshift.phot;stat.min
LePhare photo-z upper limit, 68% conf. level | src.redshift.phot;stat.max

EAZY  2.5% percentile of photo-z | src.redshift.phot;stat.min
EAZY 16.0% percentile of photo-z | src.redshift.phot;stat.min
EAZY 50.0% percentile of photo-z | src.redshift.phot;stat.median
EAZY 84.0% percentile of photo-z | src.redshift.phot;stat.max
EAZY 97.5% percentile of photo-z | src.redshift.phot;stat.max

The description is different for the two codes, but the two LePhare 
columns could just as well have been called the 16% and 84% percentile 
values, as the (central) 68% confidence interval goes between these two.

None of these percentiles are the actual min or max values (that would 
be the 0% and 100% percentiles, respectively), so the used UCDs are not 
ideal, but they are our current best idea.

Alternatively, I was considering using the very non-descriptive 
stat.value (instead of stat.min and stat.max) to signal that this is 
something other than the best-guess value, but it tells the user very 
little.

These percentiles can be used to calculate error bars: e.g. the 84% 
percentile value minus the best-guess value (say the median value) would 
be something like a one sigma error bar on the positive side. But 
unfortunately I do not think it would make sense to use the UCD word 
stat.error for the columns as they are found in the catalogue.

Kind regards, Bo

On 01/03/2022 17.41, Mireille LOUYS wrote:
> hi again ,
> please do not hesitate to send your questions to me , or to the ivoa 
> semantics list in copy .
> this helps the community and maintainers like me to be aware of the 
> trouble you may encounter.
>
> many thanks , Mireille
>
> Le 25/02/2022 à 16:59, Bo Milvang-Jensen a écrit :
>>
>> PS: I have a couple of other questions/problems with the UCD standard 
>> that I would like to ask/report. What is the best channel?
>>
>> Kind regards, Bo
>