Reporting median values

Derek Davies daviesd2 at icrf.icnet.uk
Wed Oct 21 10:01:53 EST 1998



Dear Flow World,

I have read with great interest the correspondence regarding the reporting
of median values. As my lab is multi-user and heavily used, this is the
sort of question that comes up time and time again and it appears from
the correspondence thus far that there is no definitive "right" way of
presenting and reporting differences in histograms.

I have always used the median value expressed as a "linear" value for the
reasons that have been suggested - that it keeps the relationship between
values on the right sort of scale (Ray Hicks' example of values of 10, 100
and 1000 is one that I regularly use!). However, we should all be aware,
as Howard Shapiro points out that the log amps used have inherent
innaccuracies. Ultimately though, as things stand at the moment, just how
accurate can we be and how accurate do we want to be? I have users who
have wanted me to "prove" that two histograms that have a difference in
median values of, say, 10 units are significantly different. I usually
point out that, with all the stages of data processing and their built-in
approximations, that is nigh on impossible, especially as they often want
to do that on the basis of one experiment!!

All the inaccuracies are common to most, if not all, cytometers and are
surely common to both linear and log data. My response to users
questions in regard to how best to compare histograms is to use a robust 
population statistic (in unimodal distributions this is the median) on a 
linearised scale but, most importantly, be consistent and explain to
the reader what has been done. This may not be ideal as as cytometrists we
are always saying that one of the virtues of flow is being able to make
multiple measurements on a cell-by-cell basis and yet quite often we end
up quoting a single population statistic such as median when the
population concerned spreads over a log decade or more! However until
there is an ex cathedra (1) statement is this not the best way to go?

Of a greater concern to me is how to report multimodal histograms when
researchers are after a single statistic that will define the population,
but that is for another day, I think.

Derek

(1) How *do* you do italics in Pine?

****************************************************************************
*  Derek Davies                       Voice: (44) 0171 269 3394            *
*  FACS Laboratory,                   FAX: (44) 0171 269 3100              *
*  Imperial Cancer Research Fund,     e_mail: derek.davies at icrf.icnet.uk   *
*  London, UK                                                              *
*                                                                          *
*  Web Page: http://www.icnet.uk/axp/facs/davies/index.html                *
****************************************************************************







More information about the Cytometry mailing list