Re: Contour plots and data display.

From: Ray Hicks (ray.hicks@btinternet.com)
Date: Sat Oct 20 2001 - 09:56:55 EST


Mario,
I think we've both been (annoyingly for bystanders) consistent, actually:)

While this thread is still running, and we're agreeing that the number of
events is a  statistic necessary in aiding our understanding of graphs,
could I throw in another statistic that doesn't appear very often?

Confidence limits and levels(intervals) for percentages - which is probably
even better, since it gives an idea of the reproducibility of any percentage
given.

I'm often asked "how many cells do I need to acquire to be sure about this
1% population", and I refer people to a table that I picked up somewhere (I
think it might be from a FACSDesk manual!).  It's an easy statistic to use,
and can be calculated by analysis software  (FCSPress doesn't do it yet,
admittedly, and I haven't looked to see if anything else does), or can be
looked up in tables or gathered from the web (eg
http://www.surveysystem.com/sscalc.htm -if anyone's interested), so authors
should be able to estimate it for their existing data (knowing the
calculated percentage and sample size, assuming a population size far in
excess of the sample, where appropriate).  In cases where they can't get
consistently large samples (such as when they're dealing with small cell
numbers recoverable from hard-to-get systems as in the paper you cited), it
gives an idea of how shaky the percentages might be that's even more
informative than just knowing that they had 2000 cells for one assay and
20,000 for another.

Does anyone have any views on this?

1.5 GB per article, hmm... I wonder how much journals would charge for cover
DVD's - I wouldn't fancy doing the download over my modem!

Ray

> I think the most important thing to remember about graphs is that
> they display patterns of data and should not be used to estimate or
> present absolute count information.  In reviewing my previous
> comments to the list, my position is quite consistent:  contour plots
> are a great way to estimate population frequencies, not absolute
> counts.  I think we can solve both of our criticisms of data
> presentation simply by "mandating" that the number of events
> displayed in a graphic be displayed--preferably on or near the
> graphic.
>
> And yes, I agree that the primary data should be archived and
> accessible for the general public--this would be extremely useful.
> However, it may be somewhat problematic--currently, we generate over
> 5 GBytes of data per week.  The amount of data that went into my last
> paper was 1.5 GBytes--and that did not include all of the "additional
> samples" not actually displayed in the paper.
>
> mr



This archive was generated by hypermail 2b29 : Sun Jan 05 2003 - 19:01:35 EST