Search the CONFOCAL archive at
http://listserv.acsu.buffalo.edu/cgi-bin/wa?S1=confocal

I agree, billo, but that does assume that the statistics are done 
properly.

If we are talking about data integrity and reliable reporting, then a 
whole squirming bag of worms, leeches and various other unlikable 
beasts is opened when we bring up the "S" word. As one who does a lot 
of statistical analysis for folks around here, most people do not 
understand either the principles or practice of the statistics of data 
sampling and its subsequent investigation. There are plenty of 
excellent programs around, as well as a whole raft of texts at various 
levels, that allow sophisticated data analysis with relative ease these 
days. Nevertheless, most students, and most researchers, at least in my 
field, have only rudimentary knowledge.

There have been studies done in the medical literature showing that a 
very high proportion of papers use the wrong stats, and that many of 
those papers came to the wrong conclusion as a consequence. I don't 
know if that applies equally in all fields, and in some journals at 
least, the standards of statistical analysis are now set pretty high. 
But there is still an awful lot done badly, if at all... So we get 
oversamp[ling, undersampling, biased sampling (not Nyqusit, but at the 
data level), we get inappropriate "normalisation", over-use of 
percentages, inappropriate use of simple tests (eg t-tests, or oneway 
ANOVAs) to deal with complex experimental designs, resulting in a 
tendency to get "significance" when none exists, and then the infamous 
phrase "... X and Y happened leading to Z, but it did not reach 
statistical significance ... "


Unfortunately, there's plenty more where that came from...

IAN




>
> We live in a world in which we have a limited space to demonstrate 
> results.  Most journal articles are there to demonstrate results, not 
> provide comprehensive data sets. Thus, the appropriate illustration is 
> that which best demonstrates the finding.  If you have 1000 trials and 
> of these 700 show a result and 300 show nothing, one is not obligated 
> to provide 1000 images, or even 100 or even 10.  It is more useful to 
> provide an illustration of the positive finding and note the 
> statistics, assuming that the appearance of the negative finding is 
> intuitively clear (i.e. positive shows staining and negative is dark). 
>  If the difference between positive and negative is unclear, it may be 
> useful to provide an image of each, but I don't think that's usually 
> necessry. I don't see any need to take up multiple pages of images of 
> negative results just to demonstrate statistics that are more properly 
> described succinctly in the text.
>
> There is simply no way to provide all of the data in most things in a 
> couple of pages.
>
>
> billo
>
>

* * * * * * * * * * *
Prof Ian Gibbins
Anatomy & Histology
Flinders University
GPO Box 2100
Adelaide SA 5001
AUSTRALIA

[log in to unmask]
voice: +61-8-8204 5271
fax: +61-8-8277 0085

http://som.flinders.edu.au/FUSA/Anatomy/
http://www.flinders.edu.au/neuroscience