"I suggested that data points near each other in the scattergram should be looked at to see if that general area of dose and exposure time resulted in "similar tangible results", as similar observations would tend to enhance the credibility of that subset of the data points"

This assumes that all other experimental variables are constant....you ignore all but varying dosages and duration of contact.  Thus, it is impossible to use this approach to look at "credibility".  All you can do is look at correlations.

The scattergram will give us clues, as you point out, because if some points cluster we would expect to see more similarities in those studies than exist between points that are more widely scattered.  Then we'd look more closely at those to see how they may be correlated.

Still, whatever you decide to conclude once the job is done, creating the plot seems to me worth doing.  We can then each draw our own conclusions, share them around, and proceed or not based on our individual (or collective) interpretive and investigative talents.

So here's how Jim said he'd set this up:

"We can certainly plot a scatter diagram of doses in ng per bee ( 1 ng per 100 mg bee is 1 ppb) versus exposure times in hours or days and see the spread of doses among the studies with bad outcomes for the bees. We could go on to plot the doses and exposure times in ng per bee for the studies with a claim of no bad outcome, and see how the scatter diagrams overlap. If we wanted, we could also try to rank the outcomes in order of severity, and get a sense of how dose related to severity."

So, the X axis would be dose and the Y exposure time, with each point in a color designating whether the outcome was good or bad for the bees (% survival we call "good" would have to match the controls).  If you want to get really fancy you could do a 3-D plot and put severity on the Z axis....with Z = 0 being where all the good outcomes should fall, because the test groups don't differ from controls in those.  (But I'm not like Spock, I think best in 2-D chess....)

If both colors occur in a cluster, we would have to closely review those studies to see what other factor could account for the discrepancies.  If clusters occur in just (or mostly) one color, then there is agreement.  Most astonishing would be if there was actually a linear trend to the data, and a clear threshold between the two colors.

The only other rules we need are agreement on the way to verify data extracted from each paper that is included on the graphs....and what types of experiment are to be included (contact, oral, foragers only or other bees too, etc.).

I  have found at least 50 more papers and am still digging (but I'm getting close to the bottom of the barrel I think).  Out of all these 100+ papers, we'd be lucky to get a subset big enough to be useful....but I'm game to find out.


Christina

             ***********************************************
The BEE-L mailing list is powered by L-Soft's renowned
LISTSERV(R) list management software.  For more information, go to:
http://www.lsoft.com/LISTSERV-powered.html