Let me try and clarify:

if we have a study with 40 hives involved:

in a standard study, 50% are "T"reatment, and 50% are "C"ontrol.

some kind of averaging or smoothing of data (not using data from the colonies eaten by bears, for instance) is performed and we end up with T results and C results...the study compares them and tells us if/what the result of the treatment is.

what i'm _NOT_ proposing is just making the pool of control colonies larger to come up with a better "average"...a more statistically significant result.

what i _AM_ proposing is that instead of the above experiment, run 3 SEPARATE control groups of 10 colonies each, and one treatment group, also with 10 colonies (assuming that an n of 10 is statistically significant to the study in question).

if we average out all 3 control groups, we get a large control group, 3X the "n" of the treatment group.

but, if we treat each of these control groups separately, and compare them to one another, we learn in a real world situation if any of the controls are giving us a meaningful baseline to compare with.

if controls can't be replicated, then comparing any one set of results to treatment group is more or less meaningless (depending on the spread and the numbers).

i know enough about statistics to know i know nothing about them (anyone who feels like they understand stats should try working on the monte hall problem for a while....this is the best lesson that what is obvious isn't true).  i understand the value of statistics....but how many studies that are supported by the statistics (in design and result interpretation) would stand up to the multiple control groups i've proposed?

i don't care what the stats say, if the control can't be repeated over and over in parallel with the treatment group, then it isn't a useful base for comparison.  it seems to me a real litmus test to use multiple control groups....casual reading (without a statistics background) will reveal if the control really is typical under the conditions tested.

yes, i know, $$$$...but given the inherent variability between colonies, an approach like this is cheaper than running studies over and over....researchers can see whether their controls are yielding useful enough data to make any conclusions at all....i don't believe this is always the case currently.  this can only lead to better research, no?

deknow

             ***********************************************
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

Guidelines for posting to BEE-L can be found at:
http://honeybeeworld.com/bee-l/guidelines.htm