<the problem we will have if we use several control groups with bee
research is that the control groups are likely to differ enough to cast doubt on
the results.
this is not directed towards any research or researcher...but a real
problem with the validity of results in general, and with bees in particular.>
That's the whole point. You have to design an experiment in such a way
that there is sufficient replication so that the inherent variability does
not produce false results.
The treatment effect has to differ from the control variability to the
degree that the odds of the 'treatment' result being simply due to normal
variability (that's where the controls come in) is 5% or less.
You don't have to have more controls than treatments, but each set
(control), treatment 1, treatment 2, etc. has a sufficient number of colonies (or
cages of bees) so that you can see a real difference, if there is one.
In general, you can't use 1 colony for a control or treatment, since you
can't get any measure of variability from a sample of one. If you use 2,
you've still a problem, especially if one colony is dramatically different
than the other - which one is representative of colonies in general? So,
we're usually stuck with using at least 3 colonies for each control set and
each treatment set (that way, one can hope that at least two of the three will
be similar).
In practice, there are experimental designs that use multiple control
colonies and multiple treatments all in the same apiary. That minimizes a
problem known as pseudoreplication. If you spread these same colonies over two
apiaries - you have to be confident that the environmental conditions,
forage, etc. are more or less the same. If not, you may have an 'apiary'
effect.
The opposite approach would be to have one control colony and one treatment
colony for each level of treatment (say dose of pesticide) in an apiary,
but then repeat that in say 30 apiaries.
Yes, you only have a sample of one from each locations, but you can gain
power in the analysis by replication across locations.
Now, costs enter in, so one often sees all colonies in one location with
multiple replications (colonies) or single colony sets at multiple locations.
IF cost is not a problem, you might see well replicated trials at multiple
locations - but that's rare.
Also, if you use the same colony(s) and take measurements over time, that's
something called a 'repeated measures' trial. But, those are NOT
independent replications, so one needs to recognize that and correctly apply the
stats.
There are three common mistakes made, even by experienced researchers:
1) Pseudoreplication - for example you can't have all of your controls in
Apiary 1, all of treatment 1 in Apiary 2, all of treatment 3 in apiary 3,
etc. There are numerous statistical articles addressing pseudoreplication,
which is a real problem in many environmental studies such as work done at
EPA Superfund sites.
2) Treating the results of a repeated measures trial as independent
replicates - they're not - all of the measurements from one colony still only
represent one colony. So, if you take 4 measurements in a season, that's not
four replicates.
3) Insufficient replication - this goes to the original comment. Bee
colonies are noisy systems. We all know any apiary will have weak to strong
colonies. You have to have sufficient replication (number of colonies) to get
a fix on the variability - if that number comes up to more than you can
afford to do, as a scientist, you've no option. One has to have sufficient
numbers of colonies to statistically 'see' a difference. If not, its a
wast of time and energy.
Stats books have formulas for estimating the number of replications
(colonies in this case) needed. That number also affect how small a difference
can be seen. If you have three control colonies, and three colonies treated
at the same dose of pesticide, and if that dose is greater than the
expected acute toxicity, you should see three surviving control colonies and
three dead treated colonies. In that case, three colonies for each is
sufficient. But, if you then try to find the threshold for demonstration of
sublethal effects, the numbers of colonies required will go up dramatically -
you need enough colonies to show that the subtle effect is greater than the
normal variability.
Finally, the more treatments you use, the higher the costs of the trials,
and you then up the ante. Now you have to do multi-variate analyses, and
look for interactions, etc.
Bottom line, if you don't understand stats, you need a good statistician
to design your study BEFORE you do anything. If the stats person says you
need x number of colonies to see y level of effect, and you say - I can't
afford that, then you have to make a hard decision.
Jerry
***********************************************
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
|