At 01:30 PM 2/12/98 CST, you wrote:
>Jerry raises many good points.
>I think though, that before we design the study, we shouldn't use "3 as
>a tie-breaker." Obviously in aggregate over apiaries, we can still find
>the effect with 2 (or 1) per condition.
 
 
I disagree.  3 is not ideal, but we have found within site variability to a
critical factor.  You can't address that as an aggregate over apiaries.
Mites respond to cumulative stress factors at a site.  Those will not be
the same over a group of sites.  AND WE HAVE THE DATA TO PROVE THAT.
 
We have tested hundreds of colonies over two decades for the affects of
everything from pesticides to pollutants to mites.
 
Your aggregate procedure is better fitted for some other toxic agents, but
not for these scenarios.
 
 
>
>Obviously we need to use some data (albeit sketchy) to
>estimate the statistical power of the design. We don't have any great
>data as to how no-treatment differs from Apistan/fgmo (I suspect it's a
>huge effect), but of course the more interesting question is comparison
>of the apistan and FGMO groups. I suspect we'll need a lot more than 3.
 
I agree, 3 is minimal.  But 3 gives us something that may work.  Anything
less is a waste of effort.
 
>If anyone can hazard guesses of what kind of variability we can expect
>within and across bee-yards and some general guesses as to what
>reasonable assays are of Apistan and FGMO, I'd be happy to do the power
>analysis as my small way of contributing to the general good.
 
Crank away, the numbers appear below.
>
 
I don't have to guess.  You will find on a basis of 7-12 colonies, that
most colony response data (population size, productivity, flight activity,
etc.) will fall in the range of 25-35% CV (or RSD) for health, unstressed
colonies, at a single site.  With even mild stress (pesticides, pollution,
mites) that number will increase to near 50%.  Under heavy stress, it will
often exceed 100%.
 
 
I can cite the publications, if you wish.  We also just finished 2 1/2 yrs
of data processing for groups of 7 colonies at 4 sites.
 
We also conducted a mite research project, looking at both TM and varroa
mites in 48 colonies over 3 yrs.
 
 
 
 
>Interestingly, some other statistical conditions may arise if there's an
>interaction between baseline level of infestation and efficacy of
>treatment, but we'll leave that as a possible "surprise" effect- OK?
 
Why?
 
>
>For what it's worth-
>Phil Wood
>[log in to unmask]
>
>>
>> STATISTICAL CONDSIDERATIONS:
>>
>> This is a great idea, but you need a Standard Method if you intend to
>> compare results.  As a first consideration, I offer these first step
issues:
>>
>> You should have at least 3 colonies per treatment, not 2.  One colony
>> represents an unreplicated experiment.  Two provides a bit of replication,
>> but what do you do if one of the two shows a response and the other does
>> not?  Three provides a tie-breaker (1 of 3, 2 of 3, or 3 of 3 colonies)
>> show --.  One of two doesn't demonstrate anything.
>>
>> Depending on the inherent variability in the system, even 3 colonies per
>> treatment is probably on the small side, but 3 is far better than 2.  We
>> often used 10-12 per treatment, but that is probably more than most can
>> spare for this test.  However, with enough participants, 9 colonies per
>> participant will probably be sufficient.
>>
>> You will have to test your colonies in some quantitative way for mite
>> infestation levels BEFORE you start the tests.  You should also
>> periodically monitor them during the conduct of the test, and at the end
>> (or when any of them checks out from mites).
>>
>> Drift will be a problem, so hive placement is an issue.  You don't want
>> untreated bees re-infesting treated colonies.
>>
>> Also, you shouldn't bunch the the hives up in groups.  For example, we know
>> that heat may reduce the number of mites.  So, if all of your controls are
>> in the shade and all of your Apistan or FGMO hives are in the sun, you may
>> get "mite control" from the heat of the sun (but think it was from the
>> chemical treatment - Apistan or FGMO).
>>
>> You also should choose 9 colonies of similar strength and you then have to
>> randomly assign your initial colonies to each of the 3 treatments (Control,
>> Apistan, FMGO) for statistical validity.  Let's say you pick 9 colonies.
>> Number the hives 1 through 9.  Now, put 9 slips of paper into a hat, each
>> with a number from 1 - 9.  Mix them up, then draw them out.  Let us say you
>> pull the numbers in this order 2, 7, 3, 4, 9, 6, 1, 5, 8.   We will then
>> assign the first number drawn to FMGO, the second to Apistan, the third to
>> the Control, and continue.
>>
>>
>> That allows us to randomnly assign our 9 colonies to the three test groups:
>>
>> FMGO  Hives 2,4,1
>> Apistan  Hives 7,9,5
>> Control  Hives 3,6,8
>>
>>
>> Why do this?  Again, this is a basic premise of setting up a statistically
>> valid experiment.  If we simply pick colonies, we may subconsciously pick
>> the colonies based on some bias (e.g., worst case, you pick the three
>> strongest colonies for the Control hives and assign the weakest to FMGO or
>> vice versa).  The weak colonies may be weak because they have more mites to
>> begin with.
>>
>>
>> For those of you with computers (everyone on this list), you can get your
>> computer or calculator to generate random numbers (but the hat is more
fun).
>>
>> If your colonies vary greatly in strength, you should be sure that each of
>> the 3 "treatment" groups has a similar array of hives by strength.  Again,
>> the control would have a strong, medium, and weak hive, the Apistan strip a
>> strong, medium, and weak hive, and the FMGO a strong, medium, and weak
>> hive.  However, it is simpler to start with similar strength hives.
>
>
Jerry J. Bromenshenk, Ph.D.
Director, DOE/EPSCoR & Montana Organization for Research in Energy
The University of Montana-Missoula
Missoula, MT  59812-1002
E-Mail: [log in to unmask]
Tel:  406-243-5648
Fax:  406-243-4184
http://www.umt.edu/biology/more
http://www.umt.edu/biology/bees