"No Treatment" Study

[WLC]

So when are you actually going to start making a list of statements/hypotheses about treatment-free beekeeping and the variables involved?

[Kieck]

ACBEES - I personally think you're correct, those beeks who are stationary and treat are the control group. But that's my opinion. -StevenG

As long as they do everything else just identically to the way you do, possibly.

Otherwise, you just introduce still more variables.

[Allen Dick]

> So when are you actually going to start making a list of statements/hypotheses about treatment-free beekeeping and the variables involved?

"Patience, Grasshopper"...

Or, you could go first...

[Kieck]

Step one:

Define "treatment."

[WLC]

As long as they do everything else just identically to the way you do, possibly.

Otherwise, you just introduce still more variables.

You don't need a control group, and there are many experimental designs that aren't bothered in the least by different treatments.

Have you ever heard of multifactorial studies?

Have any of you ever defended a thesis or taken a post graduate research design course?

You do realize that you WILL be under the scrutiny of others who do know a good research design when they see it, and they may not be as sympathetic to your cause as I am.

Case in point: of the articles and papers presented in support of treatment -free beekeeping that I've seen here, only 1 had a good research design, and guess what it showed was significant? Swarming!

KINDLY, list your claims and variables for treatment-free beekeeping, then go from there.

Like so:

1. Treatment free bees have a lower incidence of disease.
2. Treatment free beekeeping works best in isolation.
3. Treatment free bees develop hygienic traits.
4. etc., etc., etc. .

For variables:

• Swarming allowed.
• years treatment free.
• # of hives.
• commercial vs other queens.
• etc., etc., etc. .

You can then design a survey or use another design.

[Kieck]

Have you ever heard of multifactorial studies? -WLC

Yes.

But you're getting into a "study" here with so many variables -- most of them likely not even considered -- that even very complex statistical analyses are unlikely to tease out any real meaning.

[WLC]

Yes.

But you're getting into a "study" here with so many variables -- most of them likely not even considered -- that even very complex statistical analyses are unlikely to tease out any real meaning.

Kieck:

You need to be very objective here. Without any hypothesis to test, or variables that can be identified as dependent or independent, there's no there, there.

Claims/hypothesis and variables come first, and let the design come afterwards.

First things first.

Don't worry about the statistical tests. Those come with the experimental design which is nowhere in sight as of yet. They aren't as onerous as you might think they are. There are many techniques to make a 'mole hill out of a mountain'.

Step by step. Be objective, not subjective. Work smarter, not harder.

Yada, yada, yada.

[Allen Dick]

One of the problems of getting to ambitious is that assumptions and complications creep in.

KISS works as long as it is not overdone.

For example, we have several queen sources represented here already. Should we decide to track each stock all the way through? Is that possible, and at what cost? Or are we better off just admitting we probably cannot or will not do that given our level of interest and the time and resources available and just consider all as one group? Good question.

I had envisioned a very simple test, namely run a group of colonies for a preiod of time, record what was added and taken away and what of any significance was done, measure a few benchmarks like mite drops and nosema counts for example, and present them so that anyone can slice and dice them any which way they please.

The problem is to decide exactly what will be useful and doable and what will not since there is a cost to collecting data, and a cost to not collecting data.

[WLC]

Allen:

I think I see the problem.

You are ASSUMING that you are using a longitudinal study. Didn't I just warn everyone about putting the cart before the horse?

Bad idea for some very good reasons.

Rather than a longitudinal study, use a cross sectional study.

If your heart is set on following the progress of a group, simply give the cross sectional study yearly.

Longitudinal studies are known as a KISS off. They frequently fail because the sample size shrinks so much due to drop outs, that you don't have a valid sample size left before the study is complete.

It's the same as sending it to a 'committee'.

Just work on step 1 first. Step 2 is identifying objective and testable claims and dependent and independent variables. Step 3 is where a slick operator can group claims and variables into a powerful experimental design. Step 4 etc., etc. ...

[Allen Dick]

OK, Jim,

The point is though, that if the sample size shrinks to that point, the object of the study will have been achieved. Survival or extinction of this particular group is what this group wants to test, it seems. Perhaps not. We're discovering now what people are wanting to do and how much detail is achievable.

On the other hand, though, you may wish to help develop an experiment which can provide additional conclusions in the process. That does involve a lot more design and work, though and perhaps it is something the group can execute.

[dickm]

I did a small experiment this year with 20 colonies. Half of them got an extra slab of insulation under the outer cover. Whatever the results, I think the outcome would be meaningless because the sample size is too small. I had a few courses in statistics but don't know how that is calculated. How many colonies would I need to have a meaningful sample? This would be a place to start.

dickm

[Allen Dick]

If the difference between the two groups is obvious like 10 lived on one side and ten died on the other, it does not take a rocket sugeon to see that you proved something provided you did everything else the same. The confidence level is very high that what you see is what you see.

On the other hand, if the differences are slight and you are looking at brood area, disease levels, etc. as well to try to see the differences, the analysis gets more difficult.

Much more difficult, and it is much harder convincing any sceptics.

[sebee]

How many colonies would I need to have a meaningful sample? This would be a place to start.
dickm

If I remember correctly, you need to have a decent idea of the total population size in order to come up with some level of confidence statistically.

I think because there is so much variation in terms of genetics, geography and environmental factors, it would be difficult to "prove" much. But as Allen said, if your test shows clear results, then there is probably some correlation there that we can learn from in some way.

[dickm]

Thanks Allen,
What you say is clear. Let's see if I can ask a clearer question. If I had 2 hives and 1 died it would not carry the same weight as when 10 of 20 died. If 500 out of 1000 died It would be a much higher level of confidence. The chances of making any statements with 2 hives is negligible. One can calculate this. I've forgotten how. It is a calculation against what could be expected by chance. Commercial beeks are at an average of 30/35% losses now.
The start of this thread has 14 colonies and one treatment. Let's suppose that half of them die each year. Can we say that the dead ones died because they weren't treated? Or that the healthy ones are a result of management? I know they aren't commercial hives but If The average winter kill were 30% that would be 4+ hives. If 8 die, what are the chances that the extra four deaths are due to management style? The smaller the sample size the less likely the work is meaningful. It would be expressed as 1 chance in 10 (I just invented that number) that the result is not due to chance. I don't want to discourage anyone but many folks put a great deal of reliance on small studies.
This is basic stuff. Keick? Someone?

Dickm

[Allen Dick]

All good points. That is one reason to monitor health indicators like nosema and mites loads (both mites) because the effects of these pests are known within some broad limits and probabilities.

What you say does argue for controls, though.

[Kieck]

You need to be very objective here. Without any hypothesis to test, or variables that can be identified as dependent or independent, there's no there, there.

Claims/hypothesis and variables come first, and let the design come afterwards.

First things first. -WLC

Please see post #24.

First, we need to define terms. Let's start with "treatment."

[WLC]

Kieck:

While defining 'no treatment' is an obvious point, you can also allow for some variability in a survey by the choices provided. This also allows one to keep the sample size of the respondents from getting too low.

Some might consider feeding with sugars a treament! So, you include a question that indicates if they are feeding: 'please check any that apply: sugar, pollen, soy, yeast, ...'

[sqkcrk]

What about Michael Johnson? Isn't his "grant" to develope disease and pest resistant queens? How would he fit into this thing y'all are talking about?

[aylen]

Hi Guys

From a relatve newbee, I just wanted to say thank you for such an interesting and informative undertaking. Can't wait to follow the results!

[Roland]

I agree with Kieck. Define "treatment". I strike drone brood. Is that a treatment? Would it be possible to include me? What if I do not change my methods for the 5-6 years I think it will take for a sign of stability in events? If there are a fixed set of "can does" and "Can't does" , with some lattitude in between, can comparisions be made between groups?

Roland
Commercial with no Apistan, coumophos, Oxalic, bleach, etc.(non edibles)