From: Allen
Date: 3/19/98
Time: 10:51:50 PM
Remote Name: 198.161.229.189
I agree with you both, however let's consider this:
The accuracy of measurement in terms of being correctly calibrated is actually important only for being able to determine the true number of spores per bee. This, in turn, is important only for reporting the degree of infection if absolute numbers are required -- for comparing my infection to yours, for example.
An error of even up to 20% in measurement, as long as it is consistent (ie. stating 55 ml as 50 ml every time when the same measure is used in each case) should not compromise an experiment conducted entirely in one facility when comparing infection changes due to treatment since the errors will not affect the before/after ratios. Such errors net out in the math.
Moreover, numerically small errors in high counts do not actually seem to mean much in practice since it appears that infections should be plotted on a logarithmic scale (non linear. Levels seem to increase exponentially, and the natural development is a curve rather than the straight line necessary for easy human comprehension. Plotting on the correct scale should eliminate the illusion.
What is really important is that whatever sampling is done gives the same results in the same lab every time. The actual numbers are not important except for comparing to results from another facility.
This is the real problem we have encountered: not that we have consistent errors or even that we do not replicate the microscope results of others, but rather that when we sample different sub-sets of bees from a mass sample, that we get widely varying results -- at least that is what has seemed to happen so far.
We have been busy with other things, so we have not had the time to get out and sample some more, but we plan to repeat the first experiment to see what is up -- even if we have to sample individual bees for a while to see where the problem is. I *suspect* it is simply this (exaggerated a bit): in a 50 bee sample there are 55 bees with virtually no nosema and 5 with quite a bit -- say 80 million spores each. That averages down to 4 million spores per bee over the 50 bees. But only 5 bees have any spores.
When we grab a 50 bee sample, sometimes we get 4 bees with and 46 without and sometimes 6 with and 44 without. That amounts to a variation of 50% in the results. However chance decrees that sometimes we get 1 bee or even none with and sometimes 10 with. That means widely swinging results.
When we are using 50 bees to represent 12 hives, if one or two hives have serious nosema and the rest are clean, then the above scenario is very probable. One time we get a high reading and the next no levels at all.
I really need to study stats, but I really think our samples are too small for the number of hives and the accuracy we seek. For simply a yes/no indicator, they are adequate, but they are not good enough for a good analog readout or determining less than total success of a treatment. Our current measurements are comparable to the idiot light on a car: either you have oil pressure or you don't. That's all such coarse measures can tell.
We need better accuracy in a comparative sense and internal consistency in our data from any one sample. IMHO.
Allen
[_private/disc4_aftr.htm]