Someone brought up a very interesting kind of question to me recently.

The experiment goes like this. There are some chickens and we infect them with a parasite. We infect the chicken with one of three infection methods, A, B, C.

Now wait a bit, come back, kill chicken, and count the parasites in it.

The argument is that method C gives a more variable parasite counts, so A and B should be preferred.

The catch is that A gave much larger parasite counts in general than A or B.

The parasites undergo exponential growth in the bird, which implies that a normaly distributed initial dose would give rise to lognormal observed counts.

Do you think it is possible to test the H0 that all three methods were equally variable in their initial doses using this data. How?

Thanks for reading this very long post.