Re: test for small paired sample size

Adam schrieb:

The P value for the difference in slopes isn't significant if you log transform the data first, which I suspect would be a sensible thing to do given the way the data look. But of course you can't know whether it is or not from that sample size, unless you know whether the outcome variable is generally known to be normally or log-normally distributed.
We expect log normally (further research may be show that) but OO and zero will never be reached for a.
means a will not decrease below a steady state, even b is growing up maybe over 100.000 and minimizing b f.e below 50 will destroy the experimental setup

I'm not sure I agree that everyone can realise that "a" is increasing and "b" is decreasing. Probably, yes, but how do you know the fourth value in the series would continue the trend?

isn't this always the problem if there are only three points?
Couldn't you find a lot of fitting functions like sin cos tan and much other into three points ...

I totally agree, however, about telling the journal editor to sod off and get some common sense (again, ideally phrased in a more diplomatic manner).

I am afraid a p-value form a not fitting statistical model is more accepted than any proper illustration ...

Regards Knut
.