Whilst I'm conversant with the study. I'm unsure if the 18 % increase for the rats on OO include all the rats ? as one was noted through luck or something else living longer than all the others on OO. If that is the case the 18% is skewed and is an abnormality which for percentage purposes if done by a statistician would be ignored. this would bring the average down to a lower figure.
I am pretty sure 30% of a group living longer is no outlier that would(or should) be ignored.
Skewed does not mean we can ignore the data. It just means its not distributed normally(quite a lot of data is, actually). The 18% must have been the mean of the OO rats compared to the mean of the water rats. Makes sense that from this perspective the mean of the c60oo rats was much higher, since almost all died at the end of the maximum lifespan observed in this experiment.
See, you can't exclude data just because it looks funny. You can only discard it if it is an outlier, and thats (by convention) mathematically defined as 1,5 standard deviations below q1 or above q3. I haven't calculated it but just by looking at the data I doubt this is the case.
It was 6 rats per group. In my country it is not allowed to use more rats per group unless necessary, by law. I can't recall the exact number, but it was really low, around 6-8. It seems to be standard for animal experiments. Do not forget, human studies often tend to use way more people to have a statistical value because we normally can not control the humans in an experiment a lot so we have to increase numbers and use other tricks to reduce the effect of unknown variables. In a lab animal model, there is a LOT of control:
- we can control what and how much they eat and drink
- we have control over their environment (physically and psychologically)
- we can use special breed animals that are very alike
We can pretty much exclude a lot of stuff that would influence the experiment otherwise. That means, we can use smaller groups and still get statistically valid results. Also, do not confuse experiments with observational studies, where we need even WAY more data, because we do not only have the benefits of lab tests I listed above but in addition we also can only observe, not manipulate (i.e. give group 1 substance a, group b substance b).
From that perspective I would say that randomness may not play a that big role, given the group size.