22 September 2010

More insignificant thoughts on statistical versus biological significance

Statistically significant data are not necessarily biologically significant. That premise was the topic of 2 previous posts (here and here). A paper I read today rekindled my interest in the subject.

Giovas et al. (2010) showed that the shell dimensions of the marine gastropod Strombus gibberulus (humped conch) increased over a period of 3000 years at a site called Chelechol ra Orrak in the Palauan archipelago. How much was the increase? The mean shell length increased 1.37 mm (~4% relative to the smallest sample mean) and the mean shell width increased 1.22 mm (~8% relative to the smallest sample mean).

Shell length histograms of Strombus gibberulus over 3 time periods, Early (3000-1700 BP*), Middle (1700-1000 BP) and Late (1000-0 BP). The increase in shell heights is barely noticeable. (Fig. 4 from Giovas et al., 2010.)

Strombus gibberulus is a relatively small snail: mean shells heights are ~32 mm, the mean shell widths are ~15 mm. The reported shell size increases were statistically significant. But were they also biologically significant? In other words, does a 4% to 8% increase in the mean shell dimensions of a population of a snail species over 3000 years increase the chances of survival and reproduction of the larger individuals? That's one question the authors of the study don't address. They do discuss 8 hypotheses as possible reasons for the observed increase in shell dimensions. These include possible changes in the foraging practices of the ancient human occupants of Chelechol ra Orrak who presumably ate the snails and changes in various environmental conditions of the site.

Small increases in snail shell dimensions may indeed have biological significance. We will continue to ruminate on this problem.

*Before the present

Christina M. Giovas, Scott M. Fitzpatrick, Meagan Clark, Mira Abed. 2010. Evidence for size increase in an exploited mollusc: humped conch (
Strombus gibberulus) at Chelechol ra Orrak, Palau from ca. 3000e0 BP. Journal of Archaeological Science 37:2788-2798.


Psi Wavefunction said...

I've had statistically significant differences between two wild type ecotypes under the same treatment. Compared to all the other differences (treatment, no treatment, different mutants), this shift was negligible, and easily accounted for by slightly slower growth rates or something like it (thus, not biologically significant). Still bothers the hell out of me to this day thugh... should've had smaller sample sizes ;-)

fred schueler said...

once your sample size is adequate, the discussable measure of biological significance id r-squared, not p. In the present case, I'd like to see an inflection in the curve at the point where exploitation began.