"Thus the socially favorable mutation A2 tends to be lost or nearly lost in a random breeding population. In a population divided into many small, completely isolated groups, selection becomes reduced in efficiency as drift due to accidents of sampling increases. A2 may occasionally drift into fixation in a local group but cannot spread in the absence of migration." (Wright 1945, p. 417)Groups that drifted to fix altruism can provide more migrants to the population pool than other groups of equal size. Nevertheless, within-group selection against altruism is suppressed by within-group drift, and between-group variation can then become significant. In later multilevel selection models, within-group selection is swamped statistically by differences in average fitness between groups and drift does not feature.
For an example of later multilevel selection theory, the trait-group model of DS Wilson (1975. "A theory of group selection." PNAS 72:143-146) ignores drift within groups, lets the selfish individuals win within-group selection, but has them overwhelmed by variance in the average fitness between groups (given a cycle of groups dissolving in a panmixis and randomly re-forming from that panmictic population).
So I really see a difference between Wright’s model, where within-group drift and between-group migration rates work towards the same end, and MLS models where between-group statistics overwhelm within-group determinism.
- Wilson DS (1983) The group selection controversy: history and current status. Annual Review of Ecology and Systematics 14: 159-187.
- Wright S (1945) Tempo and Mode in Evolution: a critical review. Ecology 26:415-419. [see also the reprint in: Provine WB (ed) Evolution: selected papers. University of Chicago Press, pp. 395-399.]