280 Studies Confirm: Natural Selection Acts on Groups, Not Just Individuals

The dominant framework in evolutionary biology for the past 60 years has placed natural selection squarely at the level of the individual organism - or more precisely, at the level of the gene. A trait that helps an organism out-compete rivals propagates; a trait that harms it disappears. The elegance of this view, associated with the work of William Hamilton and Richard Dawkins among others, is that it can explain phenomena that superficially appear to benefit groups as products of individual and kin-based selection.

The view is not wrong, exactly. But a bibliometric review of 280 scientific studies published across more than a century of research now argues it is incomplete. The article, "Abundant empirical evidence of multilevel selection revealed by a bibliometric review," published in Frontiers in Ecology and Evolution, finds that natural selection demonstrably operates at multiple levels of biological organization simultaneously - at the level of the individual, at the level of the group, and sometimes at higher levels still.

What Multilevel Selection Means

"The idea of looking at selection at multiple levels is to measure whether a trait is adaptive for individuals within a group," explained co-author Anne Clark, Associate Professor Emerita of Biological Sciences at Binghamton University. "And does the frequency or existence of that trait within a group change the way the group functions in comparison with other groups?"

The distinction matters because the same trait can be simultaneously disadvantageous to the individual carrying it and advantageous to the group containing it. A worker bee that stings an intruder dies, but protects the hive. An individual that shares food rather than hoarding it reduces its own reserves while increasing the group's collective resilience. Standard individual-level selection analysis can accommodate these examples through kin selection and inclusive fitness calculations. Multilevel selection theory argues that this is a mathematical equivalence, not a mechanistic explanation - and that understanding which level of selection is actually driving a trait requires measuring both.

The Review's Scope

The 280 studies span organisms from viruses to human beings. All attempted to account for multilevel selection - meaning they measured whether traits affected fitness at both individual and group levels, not just one or the other. The review was led by Cesar Marin, a soil mycorrhizal ecologist, and included behavioral ecologists Clark and Conner Philson, evolutionary biologist Michael Wade, and Omar Tonsi Eldakar of Nova Southeastern University. Binghamton University has been a center for multilevel selection theory since 1988, through the long research program of Professor Emeritus David Sloan Wilson.

The finding that multilevel selection is detectable across such a broad taxonomic range and such a long historical span of studies suggests it is not a niche phenomenon confined to social insects or exceptional cases. It appears to be a general feature of biological systems.

Why This Remains Contentious

The scientific resistance to multilevel selection has deep roots. Since the 1960s, influential researchers have argued that claims of group selection were not subjected to rigorous quantitative measurement and should not be treated as explanatory. Some scientists banned discussion of group selection in their classrooms, calling it naive. Others argued it was vanishingly rare in practice, or that every case of apparent group selection could be mathematically reframed as individual or kin selection.

"If you measure the average increase in the frequency of a trait over generations and then say it's favored by natural selection, you're not wrong," Clark said. "But if I ask you: 'What's the mechanism for the slow increase in that trait over here and the rapid increase over there?' you're not going to be able to tell me. Whereas, if you had looked at different levels, you might see that group competition is more important in one place, or cooperation within groups in another."

The argument is ultimately about mechanistic explanation versus mathematical equivalence. Individual-level selection accounts can be constructed for most multilevel phenomena. But knowing that group competition is the dominant force driving a trait in a given population provides different and potentially more useful information than knowing only that the trait is increasing in frequency.

The Chicken Experiment

A classic illustration comes from poultry breeding. Researcher William Muir selected the most productive individual hens within cages and bred them to produce the next generation. The result was aggressive, cannibalistic birds that achieved personal productivity at the expense of cagemates - and productivity at the cage level declined. When Muir instead selected the most productive cages and bred all hens within those cages, five generations produced a docile strain with a 160 percent increase in cage-level productivity. Group-level selection became standard practice in commercial poultry and plant breeding, regardless of theoretical debates about its conceptual status.

The human relevance Clark draws out is direct. Institutional incentive structures that reward individual performance metrics can inadvertently select against traits that benefit the collective - collaborative behavior, mentorship, risk-sharing - even when those traits would make the institution more effective overall.

Limitations

A bibliometric review is only as strong as the quality and representativeness of the studies it synthesizes. Studies that found multilevel selection may be more likely to be published than studies that did not find it. The review covered studies that explicitly measured multilevel selection, which means it may miss important negative evidence from work that assumed individual-level selection without testing higher levels. The political sensitivity of the group selection debate in some corners of evolutionary biology means that publication patterns may not fully reflect the distribution of empirical results.