vol. 160, no. 2 the american naturalist august 2002 A Mechanism for the Evolution of Altruism among Nonkin: Positive Assortment through Environmental Feedback John W. Pepper 1,* and Barbara B. Smuts 2,† 1. Santa Fe Institute, Santa Fe, New Mexico 87501; 2. Department of Psychology, University of Michigan, Ann Arbor, Michigan 48109 Submitted January 10, 2001; Accepted January 18, 2002 abstract: The evolution of altruism often requires genetic simi- larity among interactors. For structured populations in which a social trait affects all group members, this entails positive assortment, meaning that cooperators and noncooperators tend to be segregated into different groups. Several authors have claimed that mechanisms other than common descent can produce positive assortment, but this claim has not been generally accepted. Here, we describe one such mechanism. The process of “environmental feedback” requires only that the cooperative trait affects the quality of the local envi- ronment and that individuals are more likely to leave low-quality than high-quality environments. We illustrate this dynamic using an agent-based spatial model of feeding restraint. Depending on param- eter settings, results included both positive assortment (required for the evolution of altruism) and negative assortment (required for the evolution of spite). The mechanism of environmental feedback ap- pears to be a general one that could play a role in the evolution of many forms of cooperation. Keywords: group composition, inclusive fitness, kin selection, group selection, multilevel selection, cooperation. Cooperation has been a central problem in evolutionary biology for several decades. The paradigmatic example is the evolution of altruistic traits, which impose a fitness cost on the actor while providing a fitness benefit to one or more recipients. One explanation for how altruism can evolve requires that individuals strategically modify their behavior in response to others’ past behavior. This forms the basis for the theory of reciprocal altruism (Trivers 1971) and a large literature on evolutionary game theory * E-mail: jpepper@santafe.edu. † E-mail: bsmuts@umich.edu. Am. Nat. 2002. Vol. 160, pp. 205–213.  2002 by The University of Chicago. 0003-0147/2002/16002-0006$15.00. All rights reserved. (Axelrod and Hamilton 1981; Axelrod 1984). However, most traits of organisms may not involve this type of cog- nitive flexibility, and here, we address the evolution of altruism in fixed traits. In this context, the most common explanation for the evolution of altruism is the theory of inclusive fitness (Hamilton 1964). The alternative frame- work of multilevel selection theory (Price 1972), although different in approach, has proven to be mathematically equivalent (Hamilton 1975; Wade 1980; Queller 1992). Under both the inclusive-fitness and multilevel-selection frameworks, whether altruism evolves depends critically on genetic similarity between actors and recipients. When populations are divided into groups such that social traits affect all group members, this genetic similarity depends on how individuals are distributed among groups with respect to the trait in question. When individuals of the same type are nonrandomly aggregated within the same groups, then individuals are, on average, more similar to other members of their own group than to the population at large, and assortment is said to be positive. When in- dividuals of the same type are nonrandomly dispersed into different groups, then individuals are, on average, less sim- ilar to other members of their own group than to the population at large, and assortment is said to be negative. The central role of assortment can be understood through either the inclusive-fitness or the multilevel- selection framework. In terms of inclusive fitness, positive assortment increases the relatedness coefficient between actors and their recipients (Hamilton 1975; Pepper 2000). In terms of multilevel selection, positive assortment in- creases genetic variance between groups and reduces it within groups, thus strengthening between-group selection at the expense of within-group selection (Price 1972; Ham- ilton 1975; Wilson 1977). Under either framework, as- sortment has a fundamental role in the evolution of co- operation. Traits that benefit only other individuals and not the actor are “strongly altruistic” (i.e., they impose a net fitness cost on the actor; Wilson 1979). Such traits cannot be selected for without positive assortment. In con- trast, traits benefiting a group that includes the actor may be only “weakly altruistic,” meaning that they impose a