New approaches to modeling primate socioecology: Does small female group size BEGET loyal males?

Journal of Human Evolution 137 (2019) 102671

Contents lists available at ScienceDirect

Journal of Human Evolution journal homepage: www.elsevier.com/locate/jhevol

New approaches to modeling primate socioecology: Does small female group size BEGET loyal males? Kristin N. Crouse a, Carrie M. Miller a, Michael L. Wilson a, b, * a b

Department of Anthropology, University of Minnesota, 395 Humphrey Center, 301 19th Ave. S., Minneapolis, MN 55455, USA Department of Ecology, Evolution and Behavior, University of Minnesota, 1987 Upper Buford Circle, Saint Paul, MN 55108, USA

a r t i c l e i n f o

a b s t r a c t

Article history: Received 30 March 2018 Accepted 16 July 2019 Available online xxx

Humans are unusual in having stable male-female breeding bonds within multi-level societies. Such societies are not found in other apes, but have evolved independently in other primates, including several African papionins: hamadryas and Guinea baboons and gelada monkeys. Stable breeding bonds have been proposed to evolve either (1) because males can monopolize females when food distribution forces females to forage in small groups or (2) because females exchange exclusive mating for male services, such as protection from infanticide. Comparative studies are needed to test these hypotheses. In the meantime, we used an agent-based computer model to test the plausibility of these hypotheses. We simulated primates living in social groups within a larger population using a model we call BEGET (Behavior, Ecology, Genetics, Evolution, and Tradeoffs), which employed decision vectors, experimental evolution, realistic trade-offs, and phenotypic plasticity. We employed experimental evolution to generate male genotypes that varied in their competitive ability and in their long-term mating strategy. “Rover” males searched for and mated with any sexually receptive females whereas “Loyalist” males formed stable associations with particular groups of females. Much like living primates, the virtual primates exhibited tradeoffs between contest and scramble competition for access to females: Loyalists evolved larger body size than Rovers. We tested the effect of female foraging group size and the presence of infanticide and infant protection on the relative success of these strategies. We found that Loyalists achieved greater reproductive success than Rovers only when females were in groups smaller than four. Both Rovers and Loyalists sometimes evolved infanticidal behavior, but the presence of infanticide benefited Rovers rather than Loyalists, suggesting that the evolution of stable breeding bonds depends on the spatial distribution of females, rather than the risk of infanticide. © 2019 Elsevier Ltd. All rights reserved.

Keywords: Stable breeding bonds Agent-based modeling Primate socioecology Multi-level societies Evolution of reproductive strategies Virtual evolution

1. Introduction Unlike most other primates, humans (Homo sapiens) live in multi-male, multi-female societies in which males and females commonly reproduce within stable breeding bonds (Rodseth et al., 1991; Chapais, 2010; Walker et al., 2011). The occurrence of stable breeding bonds within a larger society contributes to many distinctive features of human societies. High paternity confidence promotes paternal investment, improving survival of offspring, reducing interbirth intervals, and thus increasing reproductive rates (Kaplan et al., 2000; Jones, 2011). Reproducing within stable breeding bonds increases genetic relatedness among each female's

* Corresponding author. E-mail address: [email protected] (M.L. Wilson). https://doi.org/10.1016/j.jhevol.2019.102671 0047-2484/© 2019 Elsevier Ltd. All rights reserved.

offspring, promoting cooperation among siblings (Chapais, 2010). Stable breeding bonds promote cooperation among both genetic and affinal kin, which in turn promotes faster cultural evolution (Boyd and Richerson, 2009). While humans are unusual in having stable breeding bonds within larger multi-male, multi-female groups, they are not unique; such “multi-level” or “modular” societies have been reported for roughly a dozen primate species, including hamadryas baboons (Papio hamadryas), Guinea baboons (Papio papio), geladas (Theropithecus gelada), and several Asian colobines (Stammbach, 1987; Grueter et al., 2012; Swedell and Plummer, 2012). In these species, individuals aggregate in a nesting set of modular subgroups, from “one male units” (OMUs) that typically contain one breeding male and one or more females and their offspring, to larger aggregations such as clans, troops, bands and herds. Humans differ from these other primates in that human marriages are

2

K.N. Crouse et al. / Journal of Human Evolution 137 (2019) 102671

commonly monogamous, even in societies with polygynous or polyandrous marriages (Marlowe, 2000); a recent analysis of demographic data from small-scale societies found that 61% of siblings were full siblings (Ellsworth et al., 2016). Our ability to understand when, how, and why stable, mainly monogamous breeding bonds emerged in the human lineage is severely constrained by the limited information available from the fossil record. Some have proposed that monogamy emerged directly from a society with chimpanzee-like promiscuous mating, perhaps due to provisioning by males (Lovejoy, 1981; Gavrilets, 2012; Rouly, 2018). It seems likely, however, that fewer concurrent evolutionary changes in temperament and reproductive physiology would be required if monogamy emerged instead among hominins living in multi-level societies, in which stable breeding bonds already existed, but were mainly polygynous (Chapais, 2010). The multi-level societies of other primate species are thus particularly relevant for understanding the evolution of human societies (Grove et al., 2012; Grueter et al., 2012; Swedell and Plummer, 2012; Sutcliffe et al., 2016; Wilson and Glowacki, 2017). The African papionins provide valuable comparative data for understanding the evolution of male-female relations, including stable breeding bonds and multi-level societies (Jolly, 1970, 2001; Grueter et al., 2012; Swedell and Plummer, 2012). Additionally, baboons and geladas live in regions of Africa where hominin fossils are found, in habitats similar to those reconstructed for hominins; fossil papionins are commonly found at hominin sites (Foley, 1993). Baboons and their allies thus evolved in the same habitats as hominins, responding to many of the same ecological pressures, including the emergence of drier, more seasonal habitats dominated by grasses and other C4 plants, with fewer trees available for food and shelter. The stable isotope signatures of many hominins resemble those of baboons, indicating increased reliance on C4 plants for both lineages (Codron et al., 2008; Macho, 2014). Papionins and hominins, therefore, may have evolved modular societies with stable breeding bonds for similar reasons (Jolly, 1970; Grueter et al., 2012; Swedell and Plummer, 2012). Nonetheless, the evolution of multi-level societies remains poorly understood, and may involve different pathways for different lineages. Comparative data on social structure suggest that in Asian colobines, modular societies emerged from the aggregation of OMUs, whereas in papionins and hominins, modular societies resulted from the subdivision of larger societies (Grueter et al., 2012). Current evidence does not permit us to reconstruct the grouping patterns of early hominins with any confidence. Comparison with the two closest living relatives of humans, chimpanzees and bonobos, suggests that the last common ancestor of Homo and Pan lived in multi-male, multi-female groups with fission-fusion dynamics (Ghiglieri, 1987; Wrangham, 1987; Chapais, 2010). A recent phylogenetic analysis, however, suggested that Homo and Pan descend from a species with one-male units, like gorillas (Duda and Zrzavy, 2013). In this case, stable breeding bonds would be ancestral in hominins, rather than derived. Whether stable breeding bonds in hominins are ancestral or derived, however, the existence of such bonds within larger multi-male, multi-female societies requires explanation. In mammals, females tend to specialize in parental effort, freeing males to specialize in mating effort (Trivers, 1972). As such, males devote much of their time and effort to looking for and mating with fertile females. Each male must also compete against other males who are also interested in mating. Males employ various strategies in competition against other males, including guarding fertile females; competing physically for access to females prior to copulation; and competing through post-copulatory means such as sperm competition. Males incur costs when guarding

females, such as reduced feeding (yellow baboons: Alberts et al., 1996; chimpanzees: Georgiev et al., 2014); the risk of aggression from other males (chimpanzees: Muller and Wrangham, 2004; olive baboons: MacCormick et al., 2012), and opportunity costs (given that males can generally guard no more than one female at a time). Males therefore commonly guard females only when they display signs of fertility, such as sexual swellings (chimpanzees: Tutin, 1979; chacma baboons: Seyfarth, 1978). In some species, however, males guard females continuously. Whether males can do so appears to depend mainly on the spatial distribution of females (Wrangham, 1979); when females are widely dispersed, each female may be guarded by a single monogamous male, whereas if females aggregate in small groups, each group may be guarded by a single polygynous male. When females gather in large groups, males cannot effectively monopolize matings with all females, and so multiple males join groups, and males guard only fertile females (Dunbar, 1988; Nunn, 1999). The occurrence of stable breeding bonds within the large, multimale, multi-female societies of humans and some other primates is thus a puzzle. The puzzle has two parts: (1) Why do males form stable bonds with females in some societies, but not others? (2) Why do groups with stable breeding bonds sometimes aggregate in larger societies? In this paper we examine mainly the first part: stable breeding bonds. Many different explanations have been proposed (van Schaik and Dunbar, 1990; Grueter et al., 2012; Swedell and Plummer, 2012; Gomes et al., 2018). Here we focus on two main possibilities: female group size, and infanticide risk. 1.1. Female group size One possibility is that stable breeding bonds depend mainly on female grouping patterns, which in turn depend on factors including feeding ecology (Wrangham, 1980) and predation risk (van Schaik, 1983). For example, Kummer (1995) proposed that in hamadryas baboons, multi-level societies evolved when ancestral baboons colonized semi-deserts with sparse food resources, which forced females to forage in small, monopolizable groups. In most baboon species, individuals are able to travel in cohesive troops, due to feeding specializations that enable them to subsist on a diet of foods such as grass corms and acacia seeds. In northeast Africa and Arabia, however, baboons must instead disperse into smaller subgroups. These small subgroups can be defended by a single male, promoting the evolution of stable breeding bonds. At night, to avoid predation, these units aggregate at refugia such as groves of trees and cliffs. In forest habitats, trees are abundant, and baboons can sleep safely in many areas of their range. In semi-deserts and montane grasslands, however, suitable refugia are scarce. Some refugia, such as large cliffs, can support hundreds of individuals, making them both suitable for large nightly aggregations, and difficult for small groups to monopolize. Whether stable breeding bonds in other species, such as geladas and Guinea baboons, can similarly be explained by feeding ecology remains unclear. The montane grasslands inhabited by geladas appear to have more food than the semi-deserts inhabited by hamadryas, and in some seasons these grasslands support aggregations of hundreds of monkeys. Nonetheless, while the grass and herbs geladas feed upon are abundant, the availability of palatable plant parts varies seasonally (Fashing et al., 2014). When food is scarce, OMUs may forage independently (CMM, personal observation). Additionally, even when foraging in herds, female geladas typically travel in scattered subgroups, which may enable males to monopolize females effectively. We can only speculate about the foraging group size of early hominins. Marlowe (2005) notes that among modern human foragers, women commonly forage in small groups of around five

K.N. Crouse et al. / Journal of Human Evolution 137 (2019) 102671

women, gathering fruits, tubers, and other foods that would have been accessible and edible for early hominins. If females in early hominins foraged in similar group sizes, they may have been defendable by a single male; and if they also gathered at night to sleep on cliffs or in groves of trees, they can reasonably be inferred to have lived in multi-level societies (Wilson and Glowacki, 2017). 1.2. Infanticide risk Infanticide by males poses an important risk in many taxa, and has been proposed to play a major role in the evolution of social behavior and reproductive strategies (van Schaik and Janson, 2000; Grueter et al., 2012; Palombit, 2015). Infanticide risk can explain a number of otherwise puzzling phenomena, such as why female primates in multi-male societies generally mate with multiple males, advertise their fertility with brightly colored sexual swellings and copulation calls, and have ovarian cycles with long luteal phases and unpredictable ovulation (van Schaik and van Noordwijk, 1988). An important function of “friendships” between males and non-ovulatory females appears to be infanticide protection (chacma baboons: Palombit et al., 2000; Moscovice et al., 2010). Infanticide has been proposed to explain stable breeding bonds in monogamous primates (van Schaik and Dunbar, 1990; van Schaik and Kappeler, 1997). More recently, Grueter et al. (2012) argued that infanticide risk can promote the evolution of multi-level societies through the formation of stable breeding bonds, in which females stay with one male in exchange for infanticide protection. Polygynous mating by some males results in a surplus of unmated males, who may band together in bachelor groups. These males attempt to take over females or groups of females, and after doing so, may kill any nursing infants fathered by the deposed male (hamadryas baboons: Amann et al., 2017; geladas: Beehner and Bergman, 2008). While infanticide risk is widespread, and imposes severe fitness costs when it occurs, debate continues over its role in the evolution of social systems. Based on a phylogenetic analysis of mating systems in primates, Opie et al. (2013) argued that infanticide risk best explained the evolution of monogamy. In contrast, Lukas and Clutton-Brock (2013), conducting a similar analysis but including data from across mammals, concluded that monogamy is better explained by the dispersion of females: when females live apart from other females, males may benefit from finding one female and guarding her continuously. Similarly, Lukas and Huchard (2014) argued that while infanticide by males occurs in many mammals, it is a consequence, rather than a cause, of social systems. If males can monopolize matings, they will have high paternity certainty, and thus have strong incentives to protect their offspring from infanticide; and, conversely, they can be confident that infants in other subgroups are not theirs, increasing the likelihood that infanticide will provide males with fitness benefits when taking over a new subgroup. 1.3. Other factors Many additional factors may contribute to the costs and benefits of guarding non-ovulatory females. The extent to which females can be monopolized also depends on the extent to which females overlap in their cycling. A single male can more easily monopolize females when they cycle asynchronously than when many females cycle at the same time, as in seasonal breeders (Nunn, 1999). The duration of interbirth intervals likely also plays a role; when interbirth intervals are extremely long (as in orangutans: 7.2e9.3 years: (Anderson et al., 2008)), males may gain higher fitness by

3

searching for fertile females rather than waiting years for a specific female to become ready to mate again. The stability of breeding bonds thus likely depends on many aspects of primate socioecology, including how life history, group size, demography and population density vary in response to seasonality, distribution of food resources, and the availability of refugia. Even when suitable data are available, attempts to explain how primate social behavior evolves in relation to ecology face many challenges, including the large number of variables involved, the variation among study sites, the many years needed to acquire sufficient data on the life histories of long-lived, slow-growing animals, and the difficulty of quantifying key ecological variables, such as patch size. The complexity of primate socioecology makes causal relationships difficult to infer, particularly when limited to correlational rather than experimental studies. 1.4. Agent-based models Agent-based models provide a useful tool for testing the plausibility of proposed causal relationships. Agent-based modeling is a powerful method that applies computer programming to simulate interactions among a system of autonomous agents (Macal and North, 2010). These models provide three main advantages for advancing our understanding of the evolution of animal behavior. First, they incorporate more variables than are manageable in analytical models. Primate socioecological models have had some success in explaining the observed diversity of primate social organizations in different ecological contexts (Wrangham, 1980; van Schaik, 1983; Isbell, 1991; Sterck et al., 1997; Koenig, 2002). However, these efforts have been met with a range of critiques (Strier, nard, 2004; Snaith and 1994; Kappeler, 1999; Janson, 2000; Me Chapman, 2007; Saj et al., 2007) which have in common one key factor: as the number of variables increases, it becomes more difficult to understand how these variables interact. Each new dimension of information exponentially increases the number of interactions among variables, making it difficult for these models to account for complex interactions and unexpected emergent properties. Often, to reduce this complexity, researchers focus on variables that covary considerably while disregarding others (e.g., Sterck et al., 1997). Computer models can manage numerous variables simultaneously, making even complicated combinations of variables tractable for analysis. Second, agent-based models provide tools to systematically explore the emergent properties of complex systems (Bryson et al., 2007). For example, models have the ability to explore parameter space by systematically varying each combination of parameters and analyzing how the system changes under these variable conditions (i.e., Robbins and Robbins, 2004, 2005; Sellers et al., 2007; Kachel et al., 2011a, 2011b; Kim et al., 2012). Additionally, agents and their state variables are fully catalogued within a model, ensuring that they can be directly measured in a way that cannot be matched by fieldwork, where even the most thorough observers cannot capture everything. Third, agent-based models allow for essentially unlimited tests of virtual populations. Hypothetical situations under any number of environmental contexts or behavioral strategies are possible in a virtual world (i.e., Axelrod and Hamilton, 1981; Bonnell et al., 2013; Mosser et al., 2015). Despite these advantages, findings from agent-based models often have been met with skepticism within the scientific community. Critics have voiced two main concerns. A prevailing assumption is that agent-based models encourage confirmation bias, as it appears that any model can produce any outcome desired (for example, agent-based models have been used to both support (Hawkes et al., 2011; Kim et al., 2014) and refute (Kachel et al., 2011a, 2011b) the grandmother hypothesis). Second, any model

4

K.N. Crouse et al. / Journal of Human Evolution 137 (2019) 102671

represents a simplification of reality, and it can be difficult to know whether the simplifications employed in a given model have unintended consequences that reduce the validity of the model (for example, Robbins and Robbins (2004) initially used census data as input parameters to model gorilla demography, but in the model, these parameters resulted in a mismatch with the long-term demographic data). Here, we present an agent-based model, BEGET (Behavior, Ecology, Genetics, Evolution, and Tradeoffs), that explores the plausibility of evolutionary scenarios. In developing this model, we have sought to avoid the problems outlined above by introducing four novel approaches: (1) decision vectors, (2) experimental evolution, (3) realistic tradeoffs, and (4) phenotypic plasticity. To assess the extent to which our approach is novel, we searched Web of Science and Google Scholar using the search terms “behavioral ecology” with either “an individual-based model” or “an agentbased model,” for all years available. This search yielded 1351 publications, including papers and doctoral dissertations (1969e2018). From this list we narrowed the focus to publications that (i) specifically reported on a new agent-based model and (ii) specifically highlighted questions in behavioral ecology. This yielded 551 publications. Of these, only 142 models explicitly allowed parameters to evolve, and none used these approaches to the standard we describe below. We expect that this set of approaches will be useful for modeling a broad range of socioecological problems. Decision vectors Agent-based computer models typically impose a priori decision rules on agents (i.e., Te Boekhorst and Hogeweg, 1994; Hemelrijk, 1999; Pepper and Smuts, 2000; Robbins and Robbins, 2004, 2005; Evers et al., 2012, 2014; Bonnell et al., 2013; Mosser et al., 2015). While this approach can prove useful, it necessarily imposes the modeler's assumptions into the model. For example, Kachel et al.'s (2011a, 2011b) model testing the grandmother hypothesis imposed rules that were ultimately unrealistic. The model assumed no cost for increased longevity (Hawkes et al., 2011), in contrast to expectations from life history theory: to live longer, organisms must invest more energy in maintenance, reducing energy available for growth or reproduction (Charnov and Berrigan, 1993). In the present study we are interested in understanding male reproductive strategies, and do not want to assume which particular traits are encompassed in these strategies. To this end, our agents possess chromosomes with weighted evolvable preferences for performing certain behaviors within given environmental contexts. Instead of imposing specific strategies for agents to follow in our models, behavioral strategies evolve. This allows more possible combinations of behaviors than in other models. From our literature review, we found that most agent-based models contain fewer than 100 possible behavioral combinations, since most agent decisions depend on a hard-coded hierarchy of if-then statements. In contrast, our model contains over 4,285,540,224 possible behavioral combinations. Some agentbased models have a similar design, with a set of behavioral options for each agent encoded within its chromosomes, which are subject to genetic recombination and mutation (Yaeger, 1994; Holland, 1995; Key and Aiello, 2000). Our model differs from these in that our genes encode probabilistic weights, which can alter the chance of actions occurring, rather than trigger deterministic actions. In this way, our model design resembles that of Rouly (2018) in having weighted genes; however, Rouly's agents are also parameterized with many fixed values, whereas our agents possess no fixed parameters, but instead have values that can evolve over generations. Experimental evolution Agent-based models of social behavior are often parameterized through some combination of informed observation and empirical data from the study system (Robbins

and Robbins, 2004, 2005; Janssen and Ostrom, 2006; Sellers et al., 2007; Kachel et al., 2011a; Bonnell et al., 2013; Janssen and Hill, 2014). However, no model can recreate every aspect of the biological world. It is therefore more conservative to treat models as though they exist in another universe with a different set of physical laws. Often it is sufficient to explore the parameter space of a given model to test parameter settings (Robbins and Robbins, 2004, 2005; Sellers et al., 2007; Kachel et al., 2011a, 2011b; Kim et al., 2012). However, due to runtime demands and the complexity introduced by each separate parameter, this practice is not always feasible. Each parameter increases the number of assumptions imposed upon a model, and a conservative approach suggests that at least some of these assumptions are wrong. Rather than imposing parameters on the model, BEGET contains agents whose evolvable chromosomes contain most of these parameters. Thus, populations of agents evolve the parameter values, rather than having them fixed by initial settings. However, we do set a small number of global parameters, including _globalmutation-rate, _global-sex-ratio, _global-perception-range, as well as a fixed world size of 50
50 cells (Table 1). Realistic tradeoffs Rather than building a model that focuses on a few behaviors thought relevant to testing a hypothesis, in our modeling we assumed that many aspects of behavior could be relevant to the question. This is important because organisms are limited in their resources, which leads to tradeoffs in the allocation of their finite resources towards various investments. By allowing a wider range of possible behaviors, we allow these tradeoffs to evolve in a more realistic manner. These tradeoffs are summarized by the four main categories of life history allocations: growth, maintenance, current reproduction, and future reproduction (Stearns, 1992). To include these aspects of life history into the model, our agents are capable of investing in body growth, maintenance of health, sperm competition, timing of life history stages and fertility cycles, and various behaviors for maleemale competition and mating. Many researchers have recognized the crucial need to be inclusive regarding the life history mechanisms involved in their agent-based models, but have struggled in how to accurately define the relative values of these tradeoffs (Robbins and Robbins, 2004, 2005; Hawkes et al., 2011; Kachel et al., 2011b; Kim et al., 2014; Chan et al., 2017). We improve upon these efforts by imposing a cost to every possible agent action. Because these agents get nothing “for free,” we can be more confident that emergent behavioral traits occur because these evolutionary changes provide net fitness benefits. As discussed above, we avoid imposing specific fixed values for these tradeoffs by incorporating them into each agent's evolvable set of chromosomes. Phenotypic plasticity Often the evolvable traits in agent-based models represent a genotype that is indistinguishable from phenotype; an agent's adult phenotype can be predetermined from its genotype alone (Yaeger, 1994; Holland, 1995; Key and Aiello, 2000; Pepper and Smuts, 2000; Kachel et al., 2011a; Kim et al., 2014; Mosser et al., 2015). This direct mapping of genotype to phenotype limits the potential for genotypes to be differentially expressed, depending on the organism's lifetime interaction with its environment. Our model focuses on male reproductive strategies, such as investment in body size, which in living animals can be directly affected by lifetime experience. Because the adult phenotypes of our agents emerge through developmental processes, an agent's phenotype is the combined result of its genotype, its environment, and its developmental history. For example, agents that obtain more ‘food’ during growth will achieve a larger adult body size than agents with the same genotype but given less food.

K.N. Crouse et al. / Journal of Human Evolution 137 (2019) 102671

5

Table 1 Global parameters of BEGET. Variable Static parametersa _world-energy-rate _initial-number-ofgroups _females-per-group _global-mutationrate _global-sex-ratio _global-perceptionrange _infanticide-on? _female-dispersalon? Dynamic parametersb _world-energy-pool a b

Description The rate at which ‘food’ replenishes in the environment. The initial number of _groups created at the beginning of a simulation. The initial number of adult female _primates created for each group at the beginning of a simulation. An equal number of adult male _primates are also created for each group. The rate of mutation at every genetic locus. The probability that conception leads to a male or female _primate. The radius of vision for each _primate. When turned on, allows males to commit infanticide by attacking infants. When turned on, allows females to disperse from their natal group.

Replenishes at each time step at the _world-energy-rate. _Primates draw energy from this pool when they forage.

Static parameters have their values set during initialization and do not change as the simulation runs. Dynamic parameters may change values during the course of a simulation.

We use these approaches to minimize the number of assumptions we must employ. Despite this goal, we acknowledge that we cannot completely avoid imposing choices and ad hoc settings to our model design; any model must make simplifying assumptions, and these can affect the simulation outcomes in many non-obvious ways. From our literature review of agentbased models in behavioral ecology, we find it important to distinguish between quantitative and qualitative assumptions. Quantitative assumptions, which our model strives to minimize, include choices that fix parameters to particular numerical values. On the other hand, qualitative assumptions, which contribute to the bulk of our model, include some basic features of life history and behavior that are well documented from field studies of primates, such as avoiding mating with close kin. For example, in our model we force all agents to progress from a “juvenile” to “adult” life history stage (qualitative assumption), but allow agent populations to evolve the timing of this transition (rather than imposing a quantitative assumption by setting the transition time at some fixed value). Using these approaches, we simulated primates living in social groups within a larger population, allowing a diversity of male strategies to evolve. The resulting males varied along a continuum in their competitive ability and in their long-term mating strategy. We then tested the relative success of males following different strategies in the contexts of different female group sizes and risk of infanticide by males. We focused on this from the male point of view, assuming that females arrange themselves spatially to optimize factors such as feeding efficiency, predator avoidance, and social relations. 2. Methods 2.1. Model design We developed our agent-based model and ran simulations using NetLogo 6.0.2 (Wilensky, 1999, 2007) on a 2013 iMac computer running OS 10.13.1 and a 3.5 GHz Intel Core i7 processor. We report the methods used following the Overview, Design concepts and Details (ODD) description protocol for agent-based models (Grimm et al., 2006, 2010). We developed an agent-based model of evolving populations of virtual primates (_primates) in three steps. First, we evolved a viable population of _primates. Second, we generated a diversity of male strategies by allowing BEGET to run for multiple generations,

and then identified males that evolved either a “Loyalist” or “Rover” strategy. We defined Loyalists as those that rarely transferred among female groups (as in male primates that form stable breeding bonds with a particular set of females), and Rovers as males that frequently transferred among female groups (as in male primates that seek mating opportunities with any fertile female). Third, we released male _primates representing the extremes of each phenotype into environments of various female group sizes to test the relative success of each strategy. Purpose We developed this model to test hypotheses concerning the evolution of male reproductive strategies, with a focus on how female group size affects whether males attempt to guard females throughout their reproductive cycles, or only when they are at their most fertile. Entities, state variables, and scales The model included two types of entities: _groups and _primates (Supplementary Online Material (SOM) Fig. S1). All _primates are associated with exactly one _group, a spatial association of _primates. _Groups are characterized by _group-color, which facilitates visually distinguishing _primates by their _group membership. _Groups are also distinguished spatially; a _group's spatial position is calculated as the centroid of the locations of its members. _Primates are characterized by several state variables (SOM Table S1). Upon conception, _primates have several static variables that remain fixed for their lives, including _sex, _generation, references to _mother and _father _primates, and _chromosomes that contain the list of genes inherited from these parents. _Primates also possess some dynamic variables pertaining to current status, such as _age, _body-size, _life-history, and _fertility, which change throughout the course of their lives based on environmenteagent interactions. Furthermore, _primates have spatial coordinates that change during the simulation. _Primates store energy in a number of ways and have state variables to this purpose: _primates forage to obtain _stomach-energy, and then transfer this energy to _action-energy to perform actions. _Primates also track the chance of certain events occurring, such as _life-history-chance, _fertility-chance, _mortality-chance, and _conception-chance. Many of these variables, including _body-size, _life-historychance, _fertility-chance, _mortality-chance, and _conceptionchance, encode a value between 0 and 1. These values can be modified throughout a _primates's lifetime. During the “tournament” simulations (All Genotypes Tournament, Top Genotypes Tournament, and Infanticide Tournament; Table 2), we tracked the

6

K.N. Crouse et al. / Journal of Human Evolution 137 (2019) 102671

Table 2 Summary of simulation experiments. Genotype Performancea Genotype filesb # Simulation runsc # Total simulationsd Time stepse _initial-number-of-groups _world-energy-rate _infanticide-on? _female-dispersal-on? Carrying Capacityf Genotype files # Simulation runs # Total simulations Time steps _initial-number-of-groups _world-energy-rate _infanticide-on? _female-dispersal-on? All Genotypes Tournamentg Genotype files # Simulation runs # Total simulations Time steps _initial-number-of-groups _world-energy-rate _infanticide-on? _female-dispersal-on? Top Genotypes Tournamenth Genotype files # Simulation runs # Total simulations Time steps _initial-number-of-groups _world-energy-rate _infanticide-on? _female-dispersal-on? Infanticide Tournamenti Genotype files # Simulation runs # Total simulations Time steps _initial-number-of-groups _world-energy-rate _infanticide-on? _female-dispersal-on?

All 50 male genotype files (M1eM50) 40 simulations per genotype 1520 1000 6 170 true true F2, F18, F41, F48 1 600 1000 3 10e300, at increments of 10 true false 38 viable male genotypes, and F18 for females 5 games per genotype combination 80,000 100 3 Set by Equation (2) true false Top 10 male genotypes (Table 4), with F18 for females 100 games per genotype combination 36,100 100 3 Set by Equation (2) true false Top 10 male genotypes (Table 4), with F18 for females 20 games per genotype combination 8000 1000 3 Set by Equation (2) true, false false

a The Genotype Performance simulations involve recording the natural history of _primate populations seeded from each genotype and simulated in isolation. b Genotype files (M1eM50) used to seed the initial _primate populations. c Number of simulation runs conducted per genotype or combination. d The total number of simulation runs conducted for each experiment. e Number of time steps for each simulation run, data was collected at the end of each run. f The Carrying Capacity simulations involve varying the _world-energy-rate to determine the relationship between this variable and the _primate population size that it can sustain. g The All Genotypes Tournament simulations involve recording the paternity results of _primate males seeded from two distinct genotype files (taken from the pool of 38 viable genotypes) to compare the reproductive success between the two strategies. h The Top Genotypes Tournament simulations are similar to the All Genotypes Tournament but only include the Top 10 Genotypes and include 100 instead of 5 games per genotype pair. i The Infanticide Tournament simulations are similar to the Top Genotypes Tournament but in half of the simulations male _primates are not permitted to commit infanticide.

reproductive success from males with different phenotypes (a “Rover” or “Loyalist”) by their _paternal-identity, which traces all offspring through the male line. Finally, _primates have a few state variables that do not influence their behaviors but instead are used

to facilitate record keeping and data analysis, such as _countconceptions and _count-group-transfers, which track the reproductive success and number of lifetime transfers between _groups, respectively. All _primate state variables are summarized in more detail in SOM Table S1. The _primates inhabit the surface of a torus-shaped world (50
50 square cells). At each time step _primates undergo a series of behavioral processes that are outlined in detail below (Submodels). The BEGET environment currently represents ‘food’ in a non-spatial way, such that _primates can obtain ‘food’ regardless of their location. The model contains a pool of available energy, _world-energy-pool, from which _primates can draw (Table 1). Within a given time step, this energy pool is finite, and it replenishes each time step. Each _primate who takes energy from the available pool reduces the amount of energy available, such that the entire energy pool might be depleted by the time some _primates have their turn. Thus, _world-energy-rate sets the carrying capacity for each simulation. Process overview and scheduling At each time step the _worldenergy-pool is set to the _world-energy-rate parameter setting, which dictates the rate at which the world gains more energy or ‘food’ for _primates to ‘eat.’ Otherwise, the model is driven by several _primate processes, activated at each timestep, which are discussed in more detail below (Submodels). Collectively, these _primate processes correspond to realworld processes that have been observed or inferred by primatologists. These processes are executed in the following order: (I) ‘energy’ and ‘metabolism’ are updated, (II) the environment is evaluated, (III) decisions are made, (IV) actions are executed, (V) movement vectors are calculated and (VI) current status is updated (Fig. 1). Below, we refer to Ego as the hypothetical _primate currently executing these processes and Target as a hypothetical _primate that Ego can ‘see’ and thus incorporate into its decision making. Basic principles This agent-based model combines principles of behavioral ecology (Trivers, 1972, 1974; Hrdy, 1979; Wrangham, 1980; van Schaik, 1983; Terborgh and Janson, 1986; Pusey, 1987; Isbell, 1991; Stearns, 1992; Andersson, 1994; Alberts and Altmann, 1995; Mitani et al., 1996; Sterck et al., 1997; van Schaik and Kappeler, 1997; Nunn, 1999; Koenig, 2002) with concepts from virtual evolution (von Neumann, 1966; Dawkins, 1986; Ray, 1991; Yaeger, 1994; Holland, 1995) to allow _primates flexibility in behavior, but also a finite range of possible evolved strategies. From this, we evolved a series of “Loyalist” and “Rover” _primate phenotypes (see Selection experiments and Simulations sections), corresponding to high and low levels of group fidelity, respectively, and used these phenotypes to test under which group size context they achieved greater reproductive success. Emergence Strategies emerge from a given ecological context over the course of generations of selection. The ecological context depends mainly on three parameter settings: _initial-number-ofgroups dictates how many _primate _groups are initially created; _females-per-group dictates how many adult females (and males) are initially created for each _group; and _worldenergy-rate dictates the rate at which ‘food’ replenishes in the environment (Table 1). Additionally, the ecological context is shaped by the phenotypes of the _primates themselves, which may alter individual preferences for group composition. Strategies include the timing of life history events like gestation and weaning; the amount of energy allocated to body growth and maintenance; amount of time spent foraging; and others. Furthermore, _primates move spatially in response to their environment and based on their genotypes, which include weighted preferences for moving relative to other primates. In

K.N. Crouse et al. / Journal of Human Evolution 137 (2019) 102671

general, _primates tend to move toward group members and away from non-group members. With multiple conspecifics present, and indeed no spatially located food resources to influence movement, what emerges from this is a random-walk effect at the _group level. Therefore, _group spatial segregation is not guaranteed. Yet, because most _primates tend to avoid non-_group conspecifics, _group segregation usually occurs. Thus, population level spatial dynamics are not imposed by the model but emerge from individual interactions and movement preferences. Because the world wraps vertically and horizontally, the distance between the centers of any two groups ranged from 0 to about 35 cells and thus a median distance of about 17.5 cells. Within a time step, primates may move within cells, or travel to an adjacent cell. Adaptation We represent inheritance using a set of diploid ‘chromosomes’ that together form a _primate's ‘genotype.’ Each ‘behavior’ chromosome consists of a set of ‘genes’ and each gene contains (1) an environmental context (SOM Table S2), (2) a weighted preference for performing an action (SOM Table S3), and (3) a Boolean value that, when set to ‘true’, enables that gene's weight to mutate. If Ego ‘sees’ an environmental context that matches one of its genes, it performs that gene's corresponding action and invests an amount of energy to that action based on that gene's corresponding weight (see Submodels for more detail). Weights vary between 0 and 1 and variants in the weights of genes constitute ‘alleles.’ Thus, each gene can have an indefinite number of possible alleles, each with its own value from 0 to 1. Since chromosomes are diploid, each _primate

7

makes weighted decisions by averaging the weights from both chromosomes. _Primates reproduce sexually and so prior to conception, chromosomes undergo the genetic processes of recombination and mutation. During recombination, BEGET randomly selects one of two chromosomes from each _primate parent. Then, roughly 50% of alleles from the selected chromosome are randomly exchanged with alleles from the homologous chromosome. Finally, each parent provides this chromosome to produce a new homologous pair of chromosomes in the offspring. After this process, mutation at each locus occurs by chance based on the _global-mutation-rate, which is set to 10% for all simulations. This is high compared to real world organisms but promotes rapid diversification of alleles within the time scale of the simulation. Equation 1. This equation takes as input y0, a real number between 0 and 1, and x, any real number from negative to positive infinity. Output y is a real number between 0 and 1 such that positive values of x result in y > y0 and negative values of x result in y < y0.

8 < yð1þjxjÞ 0 yðxÞ ¼ : y1=ð1þxÞ 0

if x < 0 if x  0

(1)

During mutation, modifications to gene weights are calculated using Equation (1), which can incrementally increase or decrease the original gene weight while remaining within the 0 to 1 range:

Figure 1. A summary of the _primate processes that are executed at each time step. Shown here is a hypothetical scenario for Ego (E): (I) Ego updates its action-energy, (II) Ego evaluates its environment, (III) Ego makes decisions, (IV) Ego executes all actions in its queue, (V) Ego moves by (a) summing all weighted decision vectors for movement and (b) moving in the calculated direction, (VI) Ego updates its status and potentially dies.

8

K.N. Crouse et al. / Journal of Human Evolution 137 (2019) 102671

y0 is the weight's initial value, x is a randomly generated number between 1 and 1, and y is the weight's final value. Thus as a population of _primates evolves, behavioral traits change over time due to (1) variation in alleles resulting from mutation; (2) new combinations of alleles resulting from recombination and sexual reproduction; and (3) changes to the frequencies of alleles resulting from selection and drift. Additionally, _primates have ‘identity’ chromosomes, consisting of a list of ten letters, which also undergo recombination and mutation, and are used to identify related individuals. If an Ego _primate sees a Target individual with _identity-chromosomes that match its own _identity-chromosomes at greater than or equal to 25%, Ego identifies Target as ‘kin.’ Sensing _Primates can only sense their immediate surroundings within the _global-perception-range, a global parameter that dictates the radius of vision for each _primate. We set the _global-perception-range to 10 cells and kept this value constant throughout each simulation. To reduce the amount of time and computing power needed for each simulation, we constrained _primates so that at each time step, they can only perceive about 10% of their immediate surroundings. The number of conspecifics that Ego perceives is determined by taking the total number of _primates within the _global-perception-range radius of Ego, multiplying by 10%, and rounding up to the nearest whole number. This number of _primates is then randomly selected from those within Ego's _global-perception-range to be considered during Ego's decision-making process. We consider this constraint realistic, as it simulates the limited attention that real animals can devote to different features of their surroundings. It also introduces more stochasticity, which further aides in modeling reality. All information that Ego can ‘see’ by perceiving Target _primates is encoded in Target's ‘status’ (SOM Table S2). The status of Target includes factors such as its sex, life history stage, and fertility state. This status also includes factors determined using information from both individuals, such as the relative size of Target with respect to Ego, the relative health of Target with respect to Ego, whether Target is genetically similar to Ego, and whether Target belongs to the same _group as Ego. For example, the status “oafprdqn” represents an adult (“a”) female (“f”) who is currently pregnant (“p”), smaller in _body-size (“r”) and higher in _mortalitychance (“d”) than Ego, not genetically similar to Ego (“q”), and part of the same _group as Ego (“n”). Ego is also capable of perceiving itself, and so a final status factor separates whether Target and Ego are the same individual (“e”) or different individuals (“o”). See Table 4 for a detailed summary of all possible aspects of a _primate's status. While _primates are not able to recognize specific individuals, the model does provide a limited form of recognition key for our study. Males do have _group recognition: they can recognize to which group they belong, and to which groups they don't belong. When a group contains a single female, males thus effectively recognize that female. Additionally, they can recognize if a group contains their kin, and how many females are in a group. Interaction _Primates can engage in any number of interactions per time step, depending on their current _action-energy available to use, surrounding environment, and their own genotype. _Primates directly interact with other _primates in the following ways: fight, mate, groom and join another individual's _group (SOM Table S3). These interactions can only occur if both Ego and Target occupy the same cell. Some of these interactions may alter _primate state variables: the _mortality-chance variable is increased for the loser of a fight, and decreased for the receiver of a grooming event. Mating may cause conception to occur, which

results in the creation of a new _primate agent. _Primates can also decide to “join” another _primate's group and in effect transfer to that group. When this occurs, their _group state variable updates to their new _group designation. As discussed above (Sensing), group members can ‘see’ who is a part of their group and who is not. _Primates also interact with themselves to change their own state variables (SOM Table S1): _primates can invest energy into their _life-history-chance state variable, which affects the probability of transitioning to the next life history stage (for example, from “juvenile” to “adult”); maintenance decreases _mortalitychance; growth increases _body-size; conception investment increases _conception-chance; _primates can increase their _maternal-stores, the energy stores used by their dependent offspring; and investing in fertility increases _fertility-chance, or the chance of transitioning to the next fertility stage (for example, from “pregnant” to “lactating”). These state variable modifications are calculated from Equation (1): y0 is the variable's initial value, x is the amount of invested energy, y is the variable's final value, and the sign of x is negative if the effect is to decrease the state variable and positive otherwise. _Primates can also “leave” their own _group and travel alone until they join a new group. When this occurs, BEGET creates a new _group agent whose only member is the _primate leaving its current _group, and that _primate updates its group state variable to refer to this newly made _group. _Primates also indirectly interact through feeding competition by investing in foraging. When _primates invest in foraging, they are able to take a proportionate amount of energy from the _worldenergy-pool to increase their _stomach-stores. However, energy is taken from _world-energy-pool on a first-come-first-served basis (i.e., scramble competition) and there is a finite amount of energy available at each time step. If _primates attempt to forage when no energy is available they will not receive energy. _Primates can engage in any number of interactions per time step, depending on their current _action-energy available to use, surrounding environment, and their own genotype. Stochasticity As mentioned above, each _primate views its environment in a stochastic way by only ‘seeing’ about 10% of its surroundings. Additionally, the genetic processes of recombination and mutation are stochastic. Several _primate status updates are also based on chance; at each time step, a _primate checks whether to update itself based on: _life-history-chance, _fertility-chance, _mortality-chance, and _conception-chance (SOM Table S1). For example, a _primate's _mortality-chance is its chance of dying each time step. The model calculates the chance of events occurring by (1) generating a number between 0 and 1, (2) comparing that number to the chance state variable, and (3) activating the associated event if the number falls below the variable's value. For example, if _mortality-chance is activated, then the agent dies. Some _primate actions are also stochastic (SOM Table S3). For example, whether _primates join or leave _groups is based on weighted allele probabilities. The likelihood of winning a fight during an attack is directly proportional to the ratio of the opponents' _body-size: _primates with a larger _body-size have a proportionally greater chance of winning the fight, following Lanchester's linear law of battle (Lanchester, 1916; Wilson et al., 2002). Collectives Each _primate belongs to a _group. Often, multiple _primates belong to the same group and spatially aggregate based on group membership. We designed _group membership to approximate ways that group-living primates such as baboons interact within and among groups. Typically, group members cluster towards each other and avoid non-group members. We modeled these _groups as part of the same multi-level society, in

K.N. Crouse et al. / Journal of Human Evolution 137 (2019) 102671

which group members can often see and sometimes interact with non-group members. Observation At the end of each “tournament” simulation (Table 2), BEGET outputs a text file of population-level data pertaining to the relative success of the initially input genotype files (see Initialization). We calculated the reproductive success of different _primate genotypes using their _paternal-identity, which is one of either of the two input genotypes. The output data include the number of offspring produced by each identity. Additionally, the median, mean, variance and maximum values of reproductive success were collected for both genotypes. Reproductive success is measured by the mean number of _count-conceptions obtained by each _paternal-identity. Initialization Upon initialization, BEGET populates the world with _groups of _primates. The number of _groups is determined by the parameter _initial-number-of-groups. BEGET populates each _group with a number of adult females equal to the _femalesper-group value, and an equal number of adult males. However, in the tournament simulations, when _females-per-group is 1, BEGET populates each _group with 2 adult males in order to maintain an even number of males during the “tournament” simulations. Groups are randomly assigned a _group-color and are placed at a random starting location. Each _primate receives a genotype file, chosen from a list of available genotype files, which is used to populate its _chromosomes. Multiple genotype files can be uploaded to a single population. Initialized _primate males are given the _paternal-identity that corresponds to their given genotype file. Once the simulation is running, new _primates produced during sexual reproduction are given the same _paternal-identity as their _father. See SOM Table S2 for a full summary of the state variables given to each _primate upon initialization. Input data At the start of each simulation, we selected a series of genotype files used to seed the initial population of _primates. Typically, this consisted of one genotype file for all female _primates, and one or two genotype files for all male _primates. These genotype files contain all information necessary to create a population of ‘clones’ based on the individual or individuals originally used to create each genotype file. Submodels Each time step, BEGET activates each _primate in random order. Each _primate then undertakes a series of processes (Fig. 1). Once these processes have been completed, BEGET continues to the next randomly chosen _primate until all individuals have had their turn. (I) Update Energy. _Primates have two pathways to gaining usable _action-energy, depending on their current _life-history status. If Ego has a _life-history status of “gestatee,” “infant,” or “juvenile,” it transfers any available energy from the _maternal-energy of its _mother, if she is alive, to its own _stomach-energy. While the model allows juveniles to continue to take energy from their mother, it is likely that the mother no longer invests in her _maternal-energy at this stage since she is now cycling and thus ready to conceive and invest in a new offspring. The age at which mothers end investment in _maternal-energy for each offspring is a life history trait that can evolve in this model. If Ego has a _life-history status of “juvenile,” “adult,” or “senescent,” it forages to obtain energy, which it stores in its _stomach-energy. Then, Ego updates its _action-energy, which is the difference between its _stomachenergy and its basal metabolic energy cost, which is calculated based on the formula: BMC ¼ _body-size (0.762) (Nagy, 1994). (II) Evaluate Environment. Ego evaluates its environment, which includes individuals within its _global-perception-range,

(III)

(IV)

(V)

(VI)

9

including itself. As noted above, during each time step, Ego can focus its attention only on 10% of individuals within this observable range. Ego identifies the status of all Targets being considered and, in conjunction with its own known status, uses them to create a list of Ego-Target status combinations (see Sensing). Make Decisions. From the previous step, Ego has gathered a list of Ego-Target status combinations based on its current environment. We can consider each combination a ‘key’ that can be used to ‘unlock’ one of Ego's genes. As described above (see Adaptation), each gene contains an environmental context, which includes a hypothetical Ego-Target status combination. We can consider this environmental context the ‘lock’ that can be unlocked from a matching ‘key.’ During this process step, Ego checks all of its keys against each of its genes' locks. A key matches a lock if each element in the lock can be found in the key. If Ego finds a matching gene, then it puts the gene's corresponding action and weight into a queue. The weights in the queue are further modified to account for the distance from the target. In analogy with physical forces such as gravity and electromagnetism, these weighted values follow an inverse-square rule, decreasing in value with the squared distance to the target. Execute Actions. Ego executes all actions in its queue, allocating a portion of its _action-energy as calculated from the associated weights for each action. Multiple actions can be executed within the same time step. Many actions result in state variable changes (see SOM Table S3). Move. Some actions are movement-based: “Mate,” “Move Toward,” “Join Group,” “Fight,” and “Groom” trigger Ego to move toward Target, and “Move Away” and “Leave Group” trigger Ego to move away from Target. Ego calculates the direction to move based on the summation of all decision vectors based on these actions. Once this calculation is made, Ego moves in the calculated direction, the distance moved is based upon _body-size and energy from _action-energy allocated for movement. Status. Finally, Ego changes its status based upon a series of calculations. First, Ego increases _age and _mortality-chance by one unit, and decreases _conception-chance by one unit, simulating the normal aging process. Second, based upon the current values of its _life-history-chance, _fertility-chance, and _mortality-chance, Ego has a chance to update those respective status factors (see Stochasticity). If _life-history-chance is activated, Ego upgrades its _life-history to the next available stage (“gestatee” ¼> “infant,” “infant” ¼> “juvenile,” “juvenile” ¼> “adult,” “adult” ¼> “senescent,” or “senescent” ¼> “dead”). If _fertility-chance is activated, Ego upgrades its _fertility to the next available stage (“pregnant” ¼> “lactating,” or “lactating” ¼> “cycling”). If _mortality-chance is activated, or _life-history reaches “dead,” it is removed from the simulation.

2.2. Selection experiments To test our hypotheses, we evolved a set of male genotypes that varied in average number of lifetime _group transfers. To do this, we used an experimental evolution approach, similar to work done with microbes, in which iterative application of simple selection rules has led to traits such as multicellularity (Ratcliff et al., 2012). We developed genotypes from three separate and incremental phases of selection: (1) selection for individuals who were best able to survive and reproduce, (2) selection for individuals who met desirable criteria like fast life histories, large body size, and high

10

K.N. Crouse et al. / Journal of Human Evolution 137 (2019) 102671

group fidelity, (3) selection for individual males who best reproduced. We detail each phase of the selection process below: Survival selection We initially gave _primates a genotype, “Initial Simple Genotype,” with what seemed to us reasonably selected values for each allele (SOM Table S4). To reduce the likelihood of creating a non-viable genotype, this Initial Simple Genotype contained a limited number of genes, including genes responsible for regulating basic life history and reproductive tradeoffs only, and lacked any ability for _primates to fight with each other or commit infanticide. The allele values were chosen through some initial experimentation to see which best ensured the survival of the initial _primate population. Our initial guesses turned out to be poor ones, because the simulated populations repeatedly crashed within a few generations. We therefore conducted an iterated series of selection experiments to identify genotypes that led to viable populations. We ran these experiments as follows: We ran the simulation (_initial-number-of-groups: 5, _femalesper-group: 10, _world-energy-rate: 190) until the population began to decline rapidly. When the population size reached half of the initial population size, we selected the individual with the highest _generation number, indicating the maximum number of generations that had passed since the simulation began. If that generation number was equal to or higher than the generation number from the previous selection experiment, we saved this individual's genotype and used it to seed a new population of _primates. Otherwise, we ran the simulation again using the same genotype as before. In this way, we selected for populations of _primates that survived for incrementally more generations before crashing. After four iterations of this selection process, a genotype emerged that had not crashed by 30,000 time steps, and nine generations, long enough to suggest that the population could survive indefinitely. Life history selection Once we developed a genotype that resulted in viable populations, we selected for several criteria desirable for the purpose of our experiments. We did so by culling individuals who did not meet a series of requirements. First, we required _primates to remain within five cells, or half the _globalperception-range, from their _group center, thus ensuring a high level of group spatial fidelity. Second, we required _primates to maintain an adult _body-size between 0.85 and 0.95. This requirement was necessary because we found that without it, _primates gradually became smaller in body size, likely because their genotypes did not yet include fighting and thus there was no incentive for maintaining a large enough _body-size to gain a fighting advantage. Third, to increase efficiency on our modest operating system we required _primates to have shorter life histories, thereby increasing the rate of evolution and decreasing simulation runtime. Initially, _primates taken from Survival Selection lived over 6000 time steps, so every 10,000 time steps we decreased the maximum allowed lifespan, and culled any _primates that exceeded this maximum allowed lifespan, until _primates evolved a maximum longevity of 2000 time steps. We ran simulations (_initial-number-of-groups: 5, _females-pergroup: 10, _world-energy-rate: 190) using the same iterated procedure as outlined in Survival Selection; if a population crashed, we seeded a new population using the genotype that achieved the greatest number of generations from previous simulations. At this stage in the selection process, the Initial Simple Genotype did not include any “join” or “leave” actions. Thus, _primate males were not yet able to evolve preferences to either stay within or leave groups. These actions were added to the Full Genotype during the Male Diversification Selection process (detailed below). Once we developed a population of _primates that sustained each of these requirements for 10,000

time steps, we selected the individual with the highest _generation and saved its genotype, which became the “Final Simple Genotype,” the base genotype from which we evolved all male strategy genotypes in the next selection phase (SOM Table S5). In total, the Final Simple Genotype emerged from thirteen iterations of selection: four from Survival Selection and nine from Life History Selection. See SOM Tables S4 and S5 to compare Initial Simple Genotype and Final Simple Genotype values. We found that by the final round of Life History Selection, at every locus, the initial value had been replaced by a new allele at least once during mutation, resulting in a set of Final Simple Genotype alleles that are entirely different from the Initial Simple Genotype alleles. Male diversification selection During this final selection process, we evolved males in a series of simulations to generate a diversity of strategies. _Primate populations in these simulations were initialized with Final Simple Genotype, which included genes for the timing of life history, and allocation of energy for growth and maintenance, and general strategies for finding mates, including joining and leaving _groups. During this selection process, we turned off the ability for genes from the Final Simple Genotype to mutate to minimize changes to basic life history pathways during the testing phase of the project. We then added additional genes specific to male reproductive strategies (SOM Table S7), including: sex-specific investment in _body-size and _conception-chance, preferences for males fighting other males, attacking infants, and transferring between groups. Additionally, we introduced genes for female transfer between groups to allow the evolution of a wider range of possible social and spatial configurations. We changed the genes that encode mating to allow _primates to mate only with non-related individuals. We turned on the ability for these added genes to mutate, allowing for a variety of male reproductive strategies to evolve. This new genotype, “Initial Full Genotype,” is shown in SOM Table S7 and was used as the initial genotype for this selection process. Since we ultimately required a range of genotypes that produced various levels of lifetime group transfers, we initially devised a series of selection experiments that specifically selected for high and low levels of group transfers. However, this process introduced various anomalies, such as males with high group transfer rates but low levels of reproductive success. We therefore instead conducted selection experiments that selected for individual males with the highest number of conceptions. We ran simulations (_initialnumber-of-groups: 6, _females-per-group: 6, _world-energy-rate: 170) and every 1000 time steps, we selected the male with the highest _count-conceptions and used his genotype to seed a new population. We repeated this artificial selection process twenty times to generate a final male genotype to use for experimentation. We generated the final male genotype file by selecting an arbitrary adult male from the final simulation and recording his genotype information. We repeated this entire process fifty times to generate fifty distinct male genotype files (M1eM50). These genotype files were used in the “tournament” simulations. In addition to a list of alleles, each genotype file included phenotype information on _body-size, _conception-chance and _mortality-chance to be used during initialization. Thus, initialized populations seeded from these genotype files were essentially clones of the original adult from which the genotype was taken. However, once the simulation was running, subsequent generations of _primates did not reference the genotype file for their phenotypic traits, but instead developed them from lifetime experiences. Additionally, due to sexual reproduction, recombination, and mutation, novel combinations of alleles emerged each generation. We also randomly chose a corresponding female genotype file from the same population used to collect each male genotype file (F1eF50).

K.N. Crouse et al. / Journal of Human Evolution 137 (2019) 102671

2.3. Simulations Genotype performance The process described above resulted in 50 male genotypes that produced a diversity of strategies. From this pool, we chose some exemplar “Rover” and “Loyalist” genotypes to test the relative success of Rovers and Loyalists under different female group size contexts. We ran a series of initial simulations from these fifty genotype files to understand better their particular traits and classify them by _group transfer preference. These simulations gathered information on population size, female _group composition, male _group transfer preference, male reproductive strategies like _body-size and _conception-chance investment, age of _life-history and _fertility transitions, and infanticide rates. Infanticide rates were calculated by dividing the number of infants that died after an attack by a male by the total number of infants that existed during the simulation. For each genotype, we ran 40 simulations for 1000 time steps (Table 2). From these results, we identified genotypes as inviable if they resulted in a mean population size below 100 _primates, roughly 150% of the initial population size, and genotypes as viable if they exceeded 100 _primates. Twelve genotypes (M3, M4, M5, M7, M10, M12, M13, M17, M23, M27, M28, M32) fell below the viable threshold. The remaining 38 genotypes collectively had a mean (±SD) population size of 148 ± 26.3 _primates and all further reported results include only these viable genotypes. From the 38 viable genotypes, we classified each genotype as either Rover or Loyalist based on whether they fell into the top 50th percentile or bottom 50th percentile for mean lifetime _group transfers. Since genes for mating required that Ego and Target be unrelated, each male _primate transferred _groups at least once before reproducing. Mean lifetime _group transfers ranged in a continuum from 1.3 ± 0.12 transfers to 7.8 ± 1.2 transfers across genotypes (SOM Fig. S2). We found that a range of female group composition and spatial dynamics had emerged during the selection process. For the Tournament Simulations, we controlled for such socioecological factors by preventing female dispersal, maintaining a constant _group size by running short simulations, and keeping the number of _groups constant. Using the random forest method, we investigated if Rovers and Loyalists evolved differences in _body-size and _conceptionchance, which represent reproductive tradeoffs in contest and scramble competition, respectively. The random forests method builds a predictive algorithm from a set of regression trees (in this paper n ¼ 500) used to predict output values from a set of input variables. We built a random forest model using input measures of mean _body-size and mean _conception-chance and output categorical variable for assigned male reproductive strategy (Rover or Loyalist) from the Genotype Performance data. We randomly partitioned the Genotype Performance dataset into 50% training data (n ¼ 760), which was exclusively used to build the model, and 50% test data (n ¼ 759), which was exclusively used to evaluate the model's predictive performance. The Genotype Performance results contain a series of life history values. During each simulation, we tracked each _primate for the following information: _age of transition from each _life-history and (for females) _fertility stage. Using these data, we collected median values for gestation length (22.7 time steps), lactation length (31.0 time steps), age at first reproduction (110 time steps), interbirth interval (113 time steps) and maximum longevity (1450 time steps) (N ¼ (38 viable genotypes) (40 simulations) ¼ 1520 simulations). Figure 2 summarizes these results, showing the life history of _primates compared to empirical data from real baboon species.

11

Carrying capacity During the Genotype Performance simulations, we selected values for _world-energy-rate based on experimentation for what values would best sustain a population of _primates. For subsequent data collection, we opted to investigate a systematic way of determining _world-energy-rate from _initial-number-of-groups and _females-per-group. We conducted a second round of simulations to investigate the minimum required _world-energy-rate to maintain the starting _primate population size, using four arbitrary female genotype files F2, F18, F41, and F48 (Table 2). For each setting, we recorded the ten lowest values of _world-energy-rate that sustained the initial population. These data revealed a linear relationship between _females-per-group and _world-energy-rate for constant _initial-number-of-groups (SOM Fig. S3). Equation (2) summarizes this relationship in terms of the dependent variable _world-energy-rate and the independent variables _initialnumber-of-groups and _female-group-size. This equation sets the carrying capacity; it calculates how much food is needed to maintain the initial population size. All subsequent simulations use this equation to set _world-energy-rate. Equation 2. This equation sets parameter value _world-energyrate (E) based on the parameter settings for _females-per-group (F) and _initial-number-of-groups (G). This linear relationship was calculated from the data collected in the Carrying Capacity simulations.

E ¼ 10 þ 5GF

(2)

All genotypes tournament The Genotype Performance results are based on populations seeded by a single male genotype. Ultimately, we are interested in the relative success of strategies when competing against other strategies. Thus, we conducted a third round of simulations, which consisted of a series of repeated “games” between each pairwise combination of genotypes (Table 2). During these games, males competed (through contest and scramble competition) for access to fertile females. To reduce runtime, we set _initial-number-of-groups to 3. Since we expected female group size to affect genotype performance, we ran simulations with various _females-per-group (2, 4, 6, 8, and 10 females) and set _world-energy-rate using Equation (2). In each simulation, we seeded the initial _primate population with genotype F18 for females, and two of the 38 viable male genotypes for males. We selected genotype F18 because it fell at nearly the 50th percentile for mean male lifetime _group transfers, and thus was less likely to bias the behavior of offspring towards either Rover or Loyalist strategies. We ran each simulation for 100 time steps, or about one interbirth interval. Each combination of male genotypes played five total games at every _female-group-size setting, for a total of 25 games per pairwise combination at the end of the tournament. At the end of each simulation, we identified a “winner” based on which genotype sired the most offspring, using each _primate's _lineage-identity to track this. From these games, we identified the top performers based on their proportion of wins from the total number of games played, and selected the five best Loyalists and five best Rovers (Table 4). These ten top genotypes still suggested two distinct categories distinguished by mean male _group transfers (Table 3). Furthermore, we constructed a random forest model using input variables mean body-size and mean conception-chance, output categorical variable for male reproductive strategy (Rover or Loyalist), and partitioned with 50% training (n ¼ 200) and 50% test data (n ¼ 200).

12

K.N. Crouse et al. / Journal of Human Evolution 137 (2019) 102671

Figure 2. Life history values of _primates (taken from Genotype Performance results) compared to living Papio species (Jones et al., 2009; Isler and van Schaik, 2012). Living Papio values are in days and _primate values are in unit time steps, both shown here as a percentage of the maximum longevity.

Top genotypes tournament Our fourth round of simulations tested the performance of the top Loyalist and Rover genotypes in a range of female group sizes. These simulations were similar to the All Genotypes Tournament, except that now 100 games were played by each pairwise combination of genotypes (Table 4). For each simulation, we tracked the reproductive success of both genotypes by counting the number of conceptions for each. Infanticide tournament Our final round of simulations tested whether infanticide by males helped the performance of Loyalists or Rovers across a subset of female group sizes (Table 2). In half of the simulations, we turned off the ability for males to commit infanticide. We allowed these simulations to run for 1000 ticks, longer than the previous tournament simulations, to allow enough time for infanticidal events to occur in the simulation. We expected that these longer simulations would result in a wider range of final female group configurations. Having observed that the number of groups influences male strategies, we sought to control for this by excluding runs in which the final number of female groups was less than 2 or greater than 4 (we set _initial-number-of-groups to 3). We conducted all statistical analyses in R (version 3.4.3) for oneway ANOVAs, GLMMs using the package ‘lme4,’ and random forest models using the package ‘randomForest.’ We considered genotype files as random effects. 3. Results Here we report results from the three steps of our work. In the first step, the BEGET model yielded viable populations of _primates through an iterative process of experimental evolution. In the second step (Male Diversification Selection), BEGET yielded various male and female reproductive strategies. In the third step, we conducted several rounds of “tournaments” among a subset of Rovers and Loyalists selected from the second step.

3.1. Genotype performance In the worlds seeded by clones of 38 different genotypes (in which both male and female behavior emerged from the same genetic background), Rovers transferred more often between female groups than Loyalists (Table 4). Female grouping patterns also differed; worlds seeded by a given Rover genotype ended up with more female _groups than worlds seeded by a given Loyalist genotype (Table 4). The mean number of male lifetime _group transfers correlated positively with the mean number of female _groups present (GLMM: ß ± SE ¼ 0.0287 ± 0.00570, z ¼ 5.04, P < 0.001, N ¼ 38 genotypes, 40 runs per genotype; Fig. 3a). That is, when more female _groups were available, males transferred more often. Loyalists and Rovers differed significantly in the number of female _groups present at the end of these simulations (Table 4). In these simulations, males in the populations seeded by different genotypes evolved a range of values for mean _body-size (median ¼ 0.934; range ¼ 0.732e0.998), mean _conception-chance (median ¼ 0.915, range ¼ 0.345e0.991), and mean number of lifetime _group transfers (median ¼ 2.43; range ¼ 1e11; N ¼ 38 genotypes). Rovers generally had smaller body size than Loyalists, apart from one unusual Loyalist genotype, M34; (Table 3). On the whole, Rovers did not differ from Loyalists in their _conceptionchance (Table 3). Both _body-size and _conception-chance varied along a continuum, with considerable variation among genotypes for both Rovers and Loyalists (Fig. 3). These phenotypic differences affected the variance in reproductive success among the different clonal populations. Larger mean _body-size correlated positively with variance in reproductive success (GLMM: ß ± SE ¼ 2.25 ± 0.413, z ¼ 5.45, P < 0.001, N ¼ 38 genotypes, 40 runs per genotype). In most populations with large _body-size (and presumably more contest competition), winners and losers in reproductive competition thus faced more extreme outcomes. Correspondingly, greater investment in mean _conception-chance correlated negatively with variance in

K.N. Crouse et al. / Journal of Human Evolution 137 (2019) 102671

13

Table 3 Summary of Rover and Loyalist results from Genotype Performance simulations. All Genotypesa (N ¼ 38) Loyalists (mean ± SD) d

Infanticide rate Number of female _groupse Mean lifetime _group transfersf Variance in reproductive successg Mean _body-sizeh Median _body-size Maximum _body-size Mean _conception-chance Median _conception-chance Maximum _conception-chance

Rovers (mean ± SD)

ANOVAb

Cohen's d c

0.0393 ± 0.076 8.86 ± 2.84 1.94 ± 0.478 14.7 ± 13.0 0.926 ± 0.0539 0.953 ± 0.0504 0.99 ± 0.0146 0.887 ± 0.105 0.902 ± 0.155 0.992 ± 0.00915

0.0327 ± 0.0648 17.49 ± 7.55 3.63 ± 1.58 12.6 ± 14.7 0.892 ± 0.0705 0.913 ± 0.00513 0.984 ± 0.019 0.907 ± 0.0409 0.927 ± 0.0597 0.993 ± 0.00424

0.0926 1.51 1.45 0.154 0.548 0.601 0.373 0.252 0.218 0.138

NS F1,36 ¼ 26.2 F1,36 ¼ 25.6 NS F1,36 ¼ 2.88 F1,35 ¼ 4.22 F1,35 ¼ 3.48 NS NS NS

Loyalists (mean ± SD)

Rovers (mean ± SD)

Cohen's d

ANOVA

0.104 ± 0.105 8.35 ± 3.11 1.87 ± 0.522 22.0 ± 11.3 0.948 ± 0.0501 0.970 ± 0.0418 0.995 ± 0.00757 0.889 ± 0.0437 0.891 ± 0.0653 0.986 ± 0.0123

0.0371 ± 0.0553 18.8 ± 8.6 4.08 ± 2.05 14.8 ± 8.4 0.906 ± 0.0512 0.940 ± 0.00513 0.99 ± 0.0107 0.908 ± 0.0366 0.942 ± 0.0147 0.922 ± 0.00335

0.790 1.62 1.48 0.729 0.829 0.689 0.471 0.475 1.09 0.712

NS F1,8 ¼ 6.11 F1,8 ¼ 5.00 F1,8 ¼ 3.59 NS NS NS NS F1,8 ¼ 3.60 NS

P < 0.001 P < 0.001 P < 0.1 P < 0.05j P < 0.1j

Top Genotypesi (N ¼ 10)

Infanticide rate Number of female _groups Mean lifetime _group transfers Variance in reproductive success Mean _body-size Median _body-size Maximum _body-size Mean _conception-chance Median _conception-chance Maximum _conception-chance

P < 0.05 P < 0.1 P < 0.1

P < 0.1

a All Genotypes include populations seeded by the genotype files: M1, M2, M6, M8, M9, M11, M14, M15, M16, M18, M19, M20, M21, M22, M24, M25, M26, M29, M30, M31, M33, M34, M35, M36, M37, M38, M39, M40, M41, M42, M43, M44, M45, M46, M47, M48, M49, M50. b ANOVA results comparing Loyalists and Rovers. c Not significant. d Infanticide rate is calculated as the number of lethally attacked infant _primates (an infant is attacked and subsequently dies during the same timestep) divided by all infants born. e Number of female _groups is determined by counting the number of _groups with adult female _primates present at the end of each simulation. f The mean lifetime _group transfers is calculated by taking the average of all adult male _primate _count-group-transfers at the end of each simulation. g The variance in reproductive success is calculated by taking the variance of all adult male _primate _count-conceptions at the end of each simulation. h Mean, median and maximum _body-size and _conception-chance are calculated from the pool all adult male _primates at the end of each simulation. i Top Genotypes include populations seeded by the genotype files: M40, M42, M19, M46, M22 (Loyalists), M9, M48, M47, M8, and M31 (Rovers). j This calculation was taken with M34 removed and is not significant (NS) with M34 included.

Table 4 Summary of Individual Top Rover and Loyalist results from Genotype Performance simulations. Top Five Loyalist Genotypes (N ¼ 40 simulations per genotype file) Genotype file name Lifetime _group transfersa Infanticide rateb Maximum _body-sizec Maximum _conception-chanced Competitive ratioe

M40 1.3 0.16 0.998 0.974 1.024

± ± ± ± ±

0.12 0.15 1.1E-3 1.4E-2 1.2E-2

M42 1.6 0.072 0.998 0.985 1.013

± ± ± ± ±

0.34 0.064 1.5E-3 1.3E-2 1.4E-2

M19 1.8 0.053 0.998 0.984 1.014

± ± ± ± ±

0.28 0.045 1.3E-3 1.1E-2 1.1E-2

M46 2.3 0.068 0.999 0.994 1.006

± ± ± ± ±

0.36 0.067 4.0E-4 3.3E-3 3.3E-3

M22 2.4 0 0.981 0.994 0.987

± ± ± ± ±

0.36 0.0034 6.0E-3 2.8E-3 5.6E-3

Top Five Loyalists 1.9 0.10 0.995 0.986 1.01

± ± ± ± ±

0.52 0.11 7.5E-3 1.2E-2 1.6E-2

Top Five Rover Genotypes (N ¼ 40 simulations per genotype file) Genotype file name Lifetime _group transfers Infanticide rate Maximum _body-size Maximum _conception-chance Competitive ratio

M9 2.6 0.13 0.985 0.993 0.992

± ± ± ± ±

0.34 0.072 4.4E-3 3.2E-3 4.7E-3

M48 2.7 0.074 0.997 0.993 1.004

± ± ± ± ±

0.31 0.072 1.4E-3 3.1E-3 2.9E-3

M47 3.2 0.032 0.999 0.993 1.006

± ± ± ± ±

0.54 0.039 8.9E-4 3.8E-3 3.8E-3

M8 4.2 0.044 0.998 0.991 1.008

± ± ± ± ±

0.56 0.025 1.3E-3 3.3E-3 3.2E-3

M31 7.8 0.0016 0.974 0.993 0.980

± ± ± ± ±

1.2 0.0020 8.1E-3 2.9E-3 6.9E-3

Top Five Rovers 4.8 0.037 0.990 0.992 0.998

± ± ± ± ±

2.1 0.055 1.1E-2 3.4E-3 1.1E-2

a

The number of lifetime transfers among _groups (mean ± SD) for male _primates seeded from Top Rover and Loyalist genotype files. Infanticide rate is calculated as the number of lethally attacked infant _primates (an infant is attacked and subsequently dies during the same timestep) divided by all infants born. c The maximum _body-size (mean ± SD) is calculated from the pool of male _primates with the largest _body-size taken from each of the 40 simulations per genotype file. d The maximum _conception-chance (mean ± SD) is calculated from the pool of male _primates with the highest _conception-chance taken from each of the 40 simulations per genotype file. e The Competitive Ratio is calculated by taking the ratio of maximum _body-size to maximum _conception-chance (mean ± SD) taken from each of the 40 simulations per genotype file. b

reproductive success (GLMM: ß ± SE: 5.13 ± 0.372, z ¼ 13.8, P < 0.001, N ¼ 38 genotypes, 40 runs per genotype). This outcome supports the view that trade-offs exist in investment between contest and scramble competition for access to females.

To test the validity of Rover and Loyalist categories, we used a random forests model to classify male phenotypes by reproductive strategy, using data from the Genotype Performance simulations and the input variables mean _body-size and mean _conception-

14

K.N. Crouse et al. / Journal of Human Evolution 137 (2019) 102671

chance. In 78.7% of cases, the random forests model classification agreed with our earlier classification based on the mean number of lifetime _group transfers. When considering only top performing genotypes (Table 4), the random forest model classification agreed with our earlier classification in 86.0% of cases. These results support the view that these categories are meaningful rather than arbitrary. 3.2. Top genotypes tournament As we predicted, Loyalists performed best with small group sizes (1e4 females per _group) and Rovers performed best in larger _groups (10e14 females; Fig. 4). Loyalists generally achieved higher reproductive success than Rovers in small groups, while Rovers generally achieved higher reproductive success than Loyalists in large groups (N ¼ 100 runs per genotype combination, 25 genotype combinations; mean paternity percentage: ANOVA: F1,22541 ¼ 790, P < 0.001; Cohen's d ¼ 0.373). The relative reproductive success of Loyalists versus Rovers depended strongly on the number of females in each group (Fig. 5). In general, males following either strategy obtained more conceptions in groups with more females (and thus more mating opportunities; particularly since, when female group size ¼ 1, we included 2 males per group to ensure an even number of Loyalists and Rovers in the world). When competing against Rovers, Loyalists achieved slightly higher reproductive success when in groups with 1 or 2 females (Fig. 5a). However, in groups of 6 or more females, Rovers out-competed Loyalists (Fig. 5a). Strikingly in large group sizes, Rovers achieved even greater reproductive success when competing against Loyalists than when competing against other Rovers. In single strategy tournaments (Rovers vs. Rovers and Loyalists vs. Loyalists), both strategies achieved about the same

maximum reproductive success across female _group sizes (Fig. 5b). In contrast, in two-strategy tournaments (Rovers vs. Loyalists), the presence of males following a different strategy resulted in lower maximum reproductive success, although Rovers still performed better than Loyalists in groups with at least 6 females (Fig. 5b). Correspondingly, Rovers had a higher variance in reproductive success than in Loyalists in these larger groups. We expected that the Loyalist and Rover genotypes with the most extreme traits (in terms of mean lifetime _group transfers, _body-size, and _conception-chance) would be the most effective competitors, but this was not the case. For example, Genotype M40 appeared to be an archetypal Loyalist, possessing the lowest of all mean lifetime _group transfers and highest ratio of maximum _body-size to maximum _conception-chance (Table 4), but males with this genotype performed poorly against Rovers across all female _group sizes. Conversely, M22 performed best across nearly all female _group sizes, only exceeded by M47 in group sizes 8 to 14, but exhibited more Rover-like traits (higher mean lifetime _group transfers and lower ratio of maximum _body-size to maximum _conception-chance, compared to other Loyalists; Table 4). Likewise, M47, the most successful Rover strategy, exhibited more Loyalist-like traits (lower mean lifetime _group transfers and higher ratio of maximum _body-size to maximum _conception-chance, compared to other Rovers; Table 4). Thus, it appeared that the best strategies for both Loyalists and Rovers were those that hedged their bets rather than evolving extreme values for any trait. 3.3. Infanticide tournament In simulated games with Rovers and Loyalists competing against each other, Loyalists performed worse in worlds with infanticide

Figure 3. Genotype Performance results: we classified genotypes as Rovers (open circle) or Loyalists (closed circle) based on whether they fell into the top 50th percentile or bottom 50th percentile for mean lifetime _group transfers; (a) number of mean lifetime group transfers of males versus the number of female groups, (b) maximum male _body-size (a virtual analog of contest competition) versus variance in male reproductive success, (c) maximum male _conception-chance (a virtual analog of sperm competition) versus variance in male reproductive success, and (d) the tradeoffs between maximum male _body-size and maximum male _conception-chance.

K.N. Crouse et al. / Journal of Human Evolution 137 (2019) 102671

15

Figure 4. Top Genotypes Tournament results for paternity percentage: summary of Loyalist and Rover paternity percentage across different female group sizes. Rovers (dashed light) and Loyalists (solid dark) are shown as averages (thick lines) and by each genotype (thin lines).

than without infanticide (Fig. 6). Following our results above, as expected for small female _group sizes (_females-per-group ¼ 1, 2, 4, 6), when infanticide was not present Loyalists performed better (975 wins) than Rovers (443 wins: Fig. 6a). However, when infanticide was present (Fig. 6b), Loyalists still performed better (830 wins) compared to Rovers (471 wins), but this number of Loyalist wins was significantly lower than when infanticide was not present (N ¼ 20 runs per genotype combination, 25 genotype combinations; Loyalist mean paternity percentage: with infanticide: X ± SD ¼ 0.59 ± 0.28, without infanticide: X ± SD ¼ 0.63 ± 0.28; ANOVA: F1,1723 ¼ 6.88, P < 0.01; Cohen's d ¼ 0.048). 4. Discussion The BEGET model yielded viable populations of agents that developed and behaved in ways analogous to real primates. In a relatively small number of generations, a diversity of male reproductive strategies evolved.

4.1. Genotype performance Male _primates faced realistic tradeoffs: they were not able to simultaneously maximize both fighting ability (_body-size) and sperm competition (_conception-chance) but instead allocated their limited resources based on their particular evolved reproductive strategy. We found that male strategies evolved along a continuum. To identify different male strategies, we adopted an admittedly arbitrary cut-off: we categorized males that transferred more often than the median “Rovers,” and males that transferred less often than the median “Loyalists.” Loyalists invested more in _body-size, which is the main factor affecting fighting ability in this model. Rovers and Loyalists did not differ, overall, in their investment in _conception-chance, but the overall negative correlation between _body-size and _conception-chance indicates a tradeoff between contest and scramble competition, analogous to that seen in real primates, such as howler monkeys, in which species with larger hyoid bones (and thus more intimidating loud calls) have smaller testes (Dunn et al., 2015). Despite these general trends, a

Figure 5. Top Genotypes Tournament results for reproductive success: mean and maximum reproductive success of Loyalists and Rovers across different female group sizes, data from Top Genotypes Tournament simulations. Results are separated by male strategy games played against males following the same strategy (Loyalists vs. Loyalists (dark solid) and Rovers vs. Rovers (dark dashed)) and games played against males following the opposite strategy (Loyalists vs. Rovers (light solid) and Rovers vs. Loyalists (light dashed)). Loyalists vs. Rovers refers to the mean reproductive success of Loyalists in games played against Rovers. Furthermore, Rovers vs. Loyalists refers to the mean reproductive success of Loyalists in games played against Rovers (results from the same games as the previous scenario, but with data presented for Loyalists).

diversity of strategies emerged, as exemplified by genotype M34, which had a small _body-size despite being a Loyalist. 4.2. Top genotypes tournament We predicted that Loyalists would be most successful in smaller, more monopolizable female groups. During the simulation process, we controlled for various socioecological factors (namely, we kept constant the number of female groups and the size of these groups over time) and varied female group size. We found that when females were in small groups of 1 or 2, Loyalists achieved higher reproductive success than Rovers, whereas in larger female groups of 6e14, Rovers achieved higher reproductive success. Under these conditions, we infer that Loyalist males increased their reproductive success by staying in one group and monopolizing access to fertile females through contest competition. However, males following all strategies achieved greater reproductive success when they had access to more females, but males that transferred frequently among groups experienced greater gains in this context than males that transferred less frequently.

16

K.N. Crouse et al. / Journal of Human Evolution 137 (2019) 102671

Figure 6. Infanticide Tournament results: summary of Loyalist (dark) and Rover (light) paternity percentage across different female group sizes, separated by games when infanticide is allowed (solid) and when it is not allowed (striped). For all but female group size of 2, the differences were statistically significant (N ¼ 20 runs per pairwise genotype combination; 25 genotype combinations: paternity percentage: ANOVA: female group size ¼ 1: F1,287 ¼ 4.75, P < 0.05; female group size ¼ 2: F1,458 ¼ 0.063, P ¼ 0.80; female group size ¼ 4: F1,956 ¼ 5.011, P < 0.05; female group size ¼ 6: F1,974 ¼ 5.13, P < 0.05). An asterisk indicates a statistically significant result (P < 0.05).

4.3. Infanticide tournament We predicted that Loyalists would achieve greater success in games when infanticide was present. Contrary to expectations, when infanticide risk was present, Loyalists experienced a decreased, rather than increased, reproductive advantage compared to Rovers. Researchers have proposed multiple hypotheses for the evolution of stable breeding bonds in hominins (exchange of food resources: Lovejoy, 1981; exchange for infanticide protection: Swedell and Plummer, 2012; Grueter et al., 2012; response to changing ecological pressures: Grueter et al., 2012). Our results support the hypothesis that stable breeding bonds evolve in conditions in which females aggregate in small monopolizable subgroups, much as Lukas and Clutton-Brock (2013) argue that dispersed females lead to the evolution of monogamy in mammals. Loyalists achieved greater reproductive success than Rovers up to around four females per group, whereas Rovers achieved higher reproductive success in all larger groups. Thus, it is plausible that hominins similarly evolved stable breeding bonds in response to changes in ecology that favored female dispersal into small, defendable foraging parties. While infanticide risk likely affects a broad range of traits, including reproductive physiology and anatomy (van Schaik and van Noordwijk, 1988), in BEGET, the presence of infanticide risk reduced the reproductive advantage Loyalists had over Rovers in small groups. Furthermore, while provisioning by males clearly plays an important role in modern human societies, our results indicate that provisioning is not necessary for the evolution of stable associations between males and females (and this is consistent with the presence of such stable reproductive bonds in many primates, such as gibbons and hamadryas baboons, in which males do not provision mates or offspring (Fernandez-Duque et al., 2009)). In BEGET, agents evolved stable associations between males and females in response to female grouping patterns, without any provisioning by males. This is consistent with the gradual, stepwise evolution of distinctive hominin traits suggested by Grueter et al.

(2012). Given the numerous differences between the biological world and BEGET's virtual world, we do not attempt to make precise claims about female foraging group size in hominins. Nonetheless, our results support the view that stable breeding bonds emerged due to females foraging in small subgroups, rather than in response to infanticide risk. 4.4. Future directions We expect the approaches exemplified in the BEGET model will be useful for addressing a broad range of questions in future studies. A strength of this approach is that the BEGET model incorporates a large number of variables. This contrasts with a prevailing practice in agent-based modeling, which has been to limit the number of variables to those directly related to the question. While any useful model involves simplifications from reality, the specific features of living species result from interactions among many different factors. We found that the use of many different variables within the BEGET model proved tractable with available computing power, and provided useful insights beyond what a more limited approach might have provided. We expect BEGET to be useful for future studies for three main reasons. First, it serves as a generalized hypothesis testing tool. We designed BEGET not only to investigate the specific focus of this paper, but also to tackle a broad range of questions in the evolution of the behavior and ecology of primates, including the human lineage. This paper presents the first in a series of planned tests using the approach described here. Second, BEGET encompasses considerable genetic flexibility. Each gene encodes an “if-then” statement that guides agent decisions in response to features of its environment, other agents, and its own status. The features used in the current model would allow for at least 4,285,540,224 possible genes, each of which can yield an indefinite number of alleles. We have barely begun exploring the behavior space of this system. For this paper, we included features that we believed would be important for testing male reproductive strategies. Additional features could be added to investigate other traits. Third, BEGET is designed such that additional _primate status and action features can be added

K.N. Crouse et al. / Journal of Human Evolution 137 (2019) 102671

without disrupting current code. Future hypothesis testing can include new features while still maintaining the general structure of the model. In the future, we plan to implement the following additions to BEGET. First, we plan to introduce spatially explicit ‘food’ locations such that _primate success at foraging depends on the energy available at their current location. Second, we will add additional evolvable _primate abilities. For example, BEGET currently has some global parameters (_global-perception-range, _global-sexratio, and _global-mutation-rate) that could instead be handled by individual genotypes. Each _primate will have state variables that correspond to these global parameters such that we no longer need to define them globally. Third, we plan to allow all components of genes (i.e., Ego-Target combinations and actions), rather than only the weighted values, to mutate. For this paper, we explicitly chose each gene combination. However, it is likely that we did not include an exhaustive set of genes that best correspond to male reproductive strategies. When we allow the full gene to evolve, we expect that novel combinations will emerge that result in even more accurate primate behaviors. Fourth, we will give _primates the ability to remember events. We recognize that the experience that an individual acquires in its lifetime might influence its behavior in profound ways. For example, males cannot currently track the identities of the females with whom they mate. However, this kind of tracking is likely important in understanding a broad range of primate behaviors, including infanticidal attacks. Finally, we will introduce the ability of agents to eat ‘meat,’ which will be modeled as energy transferred from dead to living individuals. We expect that these agents will evolve attack behaviors that model predation. In this way, we can evolve predator agents to investigate how predation affects social organization. In BEGET, we sought to incorporate multiple features to more accurately portray the tradeoffs that real organisms face. Our results indicate that we were successful in producing a first step towards this goal. From a continuum of evolved _primates, we identified genotypes as “Loyalist” or “Rover” based on low or high levels of lifetime _group transfers. Loyalists evolved larger body size than Rovers. Across these genotypes, a negative correlation existed between _body-size and _conception-chance, indicating that tradeoffs existed between contest and scramble competition. Loyalists achieved greater reproductive success than Rovers only when females were in groups smaller than four. Both Rovers and Loyalists sometimes evolved infanticidal behavior, but the presence or absence of infanticide did not affect the relative success of these strategies, suggesting that the evolution of stable breeding bonds depends on the spatial distribution of females, rather than the risk of infanticide. Acknowledgments We thank Julia Fischer and Dietmar Zinner for the invitation to participate in the Frontiers in Baboon Research Symposium and this special issue. Our understanding of multi-level primate societies was enriched by spending time with gelada monkeys and hamadryas baboons at Guassa and Filoha, Ethiopia. We thank Peter Fashing, Nga Nguyen, and Larissa Swedell for hosting us at their study sites. Nisarg Desai provided valuable help with statistical methods. This work was supported by the Leakey Foundation (CMM) and the University of Minnesota, including a Talle Faculty Award (MLW) and an African Studies Initiative Travel Grant (MLW). Supplementary Online Material Supplementary online material related to this article can be found at https://doi.org/10.1016/j.jhevol.2019.102671.

17

References Anderson, H.B., Thompson, M.E., Knott, C.D., Perlins, L., 2008. Fertility and mortality patterns of captive Bornean and Sumatran orangutans: Is there a species difference in life history? Journal of Human Evolution 54, 34e42. Andersson, M.B., 1994. Sexual Selection. Princeton University Press. Alberts, S.C., Altmann, J., 1995. Balancing costs and opportunities: dispersal in male baboons. American Naturalist 279e306. Alberts, S.C., Altmann, J., Wilson, M.L., 1996. Mate guarding constrains foraging activity of male baboons. Animal Behaviour 51, 1269e1277. Amann, A.L., Pines, M., Swedell, L., 2017. Contexts and consequences of takeovers in hamadryas baboons: female parity, reproductive state, and observational evidence of pregnancy loss. American Journal of Primatology 79, 8. Axelrod, R., Hamilton, W.D., 1981. The evolution of cooperation. Science 21, 1390e1396. Beehner, J.C., Bergman, T.J., 2008. Infant mortality following male takeovers in wild geladas. American Journal of Primatology 70, 1152e1159. Bonnell, T.R., Campennì, M., Chapman, C.A., Gogarten, J.F., Reyna-Hurtado, R.A., Teichroeb, J.A., Wasserman, M.D., Sengupta, R., 2013. Emergent group level navigation: an agent-based evaluation of movement patterns in a folivorous primate. PLoS One 8, e78264. Boyd, R., Richerson, P.J., 2009. Culture and the evolution of human cooperation. Philosophical Transactions of the Royal Society B: Biological Sciences 364, 3281e3288. Bryson, J.J., Ando, Y., Lehmann, H., 2007. Agent-based modelling as scientific method: a case study analysing primate social behaviour. Philosophical Transactions of the Royal Society of London B: Biological Sciences 362, 1685e1699. Chan, M.H., Hawkes, K., Kim, P.S., 2017. Modelling the evolution of traits in a twosex population, with an application to grandmothering. Bulletin of Mathematical Biology 79, 2132e2148. Chapais, B., 2010. Primeval Kinship: How Pair-Bonding Gave Birth to Human Society. Harvard University Press, Cambridge, MA. Charnov, E.L., Berrigan, D., 1993. Why do female primates have such long lifespans and so few babies? or life in the slow lane. Evolutionary Anthropology 1, 191e194. Codron, D., Lee-Thorp, J.A., Sponheimer, M., de Ruiter, D., Codron, J., 2008. What insights can baboon feeding ecology provide for early hominin niche differentiation? International Journal of Primatology 29, 757e772. Dawkins, R., 1986. The Blind Watchmaker: Why the Evidence of Evolution Reveals a Universe Without Design. WW Norton & Company. Duda, P., Zrzavy, J., 2013. Evolution of life history and behavior in Hominidae: Towards phylogenetic reconstruction of the chimpanzee-human last common ancestor. Journal of Human Evolution 65, 424e446. Dunbar, R., 1988. Primate Social Systems. Cornell University Press, Ithaca, NY. Dunn, J.C., Halenar, L.B., Davies, T.G., Cristobal-Azkarate, J., Reby, D., Sykes, D., Dengg, S., Tecumseh Fitch, W., Knapp, L.A., 2015. Evolutionary trade-off between vocal tract and testes dimensions in howler monkeys. Current Biology 25, 2839e2844. Ellsworth, R.M., Shenk, M.K., Bailey, D.H., Walker, R.S., 2016. Comparative study of reproductive skew and pair-bond stability using genealogies from 80 smallscale human societies. American Journal of Human Biology 28, 335e342. Evers, E., de Vries, H., Spruijt, B.M., Sterck, E.H., 2012. Look before you leapindividual variation in social vigilance shapes socio-spatial group properties in an agent-based model. Behavioral Ecology and Sociobiology 66, 931e945. Evers, E., de Vries, H., Spruijt, B.M., Sterck, E.H., 2014. The EMO-model: an agentbased model of primate social behavior regulated by two emotional dimensions, anxiety-FEAR and satisfaction-LIKE. PLoS One 9, e87955. Fashing, P.J., Nguyen, N., Venkataraman, V.V., Kerby, J.T., 2014. Gelada feeding ecology in an intact ecosystem at Guassa, Ethiopia: Variability over time and implications for theropith and hominin dietary evolution. American Journal of Physical Anthropology 155, 1e16. Fernandez-Duque, E., Valeggia, C.R., Mendoza, S.P., 2009. The biology of paternal care in human and nonhuman primates. Annual Review of Anthropology 38(1), 115e130. Foley, R.A., 1993. African terrestrial primates: the comparative evolutionary biology of Theropithecus and the Hominidae. In: Jablonski, N.G. (Ed.), Theropithecus: The Rise and Fall of a Primate Genus. Cambridge University Press, Cambridge, pp. 245e270. Gavrilets, S., 2012. Human origins and the transition from promiscuity to pairbonding. Proceedings of the National Academy of Sciences of the United States of America 109, 9923e9928. Georgiev, A.V., Russell, A.F., Thompson, M.E., Otali, E., Muller, M.N., Wrangham, R.W., 2014. The foraging costs of mating effort in male chimpanzees (Pan troglodytes schweinfurthii). International Journal of Primatology 35, 725e745. Ghiglieri, M.P., 1987. Sociobiology of the great apes and the hominid ancestor. Journal of Human Evolution 16, 319e357. Gomes, B.V., Guimaraes, D.M., Szczupak, D., Neves, K., 2018. Female dispersion and sex ratios interact in the evolution of mating behavior: a computational model. Scientific Reports 8, 8. Grimm, V., Berger, U., Bastiansen, F., Eliassen, S., Ginot, V., Giske, J., Goss-Custard, J., Grand, T., Heinz, S.K., Huse, G., Huth, A., Jepsen, J.U., Jørgensen, C., Mooij, W.M., Müller, B., Pe'er, G., Piou, C., Railsback, S.F., Robbins, A.M., Robbins, M.M., Rossmanith, E., Rüger, N., Strand, E., Souissi, S., Stillman, R.A., Vabøg, R., Visser, U., DeAngelis, D.L., 2006. A standard protocol for describing individualbased and agent-based models. Ecological Modelling 198, 115e126.

18

K.N. Crouse et al. / Journal of Human Evolution 137 (2019) 102671

Grimm, V., Berger, U., DeAngelis, D.L., Polhill, J.G., Giske, J., Railsback, S.F., 2010. The ODD protocol: a review and first update. Ecological Modelling 221, 2760e2768. Grove, M., Pearce, E., Dunbar, R.I.M., 2012. Fission-fusion and the evolution of hominin social systems. Journal of Human Evolution 62, 191e200. Grueter, C.C., Chapais, B., Zinner, D., 2012. Evolution of multilevel social systems in nonhuman primates and humans. International Journal of Primatology 33, 1002e1037. Hawkes, K., Kim, P.S., Kennedy, B., Bohlender, R., Hawks, J., 2011. A reappraisal of grandmothering and natural selection. Proceedings of the Royal Society of London B: Biological Sciences 278, 1936e1938. Hemelrijk, C.K., 1999. An individualeorientated model of the emergence of despotic and egalitarian societies. Proceedings of the Royal Society of London B: Biological Sciences 266, 361e369. Holland, J.H., 1995. Hidden Order: How Adaptation Builds Complexity. AddisonWesley/Helix Books, Reading, MA. Hrdy, S.B., 1979. Infanticide among animals: a review, classification, and examination of the implications for the reproductive strategies of females. Ethology and Sociobiology 1, 13e40. Isbell, L.A., 1991. Contest and scramble competition: patterns of female aggression and ranging behavior among primates. Behavioral Ecology 2, 143e155. Isler, K., van Schaik, C.P., 2012. Allomaternal care, life history and brain size evolution in mammals. Journal of Human Evolution 63, 52e63. Janson, C.H., 2000. Primate socio-ecology: The end of a golden age. Evolutionary Anthropology 9, 73e86. Janssen, M.A., Hill, K., 2014. Benefits of grouping and cooperative hunting among Ache hunteregatherers: insights from an agent-based foraging model. Human Ecology 42, 823e835. Janssen, M.A., Ostrom, E., 2006. Empirically based, agent-based models. Ecology and Society 11, 37. Jolly, C.J., 1970. The seed-eaters: a new model of hominid differentiation based on a baboon analogy. Man 5, 5e26. Jolly, C.J., 2001. A proper study for mankind: Analogies from the papionin monkeys and their implications for human evolution. Yearbook of Physical Anthropology S44, 177e204. Jones, J.H., 2011. Primates and the evolution of long, slow life histories. Current Biology 21, R708eR717. Jones, K.E., Bielby, J., Cardillo, M., Fritz, S.A., O'Dell, J., Orme, C.D.L., Safi, K., Sechrest, W., Boakes, E.H., Carbone, C., Connolly, C., Cutts, M.J., Foster, J.K., Grenyer, R., Habib, M., Plaster, C.A., Price, S.A., Rigby, E.A., Rist, J., Teacher, A., Bininda-Emonds, O.R.P., Gittleman, J.L., Mace, G.M., Purvis, A., 2009. PanTHERIA: a species-level database of life history, ecology, and geography of extant and recently extinct mammals: Ecological Archives E090-184. Ecology 90(9), 2648. Kachel, A.F., Premo, L.S., Hublin, J.J., 2011a. Grandmothering and natural selection. Proceedings of the Royal Society of London B: Biological Sciences 278, 384e391. Kachel, A.F., Premo, L.S., Hublin, J.J., 2011b. Grandmothering and natural selection revisited reply. Proceedings of the Royal Society of London B: Biological Sciences 278, 1939e1941. Kaplan, H., Hill, K., Lancaster, J., Hurtado, M.A., 2000. A theory of human life history evolution: diet, intelligence, and longevity. Evolutionary Anthropology 9, 156e185. Kappeler, P., 1999. Convergence and divergence in primate social systems. In: Fleagle, J.G., Janson, C., Reed, K.E. (Eds.), Primate Communities. Cambridge University Press, Cambridge, pp. 158e170. Key, C., Aiello, L.C., 2000. A prisoner's dilemma model of the evolution of paternal care. Folia Primatologica 71, 77e92. Kim, P.S., Coxworth, J.E., Hawkes, K., 2012. Increased longevity evolves from grandmothering. Proceedings of the Royal Society of London B: Biological Sciences 279(1749), 4880e4884. Kim, P.S., McQueen, J.S., Coxworth, J.E., Hawkes, K., 2014. Grandmothering drives the evolution of longevity in a probabilistic model. Journal of Theoretical Biology 353, 84e94. Koenig, A., 2002. Competition for resources and its behavioral consequences among female primates. International Journal of Primatology 23, 759e783. Kummer, H., 1995. Quest of the Sacred Baboon: A Scientist's Journey. Princeton University Press, Princeton, NJ. Lanchester, F.W., 1916. Aircraft in Warfare: The Dawn of the Fourth Arm. Constable Limited. Lovejoy, C.O., 1981. The origin of man. Science 211(4480), 341e350. Lukas, D., Clutton-Brock, T.H., 2013. The evolution of social monogamy in mammals. Science 341, 526e530. Lukas, D., Huchard, E., 2014. The evolution of infanticide by males in mammalian societies. Science 346, 841e844. Macal, C.M., North, M.J., 2010. Tutorial on agent-based modelling and simulation. Journal of Simulation 4, 151e162. MacCormick, H.A., MacNulty, D.R., Bosacker, A.L., Lehman, C., Bailey, A., Anthony Collins, D., Packer, C., 2012. Male and female aggression: lessons from sex, rank, age, and injury in olive baboons. Behavioral Ecology 23, 684e691. Macho, G.A., 2014. Baboon feeding ecology informs the dietary niche of Paranthropus boisei. PLoS One 9, e84942. Marlowe, F.W., 2000. Paternal investment and the human mating system. Behavioural Processes 51, 45e61. Marlowe, F.W., 2005. Hunter-gatherers and human evolution. Evolutionary Anthropology 14, 54e67. nard, N., 2004. Do ecological factors explain variation in social organization? In: Me Thierry, B., Singh, M., Kaumanns, W. (Eds.), Macaque Societies. Cambridge University Press, Cambridge, pp. 237e262.

Mitani, J.C., Gros-Louis, J., Richards, A.F., 1996. Sexual dimorphism, the operational sex ratio, and the intensity of male competition in polygynous primates. American Naturalist 147, 966e980. Moscovice, L.R., Di Fiore, A., Crockford, C., Kitchen, D.M., Wittig, R., Seyfarth, R.M., Cheney, D.L., 2010. Hedging their bets? Male and female chacma baboons form friendships based on likelihood of paternity. Animal Behaviour 79, 1007e1015. Mosser, A.A., Kosmala, M., Packer, C., 2015. Landscape heterogeneity and behavioral traits drive the evolution of lion group territoriality. Behavioral Ecology 26, 1051e1059. Muller, M.N., Wrangham, R.W., 2004. Dominance, aggression and testosterone in wild chimpanzees: a test of the ‘challenge hypothesis’. Animal Behaviour 67, 113e123. Nagy, K.A., 1994. Field bioenergetics of mammals: What determines field metabolic rate? Australian Journal of Zoology 42, 43e53. Nunn, C.L., 1999. The evolution of exaggerated sexual swellings in primates and the graded-signal hypothesis. Animal Behaviour 58, 229e246. Opie, C., Atkinson, Q.D., Dunbar, R.I.M., Shultz, S., 2013. Male infanticide leads to social monogamy in primates. Proceedings of the National Academy of Sciences of the United States of America 110, 13328e13332. Palombit, R.A., 2015. Infanticide as sexual conflict: coevolution of male strategies and female counterstrategies. Cold Spring Harbor Perspectives in Biology 7, 29. Palombit, R.A., Cheney, D.L., Fischer, J., Johnson, S., Rendall, D., Seyfarth, R.M., Silk, J.B., 2000. Male infanticide and defense of infants in chacma baboons. In: van Schaik, C.P., Janson, C.H. (Eds.), Infanticide by Males and its Implications. Cambridge University Press, Cambridge, UK, pp. 123e151. Pepper, J.W., Smuts, B.B., 2000. The evolution of cooperation in an ecological context: an agent-based model. In: Kohler, T.A., Gumerman, G.G. (Eds.), Dynamics in Human and Primate Societies: Agent-Based Modeling of Social and Spatial Processes. Oxford University Press, pp. 45e76. Pusey, A.E., 1987. Sex-biased dispersal and inbreeding avoidance in birds and mammals. Trends in Ecology & Evolution 2, 295e299. Ratcliff, W.C., Denison, R.F., Borrello, M., Travisano, M., 2012. Experimental evolution of multicellularity. Proceedings of the National Academy of Sciences of the United States of America 109(5), 1595e1600. Ray, T.S., 1991. An approach to the synthesis of life. In: Langton, C., Taylor, C., Farmer, J.D., Rasmussen, S. (Eds.), Artificial Life II. Addison-Wesley, Redwood City, CA, pp. 371e408. Robbins, M.M., Robbins, A.M., 2004. Simulation of the population dynamics and social structure of the Virunga mountain gorillas. American Journal of Primatology 63, 201e223. Robbins, A.M., Robbins, M.M., 2005. Fitness consequences of dispersal decisions for male mountain gorillas (Gorilla beringei beringei). Behavioral Ecology and Sociobiology 58, 295e309. Rodseth, L., Wrangham, R.W., Harrigan, A.M., Smuts, B.B., 1991. The human community as a primate society. Current Anthropology 32, 221e254. Rouly, O.C., 2018. A computer simulation to investigate the association between gene-based gifting and pair-bonding in early hominins. Journal of Human Evolution 116, 43e56. Saj, T.L., Marteinson, S., Chapman, C.A., Sicotte, P., 2007. Controversy over the application of current socioecological models to folivorous primates: Colobus vellerosus fits the predictions. American Journal of Physical Anthropology 133, 994e1003. Sellers, W.I., Hill, R.A., Logan, B.S., 2007. An agent-based model of group decision making in baboons. Philosophical Transactions of the Royal Society of London B: Biological Sciences 362, 1699e1710. Seyfarth, R.M., 1978. Social relationships among adult male and female baboons. I. Behaviour during sexual consortship. Behaviour 64, 204e226. Snaith, T.V., Chapman, C.A., 2007. Primate group size and interpreting socioecological models: do folivores really play by different rules? Evolutionary Anthropology 16, 94e106. Stammbach, E., 1987. Desert, forest, and montane baboons: multi-level societies. In: Smuts, B.B., Cheney, D.L., Seyfarth, R.M., Wrangham, R.W., Struhsaker, T.T. (Eds.), Primate Societies. University of Chicago Press, Chicago, pp. 112e120. Stearns, S.C., 1992. The Evolution of Life Histories. Oxford University Press, Oxford. Sterck, E.H., Watts, D.P., van Schaik, C.P., 1997. The evolution of female social relationships in nonhuman primates. Behavioral Ecology and Sociobiology 41, 291e309. Strier, K.B., 1994. Myth of the typical primate. Yearbook of Physical Anthropology 37, 233e271. Sutcliffe, A.G., Dunbar, R.I.M., Wang, D., 2016. Modelling the evolution of social structure. PLoS One 11, 20. Swedell, L., Plummer, T., 2012. A papionin multilevel society as a model for hominin social evolution. International Journal of Primatology 33, 1165e1193. Te Boekhorst, I.J., Hogeweg, P., 1994. Self-structuring in artificial “chimps” offers new hypotheses for male grouping in chimpanzees. Behaviour 130, 229e252. Terborgh, J., Janson, C.H., 1986. The socioecology of primate groups. Annual Review of Ecology and Systematics 111e136. Trivers, R.L., 1972. Parental investment and sexual selection. In: Campbell, B. (Ed.), Sexual Selection and the Descent of Man 1871-1971. Heinemann, London, pp. 136e179. Trivers, R.L., 1974. Parent-offspring conflict. American Zoologist 14, 249e264. Tutin, C.E.G., 1979. Mating patterns and reproductive strategies in a community of wild chimpanzees (Pan troglodytes schweinfurthii). Behavioral Ecology and Sociobiology 6, 29e38.

K.N. Crouse et al. / Journal of Human Evolution 137 (2019) 102671 van Schaik, C.P., 1983. Why are diurnal primates living in groups? Behaviour 87, 120e144. van Schaik, C.P., Dunbar, R.I.M., 1990. The evolution of monogamy in large primates: a new hypothesis and some crucial tests. Behaviour 115, 30e61. van Schaik, C.P., Janson, C.H., 2000. Infanticide by Males and its Implications. Cambridge University Press, Cambridge, UK, p. 569. van Schaik, C.P., Kappeler, P.M., 1997. Infanticide risk and the evolution of malefemale bonding in primates. Proceedings of the Royal Society of London, Series B 264, 1687e1694. van Schaik, C.P., van Noordwijk, M.A., 1988. Scramble and contest in feeding competition among female long-tailed macaques (Macaca fasicularis). Behaviour 105, 77e98. von Neumann, J., 1966. In: Burks, A.W. (Ed.), The Theory of Self-Reproducing Automata. University of Illinois Press, Urbana. Walker, R.S., Hill, K.R., Flinn, M.V., Ellsworth, R.M., 2011. Evolutionary history of hunter-gatherer marriage practices. PLoS One 6. Wilensky, U., 1999. NetLogo. Center for Connected Learning and Computer-Based Modeling, Northwestern University, Evanston, Illinois. http://ccl.northwestern. edu/netlogo/.

19

Wilensky, U., 2007. NetLogo 4.0. 2 User Manual. Center for Connected Learning and Computer-Based Modeling, Northwestern University, Evanston, Illinois. Wilson, M.L., Britton, N.F., Franks, N.R., 2002. Chimpanzees and the mathematics of battle. Proceedings of the Royal Society of London, Series B 269, 1107e1112. Wilson, M.L., Glowacki, L., 2017. Violent cousins: Chimpanzees, humans, and the roots of war. In: Muller, M.N., Wrangham, R.W., Pilbeam, D. (Eds.), Chimpanzees and Human Evolution. Harvard University Press, Cambridge, MA. Wrangham, R., 1979. On the evolution of ape social systems. Social Science Information 18(3), 336e368. Wrangham, R.W., 1980. An ecological model of female-bonded primate groups. Behaviour 75, 262e300. Wrangham, R.W., 1987. The significance of African apes for reconstructing human social evolution. In: Kinzey, W.G. (Ed.), The Evolution of Human Behavior: Primate Models. SUNY Press, Albany, pp. 51e71. Yaeger, L., 1994. Computational genetics, physiology, metabolism, neural systems, learning, vision and behavior, or PolyWorld: life in a new context. In: Langton, C. (Ed.), Proceedings of Artificial Life III. Addison-Wesley, Reading, MA, pp. 263e298.