Among mammals, female reproduction is generally thought to be food limited, and dominance should theoretically afford high-ranking females with access to better food resources. Although the importance of dominance rank among female chimpanzees (Pan troglodytes) has been debated in the past, mounting evidence suggests that rank is very important among females (P. t. schweinfurthii) at Gombe National Park, Tanzania. In this study, we investigated the influence of season and dominance rank on female foraging strategies. We found that high-ranking females spent less time foraging and tended to have a narrower diet breadth and higher diet quality than subordinate females. In this way, subordinate female foraging strategies were consistent with how females in general adapted to periods of food scarcity. The results of this study therefore suggest that low-ranking females may face persistent “food scarcity” as a result of interference food competition. We also provide evidence that subordinates may forage less efficiently because they occupy lower quality habitats or avoid associating with dominant females in shared areas.
Feeding competition is often considered to be the most important cost associated with grouping (MacDonald 1983) and is frequently invoked to explain differences in primate social structure (Wrangham 1980; van Schaik 1989; Isbell 1991; Sterck et al. 1997). In contest competition, individuals interact directly and some obtain more of a resource due to individual advantages such as size, dominance, or motivation (Milinski and Parker 1991). The influence of social dominance on foraging has been observed in a wide variety of taxa. Dominant individuals have priority of access to food (salmon: Hakoyama and Iguchi 1997; Arctic barnacle geese: Stahl et al. 2002; cleaning gobies: Whiteman and Côté 2004), spend less time in vigilance (yellow-footed rock wallaby: Blumstein et al. 2001), and have higher energy intake rates (bears: Gende and Quinn 2004; white-faced capuchin monkeys: Vogel 2005). As a result, dominant individuals should theoretically have higher reproductive success. The relationship between rank and reproductive success is typically explained in terms of higher resource holding potential for dominant individuals (Parker 1974). In this study, we examine how dominance rank influences foraging behavior in wild female chimpanzees (Pan troglodytes schweinfurthii). To provide a context for interpreting our results, we compare subordinate foraging strategies to dietary adaptations during periods of food scarcity.
Chimpanzees live in complex fission–fusion societies in which subgroups (parties) are transient within a permanent community (Goodall 1986; Nishida and Hiraiwa-Hasegawa 1987). Compared with most other primates, dispersal in chimpanzees is female biased (Pusey and Packer 1987). Males are highly gregarious and range widely in order to defend a community range (reviewed in Wrangham 2000; Williams et al. 2004). The community structure therefore depends on how females use space inside of this male-defended territory. East African chimpanzees (P. t. schweinfurthii) generally conform to a male-bonded model in which females are less social than males. Most females concentrate their space use in small, overlapping core areas to which they have high site fidelity (Wrangham and Smuts 1980; Williams et al. 2002). Researchers have found that female core areas are clustered into neighborhoods at 2 study sites and have suggested that neighborhoods are centered around abundant food resources (Gombe: Williams et al. 2002; Murray CM, Mane SV, Pusey AE, unpublished data; Kanyawara: Emery Thompson M, Kahlenberg SM, Gilby IC, Wrangham RW, unpublished data).
Chimpanzees are ripe-fruit specialists but also consume a wide variety of other plant parts and animal matter (Goodall 1986). Studies have suggested that leaves and pith are fallback foods when high quality fruits are limited (e.g. Conklin-Brittain et al. 1998; Basabose 2002). Food resources are heterogeneously distributed and resource variation influences grouping patterns (e.g. Matsumoto-Oda et al. 1998). Several studies have reported seasonal differences in behavior which suggest that resources are scarcer during the dry season. Party sizes are smaller (Gombe: Wrangham 1977; Tai: Boesch 1996; Doran 1997; Mahale: Matsumoto-Oda et al. 1998), activity periods are longer (Gombe: Lodwick et al. 2004), and body mass decreases (Mahale: Uehara and Nishida 1987; Gombe: Pusey et al. 2005). Faced with food scarcity, we expect that females will alter their foraging strategies according to predictions from foraging models and evidence from other studies. Females may consume lower quality resources (Richard 1985; Wrangham et al. 1991), spend more time foraging (Clutton-Brock and Harvey 1977; Isbell 1991; Janson and Goldsmith 1995), and have a broader diet breadth (Charnov 1976; Stephens and Krebs 1986; DiFiore 2003).
It is generally thought that female chimpanzees spend time alone and space themselves out in order to dissipate food competition (Wrangham and Smuts 1980). However, the mode and importance of resource competition have been debated. The infrequency of female aggression (reviewed in Murray Forthcoming) makes it difficult to assess the importance of contest competition. When female aggression does occur, however, it is most often in the context of feeding (Goodall 1986; Muller 2002) or between resident females and new immigrants (Pusey 1980; Nishida 1989). These observations suggest that females compete for both food and space. This competition may have fitness consequences. Accordingly, Pusey et al. (1997) found that dominant females had higher reproductive success. Given this evidence for the importance of rank, we predict that dominance influences foraging strategies. We further predict that subordinates may compensate for reduced access to food by adopting strategies similar to those exhibited when food is scarce.
Assessing differential access to resources is difficult in wild populations. However, many studies have demonstrated that dominant individuals have higher quality territories (e.g., white footed mice: Korytko and Vessey 1991; black-throated blue warblers: Holmes et al. 1996; red squirrels: Wauters et al. 2001) or priority of access to resources in spatially cohesive groups (e.g., vervet monkeys: Whitten 1983; red deer: Thouless 1990; Arctic barnacle geese: Stahl et al. 2002). Evaluating foraging efficiency among chimpanzees is further complicated by their social system. Because females often forage alone, subordinates may avoid areas used by dominant females as has been shown in studies of other species (e.g., hedgehogs: Cassini and Föger 1995; salmon: Hakoyama and Iguchi 1997; brown bears: Gende and Quinn 2004). Females can simply avoid these areas altogether or avoid them temporally by using the same resources at different times. We predict that dominance rank will influence habitat quality and association patterns, thereby providing insight into how dominant females may obtain better food resources.
In this study, we test the hypothesis that dominant females forage more efficiently than subordinates in a wild chimpanzee population in Gombe National Park, Tanzania. We first assess how females adapt to seasonal food scarcity. We then examine the effect of dominance rank and compare our results to the seasonal analyses. We predict that subordinate female foraging strategies will be similar to those generally exhibited by females when food abundance is lower. We conclude by examining how differences in habitat quality and association patterns may cause subordinates to forage less efficiently.
Study site and period
We investigated female foraging strategies among chimpanzees at Gombe National Park, Tanzania. The climate in Gombe is highly seasonal (Goodall 1986). Based on rainfall data (Jane Goodall Institute's Center for Primate Studies, unpublished data) and the precedence of other studies, we divided each year into a dry (May–October) and a wet (November–April) season (e.g., Wallis 2002; Pusey et al. 2005). Our study focuses on Kasekela community females over a 10-y period (1995–2004). During this time, the Kasekela community contained 19–22 adult females (age ≥ 12 years).
We relied on 2 data sets for our analyses: full-day follow data from 1995–2004 and 1-h follow data from 2003–2004. Since 1973, Tanzanian field assistants and external researchers have conducted full-day follows on members of the Kasekela community. During full-day follows, researchers follow one individual for an entire day and record party composition and location every 15 min (see Goodall 1986). Foraging data on the focal individual are also recorded and include foraging bout length, food part, and food species. Dominance interactions are recorded opportunistically. In subsequent analyses, we relied on data from 1229 full-day follows on 12 adult females.
Full-day data are regularly collected on a subset of community females (Table 1, shaded) because many nulliparous females and new immigrants are difficult to follow. Most full-day focals are also middle and high ranking. To supplement these data, we therefore collected data from a broader set of females that included lower ranking and nulliparous individuals. Two field assistants and C. Murray collected these data over 21 months (January 2003–September 2004). Interobserver reliability scores revealed high levels of agreement (mean index of concordance = 0.897, N = 33 behavioral follows) (Murray 2006). During 1-h follows, we collected point samples of focal behavior (including feeding) every 10 min. After completing a 1-h follow, we rotated through females under the conditions that no female was followed more than twice in 1 day and that at least 2 h passed between successive follows on the same individual. We collected 737 follows on 22 adult females. In subsequent analyses, we included females with at least thirty 1-h follows (N = 18 females) (Table 1).
We ranked females into a dominance hierarchy via the direction of pant grunts and outcome of agonistic interactions. Pant grunts are submissive vocalizations that function as reliable indictors of dominance in chimpanzees (Bygott 1979). For analyses of 2003–2004 foraging data, we used scaled ranks that are continuous and allow for fine degrees of rank separation (following Jameson et al. 1999). The derivation of scaled ranks was based on pant grunt and agonistic data (N = 67 interactions) and was reported in another study (Murray Forthcoming). Scaled ranks could not be calculated for previous years due to the scarcity of dominance data, but we assigned categorical ranks (high, middle, and low) based on the 2003–2004 scaled ranks and ranks as reported from 1970–1992 (Pusey et al. 1997) and from 2001–2002 (Greengrass 2005). For the intervening period, we assigned categorical rank based on pant grunt data from full-day follows and assumptions with respect to the relationship between age and rank and rank stability. Several studies have reported that female rank increases with age and that relative ranks among females are stable over time (Goodall 1986; Nishida 1989; Pusey et al. 1997). Here, we first tested how rank varied with age to verify the use of an age assumption in our data set prior to performing other analyses.
Party size by season
Using 1995–2004 full-day data, we determined the party size at each 15-min point sample. We calculated average party size for a female based on the number of adults and adolescents (age 8–12 years) present at each of her point samples. We limited our analyses to females that were followed at least 30 h during a given year and season.
We divided food into the following categories: leaves, pith, fruit, flowers, sap, and animal matter. In the absence of nutritional data, we considered leaves and pith to be low quality because they are fallback foods (Conklin-Brittain et al. 1998; Basabose 2002). A previous study from Gombe also reported a significant negative correlation between monthly average body mass and the consumption of leaves and pith (Gilby et al. 2006). This relationship was stronger than the relationship between the fruit consumption and body mass, which was not significant. We therefore calculated the proportion of low-quality items within a year and season in full-day data as
Although studies often consider diet breadth as the number of species consumed per month (e.g., DiFiore 2003; Gilby 2004), they have focused on an average over all individuals. Because we were interested in individual dietary differences, considering 1 month failed to yield sufficient data. We therefore defined diet breadth within a year and season as
To measure foraging effort, we calculated the proportion of observation time in which a female was foraging. We controlled for party size because a previous study demonstrated that individuals spent less time foraging in parties (Wrangham and Smuts 1980). We assigned a party size at the start of every foraging bout based on the total number of adults and adolescents present. We did not differentiate between adults and adolescents under the assumption that they consume similar amounts of food. We treated party size as a categorical explanatory variable: small (1–4 individuals), medium (5–10), and large (11+). For full-day data, foraging effort in party size x within a year and season was given by
Data used in analyses
Because some females were followed more than others, we set minimum criteria for inclusion in analyses. For full-day data, we included females that were followed for at least 30 h within a given year and season (diet quality and diet breadth) or at least 10 h within a year, season, and party size (foraging effort). Although we would have ideally limited all metrics with the same criteria, the additional party size constraint made it difficult to apply a 30-h criterion to foraging effort. For 2003–2004 data, we did not include year because these analyses spanned such a short time period. When analyzing diet quality from 1-h follows, we included females that were targeted at least 10 times during a given season.
Habitat quality and association patterns
Habitat quality by rank
To investigate habitat quality by rank, C. Murray collected vegetation data in one hundred fifty 400-m2 plots from January–June 2004. We randomly generated plot coordinates inside of the 95% community usage kernel. Given the diversity of plant species consumed by Gombe chimpanzees (N = 141 species), we collected abundance data on a subset of 10 preferred tree species (Table 2). These species accounted for a yearly average of 48.7% of the vegetation diet. Given the well-supported relationship between diameter at breast height (DBH) and biomass production (Chapman et al. 1992), we measured the DBH for each tree. We then summed DBH inside of a plot as a proxy for plot productivity. We estimated habitat quality for females from the average plot productivity inside of her alone core area. Alone core areas for 2003–2004 were generated following Murray CM, Mane SV, Pusey AE (unpublished data). To test for habitat quality differences by rank, we performed a simple linear regression of average productivity on scaled rank.
|Species||Average proportion of diet|
|Ficus sansibarica sansibarica||0.015|
|Species||Average proportion of diet|
|Ficus sansibarica sansibarica||0.015|
We calculated the proportion of the diet for species x as the time spent foraging on species x/total time spent foraging on vegetation. We report the average yearly proportion for each sampled species for all long-term data (1974–2004).
Association patterns by rank
Although subordinates often settle away from dominant females (Williams 2000; Murray CM, Mane SV, Pusey AE, unpublished data), we were interested in association patterns for females that concentrated their space use in the same area. Accordingly, we categorized females as neighbors or nonneighbors based on the degree of overlap between female alone core areas (Murray 2006). To investigate association patterns by rank, we calculated association indices for unrelated female dyads in 2 well-defined neighborhoods. Our dyadic association measure is equivalent to a half-weight association index (Cairns and Schwager 1987) based on first arrival data. First arrivals occurred whenever a female was encountered by the full-day focal for the first time on a specific day. Two females arrived “together” when they were both encountered for the first time within 5 min of each other. We assumed that arriving together meant that the dyad was associating before being encountered. We calculated a yearly dyadic association index (DAI) as:
We performed all statistical tests with SAS version 9.1 (SAS Institute, Cary, NC). Changes in party size across years and seasons were tested with a linear mixed model controlling for repeated measures on each focal female.
To test associations of categorical rank, year, and season with each of our 3 foraging metrics for 1995–2004, we used linear mixed models that controlled for repeated observations on each focal female. When analyzing foraging effort, we included an interaction term for party size and rank because high-ranking females may obtain better resources (higher quality food items) than do subordinates when feeding in a larger party. These models were repeated for 2003–2004 using scaled rank instead of categorical rank. We fit a similar model to diet quality from the 2003–2004 1-h data using only scaled ranks. We provide effect size estimates for all predictors. Because effect size estimates (such as partial η2: Cohen 1973; Cohen 1988) have not yet been developed for linear mixed models, we estimated an ad hoc effect size using the log likelihood values for each model as follows: log likelihood for the model with the predictor (e.g., rank) minus that for the model without the predictor, divided by that for the model without the predictor. Thus, our effect sizes can be interpreted as the relative increase in the model fit due to the addition of the predictor.
Using dyadic association indices among unrelated females, we tested for association differences by categorical rank with a mixed linear model while controlling for neighborhood membership, the year, and repeated measures on the same dyad. Controlling for neighborhood membership allowed the model to incorporate higher DAIs due to passive association as a function of preferred space use.
Female rank correlated strongly with age (F = 26.8, N = 18 females, P < 0.0001, R2 = 0.626) (Figure 1). In a simple linear regression of scaled rank on age for 2003–2004, age alone explained 62.6% of the variation in scaled rank. Based on the strong relationship between age and rank and supplementary dominance data, we assigned yearly categorical ranks for the entire study period prior to performing foraging analyses that included rank as an independent variable (Table 3).
Party size by season
Average party size varied significantly by season and year (season: F1,119 = 32.71, P < 0.0001; year: F9,119 = 2.18, P = 0.03). Over all years, the average party size was 6.1 individuals (standard deviation [SD] = 4.6) during the dry season, whereas the average party size during the wet season was 9.0 individuals (SD = 4.9).
Foraging by season and rank
We summarize results in Table 4. We found significant differences in female foraging strategies by both season and rank. Females in general adapted to periods of food scarcity by broadening their diet and increasing foraging effort. Low-ranking females had foraging strategies similar to those exhibited during times of food scarcity but also consumed lower quality foods. Relationships are reported in more detail below.
Effect of season and year on foraging
Analysis of 10 years of full-day data demonstrated significant seasonal and yearly differences in female foraging effort and diet breadth (Table 4A). Diet quality did not differ by season but it did vary by year (N = 12 females, season: F1,121 = 1.59, P = 0.21; year: F9,121, P = 0.007). The average yearly female diet consisted of 16.4% leaves and pith, ranging from 9.2–20.4%, depending on the year. We further found that diet breadth varied by season and year (season: N = 12 females, F1,121 = 40.88, P < 0.0001; year: F9,121 = 1.98, P = 0.05). During the dry season, females consumed an average of 0.323 (yearly range 0.254–0.372) species per observation hour. Females had a significantly narrower diet during the wet season and consumed an average of 0.246 (yearly range 0.192–0.288) species per observation hour. Season and year also significantly predicted foraging effort (N = 12 females, season: F1,108 = 7.02, P = 0.009; year: F9,108 = 2.19, P = 0.03). Females showed increased effort during the dry season by foraging for an average 56.8% (yearly range 46.1–67.5%) of their observation time compared with 50.3% (yearly range 42.9–56.1%) during the wet season.
Effect of rank on foraging
We found a tendency towards diet quality differences in 1-h follow data (Table 4C). Lower ranking females tended to have a lower quality diet and consumed more leaves and pith (N = 18 females, F1,15 = 3.88, P = 0.07). Females with one scaled rank unit higher foraged an average of 2.6% less on leaves and pith.
We found a tendency towards differences in diet breadth by scaled rank in 2003–2004 full-day follow data (Table 4B). After controlling for season, higher ranking females tended to have a narrower diet breadth (N = 9 females, F1,6 = 3.28, P = 0.12), consuming an average 0.04 fewer species per hour for each one unit higher rank.
We found significant differences in foraging effort by rank for 2003–2004 full-day data (N = 11 females, F1,27 = 7.69, P = 0.01) (Table 4B). Lower ranking females spent more time foraging. The party size × rank interaction was also significant, as rank differences were more pronounced in medium-sized parties (F2,27 = 24.78, P < 0.0001). Party size also predicted foraging effort as females in smaller parties spent more time foraging (F2,27 = 20.72, P < 0.0001) . These differences by party size were observed for the entire study period as well (F2,108 = 11.59, P < 0.0001). From 1995–2004, females had an average foraging effort of 56.7% in small parties. They spent less time foraging in medium (44.7%) and large (45.1%) parties.
Habitat quality and association patterns
Habitat quality by rank
Higher ranking females occupied alone core areas of higher productivity (N = 18 females, F1,16 = 24.06, P < 0.0001, R2 = 0.601) (Figure 2). Rank explained approximately 60.1% of the observed variation in habitat quality as measured via cumulative DBH.
Association patterns by rank
Dyadic association indices varied significantly by neighborhood, rank, and year (year: F9,308 = 15.71, P < 0.0001). Females had much stronger levels of association with their neighbors (neighborhood: F1,308 = 21.42, P < 0.0001). The mean DAI among neighbors was 0.154 compared with a mean DAI among nonneighbors of 0.096. After controlling for the neighborhood relationship, we detected differences by dyadic rank combination (rank: F5,308 = 2.67, P = 0.02). Post hoc group comparisons of high and low association patterns revealed that low-ranking females associated least with dominants (group comparisons with Bonferroni adjustment: HL vs. LL and HL vs. LM, P < 0.008) (Figure 3). By comparison, dominant females had similar association patterns with high, middle, and low-ranking females (group comparisons with Bonferroni adjustment: HH vs. HM, P = 0.49; HH vs. HL, P = 0.92; HM vs. HL, P = 0.44).
The results of this study demonstrate that season and dominance rank influence wild female chimpanzee foraging strategies. Recent studies from Gombe have found lower body masses (Pusey et al. 2005) and increased activity periods (Lodwick et al. 2004) during the dry season. Earlier studies also reported smaller parties (Wrangham 1977), which we confirmed during our study period. These findings provide strong evidence that the dry season is a period of food scarcity, and our results demonstrate how females compensated for reduced food abundance. Females adapted to food scarcity by increasing their diet breadth and foraging effort. However, female diet quality did not differ by season. The flexibility of a fission–fusion social structure may provide a buffer, whereby females can adjust grouping and ranging patterns to maintain diet quality. Interestingly, a study of male diet quality reported a tendency for seasonal variation (Murray 2006). Males may have less social flexibility than females due to higher benefits of grouping in terms of alliance maintenance and territorial defense (reviewed in Wrangham 2000). Males may therefore be more susceptible to environmental resource variation.
In addition to seasonal adaptations, we found that female rank influenced our foraging metrics. It should be noted that categorical ranks (high, middle, and low) failed to yield significant results but scaled ranks yielded some significant differences. Although we do not propose that a linear hierarchy is necessarily most appropriate for Gombe females, finer degrees of rank differentiation such as our scaled rank may be useful in this system. Taken together, our results demonstrated that dominant females spent less time foraging on fewer foods and on higher quality foods. In this way, subordinate versus dominant foraging strategies were directionally consistent with results for how females in general adapted to seasonal food scarcity. The results of our study therefore suggest that low-ranking females may face persistent food scarcity as a result of interference food competition. This is further supported by the fact that dominant females have a heavier and less variable body mass (Pusey et al. 2005).
Numerous studies have reported that foraging efficiency increases with age (e.g., European blackbirds: Desrochers 1992), presumably due to enhanced foraging competence and resource familiarity. Given the strong correlation between age and rank in our data set, we were concerned that age could confound our results if older individuals forage more efficiently. To assess this possibility, we ran identical analyses but substituted age for rank. We did not detect any age effects for full-day follow data (Murray 2006). However, age did influence diet quality when we analyzed 1-h follow data (Murray 2006). This dataset contained several new immigrants and young individuals that were not included in full-day follows. Our results demonstrated that these females had a lower quality diet than older, more established females. This may reflect increasing familiarity with higher quality resources as young females and new immigrants learn an area.
Despite varying levels of significance, within-model rank effect sizes were generally as or more pronounced than seasonal effect sizes. Detecting significance may be limited by small sample sizes and the bias towards sampling high-ranking females. We should also note that we relied on a gross measure of diet quality that did not incorporate nutritional data or measure food intake rates. It is likely that detailed dietary data would reveal even more pronounced rank differences (e.g., pronghorn females: no rank difference based on behavioral data: Byers 1997; rank differences based on chemical analyses: Dennehy 2001).
The biological significance of the differences found in this study requires additional study. However, the influence of food acquisition on female reproductive success has been documented in numerous taxa (e.g., Columbian ground squirrels: Ritchie 1990; water striders: Blanckenhorn 1991; red deer: Conradt et al. 1999; voles: Jonsson et al. 2002). We therefore propose that the foraging differences reported here may partially account for rank differences in reproductive success in female chimpanzees (Pusey et al. 1997).
Assessing differential access to resources is very challenging in a fission–fusion society. Dominant females may 1) occupy areas of higher value, 2) use shared resources at the optimal time, or 3) consume the best resources when foraging in a group context. These mechanisms warrant further investigation, but our results demonstrate that high-ranking females occupied more productive habitats. This result is concordant with studies from various taxa which have demonstrated that dominants occupy preferred habitats (e.g. black-throated blue warblers: Holmes et al. 1996; red squirrels: Wauters et al. 2001). The detection of habitat quality differences among chimpanzees is particularly striking because female core areas are extremely overlapped (Wrangham and Smuts 1980; Williams et al. 2002).
Even though dominant females occupy areas of higher value, it should be noted that several low-ranking females live in the same neighborhood as dominant females (Williams et al. 2002; Murray 2006). In both Kasekela neighborhoods, we found that subordinates associated least frequently with dominant individuals. These patterns could result from preferential association between low-ranking females, avoidance of dominants by subordinates, or active exclusion of subordinates by dominant females. Although we cannot differentiate between the first 2 possibilities, competitive exclusion seems unlikely in light of the infrequency of female aggression (reviewed in Murray Forthcoming). Given the positive relationship between rank and reproductive success, it is surprising that female aggression occurs so infrequently. However, the potentially high costs of fights may preclude direct contests (Nishida 1989) in favor of subtler dominance interactions. Such “passive deferrals” have been observed in other species (e.g., brown bears: Gende and Quinn 2004). Regardless, the observed association patterns may indicate differential access to shared resources because high-ranking individuals often use shared resources at the best time (hedgehogs: Cassini and Föger 1995; giant kokopu: Hansen and Closs 2005).
Studies quantifying differential access to resources in wild populations have generally focused on territorial species or species living in spatially cohesive groups (e.g. black-throated blue warblers: Holmes et al. 1996; Arctic barnacle geese: Stahl et al. 2002). Here, we demonstrated rank effects despite our reliance on fairly coarse foraging metrics and the challenges of a complex fission–fusion social system. We also provided evidence that subordinates may forage less efficiently because they occupy lower quality habitats or avoid associating with dominant females in shared areas.
We thank Tanzania National Parks, the Tanzania Wildlife Research Institute, and the Tanzanian Council for Science and Technology for granting us permission to work on this project in Gombe National Park. We also thank the Jane Goodall Institute for funding data collection at Gombe, the Gombe Stream Research Center staff, and Dr Jane Goodall for granting us permission to work with the long-term data set. We thank Dr Mark Bee and 2 anonymous reviewers for suggestions on an earlier draft of this article. We thank Dr Paul Bolstad for advice on vegetation sampling and analysis. C.M.M. would also like to acknowledge her field assistants, especially S. Athumani, M. Mlongwe, and M. Msafiri, and Mete Celik for helping to extract dyadic association data. C.M.M. was supported by a grant from Milton Harris, a Dayton–Wilkie Fellowship, and the Graduate School at the University of Minnesota. C.M.M. and A.E.P. were both supported by a grant from the National Science Foundation (NSF # IIS-0431141).