Inferences concerning the lives of extinct animals are difficult to obtain from the fossil record. Here we present a novel approach to the study of extinct carnivores, using a comparison between fossil records (n=3324) found in Late Pleistocene tar seeps at Rancho La Brea in North America and counts (n=4491) from playback experiments used to estimate carnivore abundance in Africa. Playbacks and tar seep deposits represent competitive, potentially dangerous encounters where multiple predators are lured by dying herbivores. Consequently, in both records predatory mammals and birds far outnumber herbivores. In playbacks, two large social species, lions, Panthera leo, and spotted hyenas, Crocuta crocuta, actively moved towards the sounds of distressed prey and made up 84 per cent of individuals attending. Small social species (jackals) were next most common and solitary species of all sizes were rare. In the La Brea record, two species dominated, the presumably social dire wolf Canis dirus (51%), and the sabretooth cat Smilodon fatalis (33%). As in the playbacks, a smaller social canid, the coyote Canis latrans, was third most common (8%), and known solitary species were rare (<4%). The predominance of Smilodon and other striking similarities between playbacks and the fossil record support the conclusion that Smilodon was social.


1. Introduction

An icon of the North American Pleistocene, the sabretooth (Smilodon fatalis), was a formidable predator with robust forelimbs and elongate upper canine teeth. Although usually portrayed as solitary, controversy persists over its possible sociality (McCall et al. 2003). The strongest evidence for sociality comes from the late Pleistocene tar seep deposits of Rancho La Brea, California (44 000–9000 YBP; Stock & Harris 1992), where high ratios of Smilodon and the presumably social dire wolf (Canis dirus) relative to prey species suggest that both hunted in groups.

Here we present additional evidence supporting sociality in Smilodon using a comparison between playback experiments carried out in eastern and southern Africa to estimate carnivore abundance (Mills et al. 2001) and the fossil record from Rancho La Brea (Marcus 1960; Stock & Harris 1992; Spencer et al. 2003). Both playbacks (Cooper 1991; Fanshawe & Fitzgibbon 1993) and the tar seeps represent highly competitive scenarios, in that a dying or dead animal drew in multiple predators, that then potentially competed for food. In the playbacks, sounds of dying herbivores combined with the sounds of lions (Panthera leo) and hyenas (Crocuta crocuta) were used to attract carnivores for population estimation (Mills et al. 2001). At La Brea, carnivores appear to have been similarly lured by dying herbivores and trapped in sticky asphalt. The tar pit deposits represent an accumulation of these events. Carnivores outnumber herbivores nine to one in the La Brea record and commonly several individual carnivores attended playbacks. The parallels between these records, and their broader significance for understanding carnivore competition, led us to explore this in greater detail.

2. Material and methods

(a) Size and social categories

We grouped carnivore species into four categories (table 1), based on size and social status: ‘small’ (<21 kg) and ‘large’ (≥21 kg), following previous studies on carnivore diets (Carbone et al. 1999), ‘solitary’ (mostly found alone) or ‘social’ (typically in monogamous pairs or larger groups). Extant species in the fossil record (table 1) are assumed to be similar in social category to the present. Extinct species belonging to families with consistent social systems are assumed to be similar to extant relatives (Wrangham et al. 1993). Thus, bears (Ursidae) are classified as solitary and dogs (Canidae) as social. However, Smilodon, a felid, has both solitary and social relatives and we focus our analysis on examining the evidence pertaining to its social status.

View this table:
Table 1

Carnivore species present in region and those observed (numbers and percentages) attending playbacks in eastern and southern Africa and found in La Brea tar seeps in the late Pleistocene of western North America. (References for body mass, social status and population density estimates are in electronic supplementary material S1 and S2; see also text for details. *The social status of Homotherium and American lion were set as solitary and social, respectively. In the former species, its rarity in the fossil record suggests that it may have been solitary, in the American lion the nearest relatives were social. Since the two species make up less than 3% of the record, their classification has only a marginal influence on the overall pattern.)

(b) Playback data

We based our estimates of carnivores attending potential kill sites from substantial playback surveys using broadly similar methods (Mills et al. 2001; Maddox 2003; Ferreira & Funston submitted) (Kruger National Park, n=578; Serengeti Region, n=232). Pre-recorded sounds of herbivore prey species in distress, sometimes in association with lions at a kill and hyenas mobbing, were played for 30–120 min, at night, dawn and dusk and all individuals that approached were identified (Mills et al. 2001; Maddox 2003; Ferreira & Funston submitted).

(c) Community density

We used published estimates of species' densities (table 1, electronic supplementary material S1) to calculate the percentage of individuals found in the community for comparison with the percentage attending playbacks. Estimates of age ratios (juvenile to adult) for lion in the playbacks were compared with the La Brea record. Estimates of numbers of individuals for the La Brea tar seeps were based on the minimum numbers of individuals from all excavations. Juveniles were identified by the presence of deciduous teeth (electronic supplementary material S2).

(d) Statistical analysis

To compare the observed frequencies of carnivores in the four size/sociality categories with the frequencies expected based on relative densities, we used Pearson chi-squared goodness-of-fit tests (electronic supplementary material S2). To examine the importance of size and sociality in explaining the observed patterns, we used negative binomial regression with the number of individuals of each carnivore species attending as the dependent variable, size/sociality categories as the independent variables, and density as an offset (exposure) term to control for the relative densities of species. A Wald test was used to compare the coefficients for size and sociality. To obtain 95 per cent bootstrap confidence intervals for the proportions of individuals in each size/sociality category (see figures 1 and 2), the data were re-sampled 10 000 times with replacement. To compare the La Brea and playback datasets depending on whether Smilodon was social or solitary, we used the following Monte Carlo simulation. Expected frequencies were generated by multiplying the tar seep sample size (n=3324) by the proportions of individuals in the four size/sociality categories in the Africa playback dataset and this distribution was re-sampled with replacement 10 000 times. In each replicate a chi-squared goodness-of-fit value was calculated for each model, and a paired t-test was used to compare the means of the two distributions of chi-square values. The proportions of juvenile large carnivores found at La Brea were compared with the proportion of juvenile lions at playbacks using chi-squared tests. For non-significant tests, post hoc power analyses were carried out to estimate the probability of finding a significant difference (at α=0.05) if the true proportion of juveniles in the pits was 50 per cent lower than the observed proportion of juvenile lions in playbacks. The chi-squared results for coyote (Canis latrans) and American lion are questionable due to small samples, but Fisher's exact tests (not shown) yielded similar results. Statistical analyses were conducted using Stata 10.0 (Stata Corp., College Station, TX, USA).

Figure 1

Comparison of percentages of African carnivores attending playbacks from Kruger National Park (black bars), South Africa, the Serengeti region (white bars), Tanzania and the mean of all playbacks (grey bars; ±confidence intervals), against the percentages based on typical densities (hatched bars) of African carnivores for four size/social structure categories (See also table 1, electronic supplementary material S1).

Figure 2

Comparison of mean percentage (±confidence intervals) of individual African carnivores attending playbacks (grey bars) against the mean percentage (±confidence intervals) of individual North American carnivores recorded in the La Brea tar seeps (black bars) for four size/social structure categories, assuming Smilodon was (a) social and (b) solitary (See also table 1, electronic supplementary material S2).

3. Results

A comparison of the number of individuals attracted to playbacks to the expected number based on community densities reveals a consistent pattern. Large social carnivores make up approximately 85 per cent of all individuals attending playbacks. This differs markedly from the percentages expected based on community densities, where most individuals are from small solitary species (figure 1; table 1, electronic supplementary material S1; Χ32=23 3061, p<0.0001, n=4491). When analysed separately by sociality category, large species are overrepresented compared with small species (social: Χ12=18509, p<0.0001, n=4444; solitary: Χ12>358, p<0.0001, n=47); and when analysed separately by size category, social species are overrepresented compared with solitary species (small: Χ12=6864, p<0.0001, n=630; large: Χ12=1405, p<0.0001, n=3861) (figure 1). Size and sociality are equally influential (negative binomial regression; size coefficient=6.07±1.07, sociality coefficient=6.22±1.06; Wald test: p=0.88). Large social carnivores attended approximately 60 times more often than expected.

The Rancho La Brea tar seep records include 17 species of carnivore, 5 of which are extinct (table 1). If we classify Smilodon as social (figure 2a), the percentages of specimens in each category are very similar to the percentages attending the playbacks, with the large social species representing approximately 88 per cent (table 1). As in the playbacks, the second most common group comprises coyotes (C. latrans), a small social canid. Classifying Smilodon as solitary, we find a marked mismatch between playback and La Brea datasets (figure 2b). The difference in fit to the playback data between the two models (Smilodon social versus Smilodon solitary) is highly significant (Monte Carlo simulation, p<0.0001), supporting the view that Smilodon was social.

Additional evidence for sociality in Smilodon comes from the fact that a number of juvenile sabretooth cats died in the tar pits, indicating that young attended the site with adults. This was also true of La Brea lions, dire wolves and and coyotes, but not of the presumably solitary short-faced bears among which only one in 33 was a juvenile. In the playbacks, lion cubs were observed in the presence of several (2–7) adults (mean=3.4, n=5; mean=4.3, n=7, Serengeti and Kruger, respectively). Notably, the proportion of juvenile to adults for four La Brea species, which were likely to be social, Smilodon, dire wolves, coyote and American lion (27.4, 28.6, 37.5 and 33.3%, respectively), do not differ significantly from the same proportion observed for African lions at playbacks (34.4%; p>0.24, electronic supplementary material S2). This is not the case for the short-faced bear (p<0.0001).

4. Discussion

Overall we find similarities between playbacks in Africa and the fossil record in the Late Pleistocene of North America. In both the records, there are parallels in the types of species attending. In Africa large social carnivores attended playbacks 60 times more often than expected based on background densities, while large and small solitary carnivores attended less frequently than expected. The African wild dog (Lycaon pictus) is an exception, being a large social species that is mostly absent from playbacks. In the tar seep deposits, most of the species for which social status can be reliably inferred show patterns consistent with those in Africa. The large, presumably social, dire wolf is common and the presumably solitary bears and two large cats, the cougar (Puma concolor) and jaguar (Panthera onca), are rare. The social grey wolves in the fossil record, on the other hand, were rare, but this is similar to African wild dogs in playbacks. For the three remaining large cat species, the social status is unsure, but classifying Smilodon as social, we obtain strikingly similar patterns in both the playback and fossil records. This along with the presence of a high juvenile to adult ratio in Smilodon (similar to lions today) suggests that this cat was social.

Small social carnivores were the next most abundant group in both the playbacks and fossil record. In the playbacks, jackals made up between 13 and 15 per cent of the individuals attending. At Rancho La Brea, an ecologically similar species, the coyote, represents approximately 8 per cent of the records and probably filled an analogous niche. Unfortunately, in the tar seeps, the interpretation of rare species such as the American lion and Homotherium is difficult, because it is hard to determine whether they were avoiding these events because they were solitary, as leopards, or were social but existed at low densities (as African wild dogs). In these two large species of cats, we classified the American lion as social (because if its relatedness to African lions) and Homotherium as solitary, owing to its extreme rarity in the fossil record.

Because Smilodon was well armed with enormous canine teeth, perhaps single individuals could have dominated competitive interactions with social carnivores. Indeed, during contests, exceptionally large species can dominate groups of smaller species (Cooper 1991; McNulty et al. 2001). However, Smilodon was substantially smaller than American lion and the short-faced bear also present at La Brea (table 1). In addition, the African lion is social despite the fact that it represents the largest and most dominant species in the African carnivore guild. In this species, larger lion prides dominate smaller ones and intraspecific competition is seen as a major factor driving the evolution of coalitions (Packer et al. 1990). In Smilodon, therefore, given the potential for high levels of competition with other very large carnivores, including members of its own species, it seems very unlikely that as a solitary species, it would have approached these events with such a high frequency.

Our results also have broader implications for understanding the evolution of social behaviour. Carnivores have been used as a model group for developing an understanding of other meat-eating species, including early hominids (Schaller & Lowther 1969). Scavenging was an important food source for early hominids, being linked with many important adaptations, including larger body size and bipedalism (Plummer 2004). In early humans and in other carnivorous species, sociality, through greater coordination and strength in numbers, would probably have significantly enhanced their ability to compete with other species, thereby increasing caloric intake. These results support the idea that social organization and size are important factors influencing status within competitive predatory guilds.


We thank S. Cox, D. McNulty, S. Meiri. K. Safi, D. Stahler, S. Turvey, R. Wayne and the ATS Statistical Consulting Group at UCLA for their help.


    • Received September 9, 2008.
    • Accepted October 8, 2008.


View Abstract