Matriarchs As Repositories of Social Knowledge in African Elephants

See allHide authors and affiliations

Science  20 Apr 2001:
Vol. 292, Issue 5516, pp. 491-494
DOI: 10.1126/science.1057895


Despite widespread interest in the evolution of social intelligence, little is known about how wild animals acquire and store information about social companions or whether individuals possessing enhanced social knowledge derive biological fitness benefits. Using playback experiments on African elephants (Loxodonta africana), we demonstrated that the possession of enhanced discriminatory abilities by the oldest individual in a group can influence the social knowledge of the group as a whole. These superior abilities for social discrimination may result in higher per capita reproductive success for female groups led by older individuals. Our findings imply that the removal of older, more experienced individuals, which are often targets for hunters because of their large size, could have serious consequences for endangered populations of advanced social mammals such as elephants and whales.

Although there is considerable interest in the evolution of social intelligence (1–4), we still know little about how wild animals gain and store information about social companions or whether the possession of superior social knowledge enhances fitness. When mammals live in complex fission-fusion societies (5–7), individuals may encounter hundreds of others during their daily ranging patterns, and time intervals between repeated exposures to the same individuals may be extremely long. Under these circumstances, the problem of distinguishing genuine strangers from a wide range of more regular associates is far from trivial.

Female African elephants live in matrilineal family units led by the oldest female, or matriarch. Within our Kenyan study population, a single family unit directly encounters, on average, 25 other families in the course of a year, representing a total of around 175 adult females (8). Previous paired playbacks indicated that adult females are familiar with the contact calls of around 100 others in the population, being able to discriminate between calls on the basis of how often they associate with the caller (9). However, families differ in how good they are at this task. Here we assess the discriminatory abilities of particular families in detail as a way to investigate the causes and consequences of social knowledge acquisition. In particular, we test the hypothesis that family units with older matriarchs are superior at discriminating the calls of close associates from those of distant associates, statistically excluding other potential explanatory factors. We also examine the relationship between matriarch age and reproductive success, linking reproductive success to matriarch age and response to playbacks. This study provides an unusual opportunity to investigate the role that memory can play in the social behavior of a long-lived vertebrate species.

In our study population in Amboseli National Park, Kenya, data on life histories and association patterns have been obtained for more than 1700 individual elephants over 28 years by the Amboseli Elephant Research Project (7, 9, 10). Vocal discrimination abilities were tested by giving each of 21 family units, over the course of 7 years, a series of playbacks (11) of contact calls from adult females in other families in the population with whom they had a range of association indices (12). Contact calls, which have infrasonic fundamental frequencies but harmonics that extend well into the audible range, are the most common call that adult females use to advertise their location to widely spaced social companions (9, 13). The probability of families of subjects bunching into defensive formation on hearing playbacks of calls from other families decreased with the association index with the caller, conforming to a logistic curve [logistic model on the binary variable of bunching (14): the effect of association index χ2 = 15.30, df = 1,P < 0.001]. This curve describes the probability of bunching as the association index increases, generally leveling to zero at high indices. The detection of the presence of less familiar females (low-association-index families) is potentially beneficial, because this section of the population is more likely to initiate agonistic disputes or harass young calves (9).

The age of the matriarch (10, 15) had a significant effect on the probability of bunching when controlling for the association index with the caller and family identity, and there was a significant interaction between the age of the matriarch and the association index (Table 1). Other potentially confounding variables—the number of females in the group, the mean age of females other than the matriarch, the number of calves, the age of the youngest calf, and the presence or absence of adult males—were excluded from the model because their effects were not found to be statistically significant. The probability of bunching decreased with increasing matriarch age, suggesting that families with older matriarchs may have either larger networks of vocal recognition or greater social confidence than families with younger matriarchs. However, of greater importance with respect to our hypothesis was the interaction between the age of the matriarch and the association index with the caller. Specifically, the sensitivity of the bunching response to the association index increased with the age of the matriarch, so that families with older matriarchs were relatively much more reactive to females with whom they had a low association index than to those with whom they had a high association index (Table 1 and Fig. 1A). For example, although families with old matriarchs (55 years) were several thousand times more likely to bunch in response to calls from families with whom they had a low association index (0.01) than to those with whom they had a high association index (0.1), the probability of bunching for families with young matriarchs (35 years) increased only marginally (×1.4) across these conditions (Fig. 1A). If families with older matriarchs were simply more confident, we would predict their lower level of defensiveness overall but not these marked differences in the slope of the logistic curves. Rather, families with older matriarchs appear considerably more adept at using auditory signals to correctly discriminate between familiar and unfamiliar females in the vicinity and respond appropriately.

Figure 1

The variation of response to playback calls as a function of the association index for families with matriarchs of differing ages. Values depicted are those from a logistic regression model (as described in Table 1) for families with matriarchs of 35 years (young matriarchs: dashed line) and matriarchs of 55 years (old matriarchs: solid line). Although age was a continuous variable throughout our analyses (the range of ages in our sample was 27 to 67 years), here we focus on two age groups that are representative of young and old matriarchs in order to clearly illustrate the interaction between the age of the matriarch and the association index with the caller. Standard error bars predicted from the models are depicted as a guide. The graphs describe probabilities of (A) bunching and (B) smelling.

Table 1

Deviance tables obtained after fitting binary responses to playbacks to a logistic regression model.

View this table:

Although bunching in response to a playback is primarily a defensive behavior, smelling must constitute a means of further exploration: When subjects smell after playbacks, they attempt to gather olfactory information on the caller's identity. An examination of the binary variable “smelling” in response to playbacks showed the same interaction between the age of the matriarch and the association index that characterized the bunching response (Table 1). Subjects in groups with older matriarchs were much more likely to use their trunks to smell if played calls from low- rather than high-association-index families (Fig. 1B). In contrast, groups of subjects with younger matriarchs failed to show this relatively inflated probability of gathering olfactory information about infrequent associates. It is important to note that here families with older matriarchs were at least as reactive as families with younger matriarchs to callers with low association indices. However, their ability to distinguish between calls from low- and high-association-index families was apparently much greater.

Neither the smelling nor the bunching response was affected by the number or age of the other females in the group (Table 1), indicating that it is the matriarch who signals to the rest of her group whether defensive or exploratory behavior is necessary. Although determining the mechanism by which the matriarch orchestrates this change in group behavior is not within the scope of the current study, a subtle acoustic or olfactory cue is the most likely possibility. There are anecdotal reports in the literature of females following the lead of their matriarch in other coordinated group activities (7,13, 16).

The superior discriminatory abilities of older matriarchs should translate into reproductive benefits for the family unit, because time is more efficiently allocated by reserving defensive behavior for circumstances where it is appropriate, and because opportunities for cooperation with more frequent associates are provided. In support of this hypothesis, using an analysis that controlled for the number of females per family, the age of the matriarch was a significant predictor of the number of calves produced by the family per female reproductive year over the course of the study (17), our standardized measure of recent reproductive success (Table 2). Additional factors, including knowledge that older matriarchs had accumulated in a variety of other domains, might have contributed to this association. To explore the relationship between responses to playbacks and reproductive success more specifically, we calculated inflection points for the logistic curves of bunching and smelling on the association index for each family [as –α/β, see (18)]. These were used as aggregate statistics describing between-family differences in response to playbacks. Entering these into an analysis of the residual variation in reproductive success after removing the variance due to the age of the matriarch (and the number of females) showed that families appeared to derive a reproductive benefit (P = 0.05 to 0.08) by bunching and smelling more readily; that is, by showing greater caution and exploratory behavior when presented with another female's call. Thus, it is likely that families with old matriarchs benefit reproductively because their matriarchs target caution at the appropriate individuals: callers that are strange to them. Also, the independent reproductive benefit from exhibiting more caution and exploratory behavior (once the effects of matriarch age have been removed) may explain why families that have young matriarchs, with their reduced social knowledge, show a greater overall tendency to bunch and smell.

Table 2

Regression analysis of factors affecting the reproductive success of elephant family units (calves per female reproductive year) over the course of our study.

View this table:

Previous researchers have speculated that individuals can derive fitness benefits from an improvement in ecological knowledge that accompanies aging (19, 20). Our results suggest that aging may also influence reproductive success through its effects on the acquisition of social knowledge and that the possession of enhanced discriminatory abilities by the oldest individual in a group of advanced social mammals can influence the social knowledge of the group as a whole. These findings have important implications for conservation as well as evolutionary biology. Tusk size in elephants is related to age, and hunters focus their efforts on individuals that have large tusks (21). In view of our results, it is clear that the removal of matriarchs from elephant family units could have serious consequences for the conservation of this endangered species. Indeed, in many mammal societies, the oldest individuals are also the largest, and these tend to be particular targets of hunters (22) and poachers. If groups rely on older members for their store of social knowledge, then whole populations may be affected by the removal of a few key individuals.

  • * To whom correspondence should be addressed. E-mail: karenm{at}


View Abstract

Navigate This Article