Introduction

Territoriality plays a key role in determining the structure, distribution, and behavior of individuals, groups, and populations over available space, as it regulates how organisms share a particular area (Brown 1964; Dugatkin 2009; Maher and Lott 2000). However, we do not yet have a good understanding of the pressures leading to differences in territoriality within and among species (Packer et al. 2005; Potts and Lewis 2014; Shonfield et al. 2012; Sorato et al. 2015). Fundamentally, territoriality refers to the defense of an area against intruders through physical defense or tenure advertising displays (Bartlett and Light 2017; Bates 1970; Börger et al. 2008; Brown 1969; Burt 1943; Carpenter and MacMillen 1976; Maher and Lott 2000; Mai et al. 2005; Nice 1941; Noble 1939; Potts and Lewis 2014; Powell 2000; Tinbergen 1957; Wilson 1975). Territorial displays consist of auditory, visual or olfactory signals that advertise area tenure to others, which act as a proxy of the riskier and more energy consuming overt defense (e.g., Presbytis johnii: Pourier 1968; Hylobates agilis: Mitani 1987; Lemur catta: Kappeler 1998). Consequently, territorial behavior is usually associated with a fixed and exclusive area called a territory. However, there are a wide variety of conceptual definitions of territoriality, making it difficult to operationalize and compare between researchers (reviewed by Maher and Lott 1995). Furthermore, depending on the definition used in a given taxon, the study of territoriality may need quite detailed behavioral data (e.g., on ranging, home range overlap, intergroup behavior, critical resources, etc.), which can be difficult to acquire.

One related, but more accessible, parameter that studies have frequently used to understand territoriality among animals, and particularly primates, is range defensibility. Range defensibility refers to the capacity for an animal to move over a given area to monitor and defend it from intruders (Lowen and Dunbar 1994; Mitani and Rodman 1979). Mitani and Rodman (1979) operationalized range defensibility using the ratio of mean day journey length to the diameter of the home range: the D index. Lowen and Dunbar (1994) further developed the index by taking into account the detection distance of intruders, but this study reached the same conclusions as Mitani and Rodman (1979): animals are more likely to be territorial when the ratio of daily movement to home range is >1.

Territoriality should occur when critical resources are sufficiently abundant and predictable in space and time (i.e., when they are economically monopolizable), so that the benefits from defending the area containing them outweigh the costs (Brown 1964; Carpenter and MacMillen 1976; Grant et al. 1992; Pulliam and Caraco 1984; Schoener 1987). Similarly, range defensibility should follow this economic notion of space; the benefits of being mobile should be more rewarding than its costs (Kinnaird 1992; Maher and Lott 1995; Schoener 1987). Kinnaird and O’Brien (2007) found a clear relationship between hornbill body size and D, supporting a classification of territorial and nonterritorial hornbills by linking D to fighting ability. In primates, Mitani and Rodman (1979) and Lowen and Dunbar (1994) both found that that D positively discriminated territorial from nonterritorial, with a tendency to defend a range when it is more economically practicable to patrol it (D > 1). Several other studies of primates have also associated D to territoriality (e.g., Chapman and Fedigan 1984; Peres 2000; Van Schaik 1992; Wich and Nunn 2002; Willems and van Schaik 2015). However, none, to date, have examined the selection pressures driving it.

Howler monkeys (Alouatta spp.) may indicate territoriality in their eponymous roars (e.g., Bernstein 1964; Carpenter 1934; Da Cunha and Jalles-Filho 2007; Horwich and Gebhard 1983), scent marking (Braga Hirano et al. 2008; Sekulic and Eisenberg 1983), and collective defecation (Braza et al. 1981; Drubble and Gautier 1993; Shoemaker 1982). However, there could be different interpretations of these behaviors. Rather than having a territorial function, roaring could act as a mechanism for intergroup spacing and mutual avoidance (Chiarello 1995; Da Cunha and Byrne 2006; Kitchen 2006; Sekulic and Chivers 1986). Scent marking, defecation, and intergroup encounters can occur anywhere in the home range, and are not clustered at boundaries (Chivers 1969; Cornick and Markowitz 2002; Gavazzi et al. 2008; Gittins 1980). Finally, the home ranges of neighboring groups often overlap substantially (Gavazzi et al. 2008; Klein 1974; Whitehead 1989), which violates the essence of a territory (Brown 1969).

Those who argue that howler monkeys are not territorial have suggested that their highly folivorous diet and physiological inability to process cellulose do not allow them to spend time in aggressive intergroup encounters or to patrol home ranges (Crockett and Eisenberg 1987; Strier 1992). The traditional argument is that leaves are evenly distributed in the landscape and do not to permit monopolization and food competition (Janson and van Schaik 1988). However, howler monkeys are more correctly defined as folivore-frugivores and evidence exists for competition over limited fruit resources (Chapman 1988; Chaves and Bicca-Marques 2016; Jones 1980). In addition, there is also support challenging the idea that folivory does not create food competition (Sayers 2013; Snaith and Chapman 2005); howler monkeys are highly selective in the leaves they consume (they forage from young leaves from a wide range of species, to avoid intoxication from secondary compounds) and, thus, competition for limited leaves can also arise (Arroyo-Rodriguez and Mandujano 2006; Knopff and Pavelka 2006; Milton 1979). Moreover, there is evidence that howler monkeys sometimes behave aggressively towards neighbors (Cristóbal-Azkarate et al. 2004; Degusta and Milton 1998).

Thus, there is no general agreement about the extent of territoriality among howler monkeys, or the factors mediating range defensibility. Here we studied range defensibility among howler monkeys using 63 groups, representing 8 species, and examined its socioecological correlates using linear mixed models.

Methods

Study Groups

We used the database published by Fortes et al. (2015) on the movement ecology of 91 groups of howler monkeys, belonging to 8 species. Only 64 groups, however, had sufficient information on home range size and daily path length to allow us to calculate D (Electronic Supplementary Material [ESM] 1). For these 64 groups, we extracted the following socioecological variables: study area (km2), group size, number of adult females, number of adult males, number of immatures, howler monkey density (individuals per hectare), percentage of fruit in the diet, forest type (araucaria forest, dry deciduous forest, rainforest, semideciduous forest, seminatural forest, and subtropical forest), and home range overlap (ESM 1). If the database provided a range of values instead of a single data point for any of the variables, we used the median for posterior statistical analyses. To calculate D we used kernel home range sizes when available; otherwise we used that estimated by other methods (i.e., minimum convex polygon or grid cell).

Defensibility Index

We calculated the defensibility index (D) for each howler monkey group. This value describes the likelihood that a group will encounter its own home range boundary as it moves within it on an average day: D = d/d′, where d is equal to the average daily path length and d′ is equal to the diameter of a circle with the area equal to that of the observed home range (Mitani and Rodman 1979). A territorial species will theoretically have a D index >1, which represents the ability to cross the full width of the home range during a day of normal travel.

Statistical Analyses

We used analysis of variance (Kruskal–Wallis), followed by pairwise comparisons using Tukey and Kramer (Nemenyi) post hoc tests, to examine differences in D across the eight howler monkey species. We then ran linear mixed effect models (LMMs) to investigate how the different independent variables (study area, group size, howler monkey density, percentage of fruit in the diet, and forest type) affected D, with species fitted as a random factor. We eliminated an outlier point that was two standard deviations below the mean and thus the database was finally equal to 63 groups. We also discarded home range overlap from the LMMs because of the very low sample size (N = 20) to compare across different species and predictors. Therefore, we compared the relationship between home range overlap and D using a Spearman rank test. As we found high autocorrelation between group size, number of females, number of males, and number of immatures, we used only group size for analyses, as it had the greatest sample size. We applied natural logarithmic transformations to all covariates, except proportion of fruit, which we arcsine square root transformed. Prior to LMM analyses we also standardized all independent variables by first subtracting the mean from each value and then dividing by its standard deviation.

Because information criteria–based approaches require complete cases (i.e., no missing data) in order to compare among different candidate models, we employed a multiple imputation approach (ESM 2) to handle missing data while running LMMs (Nakagawa and Freckleton 2011; van Buuren and Groothuis-Oudshoorn 2011). The multiple imputation method fills in missing values ensuring that imputed data values are still plausible, as they are drawn from a distribution specifically designed for each missing value. When used together with information criteria approaches, multiple imputation provides consistent parameter estimation and confidence intervals that fully incorporate uncertainty due to missing data in a lambda score (Rubin 1987). We used predictive mean matching (PMM) as the imputation method, which considers a scenario of data missing completely at random to estimate 20 plausible imputed datasets (m = 20) based on 99 bootstrap replicates of the mean (Morris et al. 2014). PMM fills in missing values randomly from the donors’ closest matches, then randomly samples one of the donors, and finally returns the observed value of the match (Morris et al. 2014).

We fitted independent variables in all possible combinations to create a list of explanatory models, with species fitted as a random factor to control for data dependency and interspecific variance. Then we ran each of the explanatory models with each of the 20 imputed databases and calculated the mean (SE) Akaike information criterion (AIC, Tabachnick and Fidell 2007) for each model following Nakagawa and Freckleton (2011). We considered the models within an AIC distance of ≤2 (∆AIC ≤ 2) to indicate substantial explanatory support and models with ∆AIC between 2 and 7 to indicate moderate support (Burnham and Anderson 2002). Then, from this set of a priori models (∆AIC < 7) we used further AIC related statistics that consider model selection uncertainty (Snipes and Taylor 2014):

$$ \mathrm{Akaike}\ \mathrm{weight}\ (w)=\frac{\exp \left(-\frac{1}{2}\Delta {AIC}_i\right)}{\sum_{r=1}^R\exp \left(\frac{1}{2}\Delta {AIC}_i\right)} $$
$$ \mathrm{Evidence}\ \mathrm{ratio}\ (ER)=\frac{w_{\mathrm{best}}}{w_i} $$
$$ {\mathrm{Log}}_{10}(ER):{LER}_i={\log}_{10}\left({ER}_i\right) $$

where AICi is the individual AIC score for each of the models, R is the total number of models, r is the model being considered, wbest is the weight of the best model, and w is the weight of the other individual models. Following Kass and Raftery (1995), we ranked models as minimal, substantial, strong, and decisive to correspond approximately to log evidence ratios (LERs) of 0, 0.5, 1, and 2, respectively. We reported all factors included in the best-supported models as applicable contributors to model outcome, including those that did not give significant P values. We further investigated the factors in the models by assessing their respective estimate coefficients, standard errors, and λ values (i.e., the proportion of total variance that was attributable to the missing data). We visually inspected residual probability plots for deviations from homoscedasticity and normality. We ran all statistical analyses in R 3.3.1 (R Development Core Team 2008), using the libraries lme4 (Bates et al. 2012), MICE (van Buuren and Groothuis-Oudshoorn 2011), and MuMIn (Bartoń 2016).

Ethical Note

All the data we used are from published papers. The authors declare they have no conflicts of interest.

Results

All howler monkey species scored as territorial (D ≥ 1) with the exception of Alouatta palliata (D = 0.96, ± SD = 0.49; Fig. 1). Nonetheless, there was considerable variation in the mean D values across species (Fig. 1). We found significant differences in D across the eight howler monkey species (Kruskal–Wallis χ2 = 20.3, P = 0.005, df = 7). In particular, A. caraya (Kruskal–Wallis post hoc test, P = 0.007) and A. guariba (P = 0.006) had significantly higher D values than A. palliata. Other pairwise post hoc comparisons were not significant.

Fig. 1
figure 1

Mitani and Rodman defensibility index (D) across different Alouatta species based on a review of 64 studies of 8 howler monkey species. Open circles indicate data points. The horizontal black lines and cross represent the mean and median values, respectively. Boxes extend to data points that are less than 1.5 × IQR away from 1st/3rd quartile. Whiskers extend to minimum and maximum values.

Full size image

Mean home range overlap (± SD) with neighboring groups was 30.2% (± 5.5%) and was highly variable (range 0% to 87%). We found no correlation between home range overlap and D (rs = 0.17, P = 0.45, N = 20).

We first obtained 13 potential explanatory LMM models: three with strong support (∆AIC < 2) and 10 with moderate support (∆AIC = 2–7, Table I). There was substantial evidence in favor of the model with the lowest AIC (i.e., the full model) relative to the other models (LER > 0.5), with the exception of the next two best models (model 2 and model 3 hereafter) with delta AIC < 2. The best three models together received 69% of the weight of all models considered; thus we selected these models as the final best explanatory models. Visual inspection of residual plots of the best three models did not reveal deviations from homoscedasticity or normality. In all models, group size had a positive effect on D, whereas study area had a negative effect (Fig. 2; Table II). Percentage of fruit in the diet had a weak negative effect on D. The full model also included forest type among its explanatory variables for D, but the standard errors of the effect of different forest types were too large to conclude any clear pattern. Howler monkey density was the parameter with the strongest effect on D in all the three models. Both model 2 and model 3 contained study area, group size, and howler monkey density with effects on D similar to those of the full model (Table II).

Table I Summary of Akaike information criterion (AIC) results for a priori models (∆AIC < 7) explaining the defensibility index in howler monkeys for different models based on 20 imputed datasets (m = 20)
Full size table
Fig. 2
figure 2

Effects of area, group size, howler monkey population density, percentage of fruit in diet, and forest type on the Mitani and Rodman defensibility index (D). The distributions of the predictors (covariates) are displayed with vertical short lines on the horizontal axis and 95% confidence intervals are shown in gray around the regression line. For forest type (the only categorical predictor) dots indicate the expected mean D.

Full size image
Table II Summary of estimates (β) and standard errors (SE) of individual parameters explaining range defensibility index (D) in howler monkeys considering the 20 imputed dataset (m = 20)
Full size table

Discussion

We found considerable variation in range defensibility both within and among howler monkey species. This is, not all of the group home ranges were equally defendable from potential intrusion. These results highlight the fact that range defensibility can be expressed differently not only across closely related species, but among groups of the same species. In addition, these results suggest that howler monkeys are theoretically able to move in a manner that allows for the defense of a home range from rival neighbors.

Alouatta palliata was the only species that could theoretically be nonterritorial according to D. Bergman et al. (2016) suggested that A. palliata is a species with high levels of intragroup competition and low levels of intergroup competition, which could minimize their interest in defending territories against neighbors. Mitani and Rodman (1979) also found a similar pattern of D, with the variance spanning zero—from territorial to nonterritorial—in black-and-white colobus monkeys (Colobus guereza) and gray langurs (Presbytis entellus). A. palliata live in larger social groups than other howler monkey species, composed of several males and females (Arroyo-Rodríguez et al. 2008; Asensio et al. 2007; Dunn et al. 2015; Estrada et al. 2006). Large groups could deplete food sources faster than small groups (i.e., scramble competition, Dias and Rangel-Negrín 2015), which could lead to larger home ranges (Isbell 1991) and in turn smaller D scores. However, indication of territoriality in A. palliata was borderline, with a high intraspecific variability (mean D = 0.96, SD = 0.49) for precluding any definitive territorial behavior in the species. Moreover, LMM analyses, taking all the members of the genus together, found a positive relationship between D and group size while controlling for the effect of species. This general relationship supports the resource holding potential hypothesis (Parker 1974) for Alouatta, which states that groups with higher fighting ability (i.e., larger group sizes) will generally exhibit a higher ability to defend a territory (e.g., Mosser and Packer 2008; Wilson and Wrangham 2003).

The percentage of fruit in the diet was negatively associated with range defensibility in howler monkeys, although it had a weak effect and a relatively large part of its variance was attributable to missing data, and thus interpretations should be taken with caution. The contribution of fruit to the diet should theoretically have a positive impact on range defensibility as fruit is a defendable resource (Brown 1964; Clutton-Brock and Harvey 1977; Davies and Houston 1984; Emlen and Oring 1977). We should also expect a high D with a frugivorous diet owing to the relatively high energy provided by such a diet, which permits more movement in comparison to that based on leaves (Milton 1980). The observed weak effect of frugivory on range defensibility in howler monkeys could be related to the relationship between resource value and fruit abundance not being necessarily strong and linear (Grant et al. 1992; Maher and Lott 2000). In addition, the stronger effect of howler monkey density on D hindered a clear relationship between diet and range defensibility. It is also possible that there are effects of food abundance and distribution not accounted for in such a relationship owing to potential differences in seasonal territoriality not detected in the overall large temporal scale of the studies (e.g., Golabek et al. 2012; Marler and Moore 1989).

Study area had a negative effect on the defensibility index of howler monkeys. A large area should provide more available territories compared to a small area (cf. Fretwell 1972; Pen and Weissing 2000). In turn, a scenario of groups with territories in a large area would create little overlap between neighboring groups, and thus low need for territorial patrolling. This is related to the ecological concept of the ideal free distribution (Fretwell 1972). Interestingly, groups living in very small and isolated fragments (< 10 ha) in this study may have not had neighbors close by, but still displayed a high D, which might give a false impression of high territorial defense. Instead, high defensibility scores in these cases might have been simply related to the necessity of moving constantly across a small area in search of limited resources.

The density of howler monkeys was the parameter with the strongest effect on D in the three supported models, with greater D at higher densities. Animals deplete food more quickly when more individuals are in the landscape, and greater D values are probably the result of a greater need to move in search of food. In addition, howler monkeys might display territorial behavior in landscapes with high densities of individuals in response to the greater intergroup competition, as occurs in ring-tailed lemurs (Jolly et al. 1993). Kitchen et al. (2015) reasoned that Alouatta palliata may actively compete with neighboring groups when living at high densities. Holzmann et al. (2012) suggested that A. guariba make lower frequency calls at higher population densities than lower population densities. Similarly, Shonfield et al. (2012) observed that red squirrels (Tamiasciurus hudsonicus) were more likely to produce territorial vocalizations when surrounded by higher densities of conspecifics.

Range defensibility in howler monkeys varied among socioecological scenarios (i.e., large group size, small area, frugivorous diet, high population density) that favored competition. To better understand the territorial nature of howler monkeys, detailed analyses of whether different species and groups have site-specific aggression and range exclusion assessed through direct behavioral observations are needed. Studies using D have not typically addressed the limitations of associating D directly with territoriality, and a high D is not necessarily an unconditional proxy of territoriality. For example, Mitani and Rodman (1979) also found that some primate species presenting high D values were in fact not territorial, as they did not engage in aggressive encounters. Nonetheless, as the term defensibility denotes, D has an inherent nexus to the spatial organization of territorial maintenance, as it indicates whether mobility permits area monitoring. Thus, our results on howler monkeys’ range defensibility still suggest a territorial gradient depending on the competitive circumstances of each particular site and the cost–benefits of being territorial.

Animals recognized as territorial show high plasticity in their degree of territoriality in response to variation in food abundance and distribution (e.g., Motacilla alba: Zahavi 1971; Turdoides bicolor: Golabek et al. 2012), habitat quality (e.g., Canis lupus: Kittle et al. 2015; Cordylochernes scorpiodes: Zeh et al. 1997), mates and breeding sites (several carnivore, ungulate, and primate species: Grant et al. 1992; Passerella iliaca: Alcock 2005), seasonality (Sceloporus jarrovi: Marler and Moore 1989), or various mixed factors (Pteronura brasiliensis: Leuchtenberger et al. 2015; Pan troglodytes: Moore et al. 2015; Pomatostomus ruficeps: Sorato et al. 2015). Further, a species generally assumed to be nonterritorial such as the pigtailed macaque (Macaca leonina) can show territorial patterns under particular circumstances that force high levels of competition (José-Domínguez et al. 2015). There is evidence that territoriality does not need to be a species-specific trait (Carpenter and MacMillen 1976; Powell et al. 1997). Animals may present “facultative territoriality,” which means that they may exhibit territoriality only under particular social and environmental circumstances such as high population density and small area available (Chapman and Fedigan 1984; Kinnaird 1992). Therefore, a continuum that goes from defending to not defending a territory might reasonably exist in howler monkeys, which would adjust to particular socioecological factors that require or favor area defense.