 |
|||
| Volume 1, Issue 1, 2007 | |||
|
Vicki K. Bentley-Condit, Grinnell College, bentleyc@grinnell.edu Timothy S. Hare, Morehead State University, t.hare@morehead-st.edu |
|||
|
Introduction Numerous animal behavior studies have examined environmental influences upon behavior. This research has included the influence of resource availability on grouping patterns, (e.g., Wrangham 1980; Johnson et al. 2002), resource availability on behavior (e.g., Crocket & Janson 2000; Archie et al. 2006), and population density upon behavior (e.g., Judge & de Waal 1997; Benson et al. 2006). However, with a few notable exceptions (Altmann 1979; Rhine & Westlund 1981; Robinson 1981; Janson 1990), very little animal research, particularly that with primates, has attempted to systematically examine spatial context in social behavioral analyses. This lack with primates is particularly noteworthy given their abilities to discern, distinguish, and differentiate between microhabitats (Gonzalez-Kirchner 1999), personal space (Ihobe 1989), and spatial relations (Wilson 1972; Bowe 1984). Here, we use Geographic Information Systems (GIS) technology and inferential spatial statistics to explore the spatial dimensions of animal social behavior using a captive baboon group as our subjects. We, thus, simultaneously examine the applicability of these methods in animal social behavior research and address a gap in the primate behavior literature. Specifically, we examine lactating female baboons’ space use using an integrated GIS and spatial statistical approach. Lactating females were chosen as subjects because the data were collected as part of a larger study of baboon infant development (see Bentley-Condit 2003). Examining space utilization with a captive population living as a large group in a large enclosure (see Methods) offers the advantages/caveats of 1) a circumscribed area amenable to detailed mapping, 2) no predation, 3) reliable resources, and 4) ample social interaction opportunities. Although this captive group’s habitat removes several of the environmental factors that influence an individual’s use of space, it presents itself as a unique opportunity for assessing the applicability of GIS technology. GIS offers several advantages over traditional methods of spatial analysis and provides a robust approach to animal behavioral data exploration as it has the ability to store, manipulate, and display data linked to locations (Longley et al. 1999). It is being used extensively by geographers, geologists, and archaeologists and is beginning to enter the realms of biologists and animal behaviorists (e.g., Wieczorek & Hanson 1997; Ganskopp et al. 2000; Beauplet et al. 2004). However GIS use in animal behavior studies has been limited, primarily focused on mapping habitats, calculating home ranges, and examining resource distribution, and is often linked with Global Positioning System (GPS) technology (e.g., Campbell & Wentz 1993; Zinner & Torkler 1996; Formica et al. 2004; Scholz & Kappeler 2004; Karavas et al. 2005). Nevertheless, with GIS one can conceivably examine any behavior(s) or interaction(s) by sex, age, status and other factors as they occur in space in ways that were previously impossible. It allows more accurate pinpointing of behaviors and more powerful statistical analysis of the spatial context of behavioral data – particularly those data that violate the assumptions underlying traditional statistics. GIS resolves problems encountered using non-spatial techniques as it reveals the presence of spatial effects and measures the extent to which data values in close proximity vary together. As with non-normally distributed data, the presence of spatial effects in data produce biased and overly precise results from common statistical tests such as correlation coefficients and standard regression. This problem is alleviated through the use of inferential spatial statistics. An integrated GIS/spatial statistical approach stands to reveal previously unseen or vague patterns such as: 1) an individual animal’s “personal” space, 2) the spatial dimension of particular activities by an individual, and 3) the interaction between an individuals’ personal characteristics and her use of space. We anticipate that GIS-based spatial statistical behavioral studies will become more common as the process of data collection and processing is streamlined through the use of hand-held computers to record both map and behavioral data. 1A) As part of this selective use of space, individuals’ ranks should influence their behaviors’ spatial context as it is well established that dominance rank affects behavior among many animals (e.g., Schino 2001; Vervaecke et al. 2005; Arakawa 2006). Due to priority of access to resources related to high rank, high-ranking females should have greater access to, and be able to restrict others’ access to, “preferred” areas of the habitat. Thus, we predict there will be a correlation between females’ ranks and their behaviors’ spatial context. 1B) Similarly, individuals’ ages are predicted to affect the spatial context of their behavior. That age also affects behaviors is well established (e.g., Baker 2000; Reader & Laland 2001; Pelletier & Festa-Bianchet 2004). Whether age, within the age-category of adult, is specifically correlated with the spatial aspect of behavior has yet to be investigated. However, it stands to reason that age might affect how or where a given female performs a given behavior in space. Therefore, we predict that the spatial context of young vs. middle aged vs. older females’ behavior will vary. 2) Finally, although baboons may transport their food (in hands or cheek-pouches) before consuming it, there should be an obvious spatial context of ingestion in this group with most occurring in areas near food hoppers, all located in Zone A of the habitat (see Fig. 1). Rest, social, and infant behaviors are predicted to be correlated with Zone B (see Fig. 1). It is this zone that contains most of the wooden and metal climbers and concrete culverts available to the population and captive primates generally prefer to orient themselves near objects (Menzel 1967). Therefore, we predict that these behaviors should be associated with objects of orientation; travel should occur throughout the corral and not be space-specific. Methods Study Site, Subjects, and Data Collection
The first author collected
data for this project at the Southwest Foundation for
Biomedical Research in San Antonio, TX during the final two
months of a six-month study of infant development in 2001.
The subjects are a subset of the olive baboon (Papio
anubis) breeding population at this site, housed in a
large corral (»2.4
hectares; Fig. 1). Shade and protection from inclement
weather are provided by the angled corral walls and by
apparatuses located primarily in Zone B (see below). The
population was established in 1979 (Goodwin & Coelho 1982)
and there are currently 500+ individuals in the corral
(population density
»20,600
per km2). Females have very few direct
matrilineal kin ( Eight food hoppers and five water spigots are located at uneven intervals around the periphery of the corral (Fig. 1). The baboons receive a basic regime of LabDiet® occasionally supplemented by fruits and vegetables and have ad libitum access to water. LabDiet® is provided twice daily with a heavier feeding in the a.m. (≈ 0700 hours) and a lighter feeding in the p.m. (≈1600 hours). Food is thus available in particular locations at particular, and for relatively limited, times. The entire group cannot feed at any one hopper and the food at each hopper is depleted within a couple of hours. Consequently, although predictable temporally and in terms of size, the resources/food in this environment are spatially and temporally clumped.
Figure 1. The Corral and its Features. In addition to the individual continuous focal observations (Altmann 1974) collected as part of the previously mentioned infant development study, space/map data were collected on a subset of 25 lactating females from late-September through late-November 2001. The first author worked inside the corral on foot and collected data once per day between approximately 0645 hours and 1500 hours, 6 days per week (Monday-Saturday). Observations of each female were scheduled so that all 2-hour blocks of time (i.e., 0700-0900, 0900-1100, etc.) were approximately equally represented (see Bentley-Condit 2003) to avoid observational biases.
Those females (n=8)
near the end of their observation cycles (i.e., with less
than 20 remaining observations) when the space project began
were later rejected due to insufficient representative data
leaving a sample of 17 females – each with ≥21
maps/observations ( During the continuous individual focal, all of the female’s movements, point locations, routes followed, and times were mapped via pen and paper on a corral drawing each time she moved. The corral has 12 numbered sides – much like a large pie with 12 “slices” (Fig. 1). The corral space was divided into three zones for mapping. Zone A, an outer ring extending inward from the walls for »15m, contains all food, water, and windows but very little else. Zone B, extending inward from the first for another »25m, is where most climbers and concrete culverts are located. Zone C, a central ring extending another »40m to the center point of the corral, contains little but has a popular, slightly elevated rocky area and some logs. Each corral “slice” was thus divided into these three zones for a total of 36 areas (i.e., 1A, 1B, 1C, …12C) within which a female could be mapped at numerous points and at any time. The maps were subsequently digitized into a GIS database where a digitized blueprint of the corral with its features had been used to establish exact coordinates. Though not available to the researcher during data collection, as mentioned previously, a hand-held computer would substantially streamline this process and allow for even higher data analysis resolution. The maps, combined with their associated behavioral focal data, provide information on precisely where individuals were located (i.e., points within zones within “slices”), length of time there, behaviors while there, and with whom they interacted and, thus, allow us to assess spatial utilization and behaviors’ spatial distributions. To facilitate analyses, the behavioral data were collapsed into five broad categories: rest, feed, travel, social, and infant. The first four categories are self-explanatory and generally used in animal time-budget studies and the fifth category, infant (i.e., mother’s interactions with her infant), was added due to the focus of the overall project being infant development. (The time budget subsequently established for these females’ is: rest=52.12%, travel=18.37%, feed=14.13%, social=8.8%, infant=6.59%.) Females’ ranks were determined based upon their interactions with all other females following the methodology of Bramblett (1981) and subsequently grouped into high, middle, or low rank at naturally occurring breaks in the dominance matrix. Females were similarly grouped by their ages into three age categories [young (range=5-9yrs), middle (range=11-14yrs), and old (range=16-18yrs)] where natural breaks occurred between the groups. Each dominance rank group contained members from each of the three age ranges; likewise, each age range had females representing all three dominance rank groups. GIS Inferential Statistics As the GIS statistical data analyses are somewhat complex, a detailed description follows. However, GIS is currently integrating these techniques into easier to use packages. We initially investigated the raw spatial data using descriptive statistics that indicated asymmetrical distributions for all variables. The data were logged to yield a closer approximation to normality (Dowdy & Wearden 1985). With the outline and features of the corral defining the context (Fig. 1), our initial mapping of “all data points”, “all data points by individual”, and “all data points by behavior by individual” revealed a general distribution of data points with local areas of high and low densities (Fig. 2A). We explored several ways to facilitate comparisons between behaviors and individuals including using the 36 polygonal zones to create choropleth maps (see Fig. 2B and Fig. 2C) where the data were normalized to account for the different spatial area sizes of the corral zones.
Figure 2. Examples of Thematic Maps. 2A - Total data points for all focal females including 75% kernel density contours. 2B - Choropleth map, normalized by area, of total time of social behavior for female 8145. 2C - Choropleth map, normalized by area, of time by behavior for female 10667. To reduce subjectivity in the analysis of locational data, we used several types of spatial statistics. Nearest neighbor analysis (NNA) characterizes the degree of clustering or dispersion of point locations in a region (Kaluzny et al. 1998) and produces a global statistic (range= 0-2; 0=perfect clustering, 1=randomness, 2=perfect dispersion) with a significance test that indexes the degree of clustering/dispersion. However, NNA does not account for variation in time at each location so we also utilized spatial autocorrelation. Spatial autocorrelation is the lack of independence between elements in a spatial dataset (Kaluzny et al. 1998) and provides a means for assessing the degree of clustering in the time spent at particular points in the corral. A positive spatial autocorrelation indicates females spend longer periods of time at points in particular zones; a negative spatial autocorrelation indicates the times females occupy points in particular zones are dissimilar. We normalized the autocorrelation data to account for the different sized areas represented by the corral zones and utilized Moran's I and Geary's C statistics for assessment (Anselin 1995).
As the assessment of our predictions depends on comparisons
between variables, we used kernel density estimation methods
to compare pairs of point distributions with attached
continuous variables (Kie et al. 1996). Kernel density
estimation is a nonparametric statistical method for
estimating probability densities where points can have
attached continuous values or weights. It has been
incorporated into some animal home range studies (Worton
1987; Taulman & Seaman 2000) and has larger values in areas
with more observations and lower values in areas with few.
It describes the probability of finding an individual in any
one place. The two-dimensional kernel density estimator is
in the following general form where
The most useful output from kernel density estimation is a two by three matrix of kernel density probability with 75% contour lines (Fig. 2A) that delimits where the individual can be expected to be found 75% of the time while engaged in that activity. Seventy-five percent contours were a meaningful compromise between 50%, which represented too small an area, and 95%, which represented too large of an area to be informative.
Finally, we implemented a bivariate comparison between
paired kernel density estimates using a bootstrap hypothesis
test of equality to test the significance of the
correlations between pairs of individuals or behaviors
(Efron & Tibshirani 1998). To compare two density estimates
(
The textual output contained the smoothing parameter and the p-value of the test; the graphical output, a 2x2 matrix of maps, provided a comparison between data sets (Fig. 3) and can be presented as contours (Fig. 3D). The hypotheses of the bootstrap hypothesis test of equality were:
The density surface of the interaction between individuals was based on the following equation which represents the standardized difference between the two individuals’ kernel density estimates divided by the bivariate variance (see Bowman & Azzalini 1997:1114-115).
In tandem, the matrix of maps described above highlights particular characteristics of behaviors’ spatial dimensions.
Figure 3. Examples of Kernel Density Comparisons of High and Low-Ranked Females. 3A and 3B - Lighter areas are used the most by that particular rank for all behaviors. 3C - Interaction of the ranks in their use of space with lightest areas used by High Ranked, darkest by Low and gray areas used equally by both. 3D - Positive and negative, two and four standard deviations, of the density estimate of the interaction between the two ranks. Positive depicts High Ranked areas, negative the Low Ranked areas, and the space between indicates areas used equally by both. Results Per Prediction 1, across behaviors, each individual consistently uses a subset of the corral area. NNA results indicate moderate clustering for all behaviors combined and particular behaviors independently for all of the focal females except female 9599 (Clark & Evans NNA “all behaviors/all females”: = 0.79, Z=17.51, P£0.0001; Clark & Evans NNA “all behaviors/ Female 9599=0.99, Z=0.15, P=0.43) with the highest level of clustering present in feeding (Clark & Evans NNA “feeding/all females”: =0.59, Z=18.61, P£0.0001). Thus, individual females “cluster” their behavior in particular areas – feeding here, resting there, etc. Moran’s I and Geary’s C indicate that each female has a low to moderate level of positive spatial autocorrelation for total time and by particular behaviors (again with the exception of Female 9599), i.e., longer periods of time at points in particular areas (Moran’s I “all females/all behaviors”=0.437, Z=6.282, P≤0.0001; Geary’s C “all females/all behaviors”=0.519, Z=-6.445, P≤0.0001). Finally, the comparison of kernel density estimates between areas indicates that the space consistently used by each female is different from that used by other females with only eight of the possible 136 different pairings of females revealing overlapping kernel density estimates and a P>0.05. Per Prediction 1A, there are consistent differences in the use of space by the focal females according to their ranks. Kernel density estimates indicate that space used by the high, mid, and low-ranking females for all behaviors is consistently different and all bootstrap tests are highly significant (bootstrap pair-wise comparisons: High to mid = 0.00000000000153, P≤0.0001; High to low=0.00000000000232, P≤0.0001; Mid to low=0.000000000003242, P≤0.0001). This pattern can be seen in Figures 3C and 3D comparing high to low-ranking focal females. High-ranking females consistently use areas to the left (north and west) while low-ranking females consistently use areas to the right (south and east). Not shown is that mid-ranking females are primarily found in the area between the positive and negative second standard deviations of the high and low-ranking females (i.e., in the middle of the north and west and extending slightly into the south and east). Thus, not only do females reliably use a subset of the available space but the locations of those areas vary by their ranks. Per Prediction 1B, focal females’ use of space also varies by age group. Kernel density comparisons indicate the space used by the three age groups (young, middle, old) for all behaviors differ significantly (bootstrap pair-wise comparisons: Young to middle=5.04, P£0.0001; Young to old=3.87, P£0.0001; Middle to old=4.73, P£0.0001). Similarly, comparisons for each individual behavioral category, with the exception of feeding for young and old females, were significantly different (bootstrap pair-wise comparisons: P£0.001; Young to old “feeding”=1.43, P=0.144). This pattern can be seen in Figures 4C and 4D comparing middle to old aged focal females. The middle age females are using areas to the north (upper left) while the old females are using areas to the south (lower left and right). The young females use areas more to the east (far right), primarily between the positive and negative second standard deviations shown in Figure 4D. Given the size of our sample, we compared the rank group and age group kernel density results for independence. Seven of the nine bootstrap pair-wise comparisons show the two distributions to be drawn from different populations (e.g., Young Age/Low Rank=0.00000000000112, P=0.001; Middle Age/High Rank=0.00000000000146, P£0.0001). However, there is distribution overlap between Middle Age/Low Rank (bootstrap pair-wise comparison=0.0000000000009, P=0.117) and between Old Age/High Rank (bootstrap pair-wise comparison=0.00000000000054, P=.283). These data indicate less variation between females by age group than between females by rank, i.e., there is more spatial overlap between differently aged females than there is between differently ranked females.
Figure 4. Examples of Kernel Density Comparisons of Middle and Older Aged Females. 4A and 4B - Lighter areas are used the most by that particular age for all behaviors. 4C - Interaction of the ages in their use of space with lightest areas used by Middle Age, darkest by Old and gray areas used equally by both. 4D - Positive and negative, two and four standard deviations, of the density estimate of the interaction between the two ages. Positive depicts Middle Age areas, negative the Old Age areas, and the space between indicates areas used equally by both. Per Prediction 2, there is a correlation between feeding and the locations of the food hoppers. Feeding by all sampled females occurs in four small areas that are closely associated with six of the eight food hoppers. Two hoppers, 6A and the first hopper at 8A, were not often utilized. The second part of Prediction 2 is partially supported. While the distributions of rest, social, and infant behaviors do include the enriched Zone B, they are not evenly distributed throughout (Fig. 2A, 75% kernel density estimate contours) and are associated with the preferred Zone A food hoppers. The north-south central axis of the corral and Zone C are the least used areas; the western half of the corral is the most commonly used for all behaviors (Fig. 2A). As shown above, all females consistently use these broad areas but each female exploits a unique smaller space within the broad areas (see Prediction 1). Finally, it is important to understanding the spatial dimension of these subjects’ behaviors to note two basic behavioral patterns. First, while there is a relatively high rate of female-female agonism (i.e., agonistic events per female per observation hour: X±SD=2.12 ± 0.849, N=444), the rate of agonistic events received by the focal females is low (i.e., agonistic events received per female per hour: X±SD= 0.664 ± 0.651, N=132). Second, females engage in affiliative events at a higher rate (affiliative events per female per hour: X±SD=3.287 ± 2.268, N=613) and affiliative events occur at a significantly higher rate per hour than do agonistic events received (t test: t32=4.57, P<0.01). The spatial context of behavioral patterns can be seen in how individuals in this group use different areas. The data demonstrate the statistical tendencies of area use by both female and behavior. As behaviors necessarily occur in a spatial context, use of space is an important, yet often ignored, dimension of behavior. Given the high population density of this group and the lack of matrilineal support networks, females should attempt to focus their behaviors spatially such that they can meet their basic needs, minimize contact and potential altercations with those females with whom they do not have affiliative relationships and promote access to those with whom they do. The patterns seen in the behavioral data of affiliative events occurring at a significantly higher rate than do agonistic events received appear to support this interpretation. The spatial dimension of behavior is affected by both a female’s age and rank. In this group, rank appears to have a stronger influence than age. Just how particular areas come to be established as preferred by the rank groups is likely due to history/tradition (Wilson 1972). Given that the corral could be divided in numerous ways and females’ requirements still met, that high-ranking females use the north-west quadrant and low ranking the south-east is, in this habitat, environmentally arbitrary. However, among other animal populations, similar spatial dominion could have an impact on survival and/or reproduction. While numerous studies of primate behavior have commented upon the abilities of higher ranked individuals to influence others’ access to resources, our research indicates that it may now be time to pay closer attention to the spatial dimension of individual’s and categories of individuals’ behaviors. The behaviors performed by the subject females occurred in relatively circumscribed areas of the corral. Ingesting was the most spatially limited behavior – surprisingly focused in four relatively small areas of the corral. All other behaviors are correlated with the areas of Zone B that are contiguous with the preferred feeding sites in Zone A. As shown in Figure 2A, the focal females’ behaviors are not randomly distributed in space. Rather, there is a statistically predictable pattern to the spatial distribution of behaviors – by female, by the female’s rank and age category, and by the behaviors themselves. Finally, our research may offer some implications regarding how the spatial distribution of behavior can impact inter-individual relationships. Females in this group lack extended matrilines. Instead, they cluster with like-ranked females and engage in social interactions at a level similar to that observed in wild populations between closer kin. Females in this group appear to be making the best of a less-than-ideal situation through their spatial arrangement. The implications of these data are that how and with whom one interacts may be strongly influenced by relative spatial arrangements – whether in circumscribed areas or in the wild. These ideas deserve further examination. In sum, these data tell us that we need to examine, and look for patterns in, not only what our subject animals do but also where they do what they do. A systematic inclusion of the spatial dimension variable could ultimately lead us to a broader understanding of animal social behavior and social organization. Although currently somewhat labor-intensive, it is the use of GIS technology and the associated inferential spatial statistics that made the above detailed spatial context analyses possible and that will allow us to rigorously examine these issues in other populations. GIS thus stands poised to explicate the previously under-represented spatial dimension of animal behavioral research. Acknowledgements This project was funded by grants from the Committee on Faculty Support at Grinnell College to VB-C and a Mellon Post-Doctoral grant to TH. VB-C is grateful to the SFBR for their permission to conduct behavioral research inside their breeding corral and for their continued support of her research. Special thanks to Linda Brent, Thomas Butler, Bill Cummins, Scott Chambers, and other current/past members of the Department of Laboratory Animal Medicine and to Thomas Moore, Grinnell College, for reading a draft of this manuscript. Crockett, C. & Janson, C. 2000. Infanticide in red howlers: female group size, male composition and a possible link to folivory. In: Infanticide by Males and its Implications, (Ed. by C. van Schaik & C. Janson), pp. 75-98. Cambridge: Cambridge University Press.
Dowdy, S. & Wearden, S. 1985. Statistics for Research. 2nd ed. New York: John Wiley & Sons.
Efron, B. & Tibshirani, R. 1998. An Introduction to the Bootstrap. Boca Raton: Chapman & Hall.
Formica, V., Gonser, R., Ramsay, S. & Tuttle, E. 2004. Spatial dynamics of alternative reproductive strategies: the role of neighbors. Ecology, 85, 1125-1136.
Judge, P. & de Waal, F.B. 1997. Rhesus monkey behaviour under diverse population densities: coping with long-term crowding. Animal Behaviour, 54, 643-662. Kaluzny, S., Vega, S., Cardoso, T. & Shelly, A. 1998. S+ SpatialStats: User's Manual for Windows and UNIX. New York: Springer.
Karavas, N., Georghiou, K., Aranoutsou, M. & Dimopoulos, D. 2005. Vegetation and sand characteristics influencing nesting activity of Caretta caretta on Sekania Beach. Biological Conservation, 121, 177-188.
Kie, J., Baldwin, J. & Evans, C. 1996. CALHOME: a program for estimating animal home ranges. Wildlife Society Bulletin, 24, 342-344.
Longley, P., Goodchild, M., Maguire, D. & Rhind, D. 1999. Geographical Information Systems: Principles, Techniques, Management and Applications. New York: John Wiley & Sons.
Menzel, E. 1967. Naturalistic and experimental research on primates. Human Development, 10, 170-186.
|
|||
 |
=1.66)
with whom to associate due to infant culling. With unknown
paternity, the exact role of kinship cannot be assessed.
=35
per female). For these 17 lactating females (n=10
with ♂ infants, n=7 with ♀ infants) on which the
following analyses are based, there are 593 mapped 20-minute
focal observations (
denote
the data and 
denote
the bivariate smoothing parameters (Bowman & Azzalini
1997:6-7).
and
)
using a significance test we used the following equation
(see Bowman & Azzalini, 1997:108 for details):
| | Home | Contact Us | Editorial Board | Current Issue | Submission | |
|
|
|
© Copyright 2006, Scientific
Journals International. All Rights Reserved. |