Introduction

It is not certain how the rhesus macaques got to Silver Springs, Florida. Some say that they escaped during the filming of Tarzan movies in the late 1930s. Others claim that during this same time period a “Colonel Tooey” who managed the Jungle Cruise boat rides along the Silver River intentionally released a small colony on an island in the river believing the macaques couldn’t swim and would therefore be contained. Rhesus macaques can swim, and they did so to the shores of the river. However it happened, there has been a free ranging population of rhesus macaques, a species endemic to Asia, inhabiting the riverbanks of the Silver River in Florida since at least 1938. During the 1980s researcher Linda Wolfe studied the makeup of this population and some of her field data is now publicly available.

The visualizations in this report represent a small network of rhesus macaques along the Silver River. Based on a paper published in 1987 by Wolfe and Peters on the subject, these data likely represent a small group along the north side of the river that was once part of a larger group, but because of a variety of circumstances, separated. This small group was observed by Wolfe for a three month period and co-occurences of group members along the river bank were recorded.

Inspiration

References for design came from Caleb Jones’s visualization of his LinkedIn network and Evelina Gabasova’s Star Wars networks. Caleb’s visualization is visually striking for its symmetry and strong use of color. Nodes on the graph are spaced appropriately without occlusion and yet still give the impression of spatial clustering. One issue with the graph, is that it doesn’t contain labels or any information regarding the nature of the network, so viewed outside of the blog post, it is difficult to interpret.

Evelina’s Star Wars network uses a lot of similar design principles as Caleb’s graph, however with the addition of labels and selective color, it becomes more intuitive to read and understand without additional information. One can quickly understand what network the graph depicts and identify nodes that play key roles.

Network visualization is being used more and more to understand non-human animal social networks and the paper The Neuroethology of Friendship by Brent, Chang, Gariepy, and Platt served as inspiration for how to successfully implement this strategy to understand relatively abstract biological phenomena such as friendship and it’s role in evolutionary fitness. In this paper the authors studied the social networks of several non-human primates as well as humans as one way to understand how friendship plays out in species, forms the shape of social networks, and ultimately may provide adaptive benefits to the individual and group. There is considerable neuroscience in this paper that falls outside the scope of its inspiration for the Silver Springs rhesus macaque visualization, however it is a good example of how network visualization can become an important component of scientific study.

Materials

In order to create this network the initial datasets were sourced from CASOS and UNICET. The files contained both node and edge tables and Microsoft Excel and OpenRefine were used to clean these data. Gephi was used for network statistical analysis and primary visualization and Adobe Photoshop was used to add titles and keys to the graphs.

Methods

This network started initially with only one dataset from CASOS. These data contained a node table that listed all of the members of the rhesus macaque group as numbers (1-20), and an undirected edge table containing information on group member co-occurences by the river bank. Each edge was weighted by the number of times group members appeared together over the three month observation period. Although rich enough to form the basis of a network visualization, these data were not detailed enough to be able to draw useful conclusions about the makeup of the group. The UNICET dataset was then used to fill in node attributes of an individual’s sex, rank, and age. The CASOS and UNICET node table data were combined in OpenRefine and Excel and then, along with the cleaned edge table, imported into Gephi.

Because of the small nature of the network, it was initially visualized in Gephi using the Fruchterman Reingold layout. This allowed for more space between nodes which was necessary as inspection of each node and lack of occlusion proves critical in the analysis of the visualization. Statistical measurements of average degree, network diameter, graph density, and modularity were calculated for the network and nodes were sized proportionally to their degree. Modularity was calculated with a radius of 1.75 which produced 6 sub-groups within the network.

Nodes in the first graph produced were colored by sex of the macaque so that users could easily view the ratio of male to female within the group. Node labels that consisted of the individual’s rank and age were placed over the nodes and sized proportionally. This graph can be seen below.

Although coloring nodes by sex is a quick way to ascertain the male to female ratio in the group, it was thought to be potentially misleading as users may interpret the colored groups to represent connectedness, which is incorrect. Therefore a second graph was produced, this time incorporating sex data into the node labels and coloring nodes according to modularity class. This graph can be seen below.

Finally, a third graph was produced using the Forced Atlas 2 layout in order to spatially separate nodes that have lower weights on each of their edges. Because of the small size of the network, default settings in this layout lead to an indecipherable graph with large amounts of occlusion so the scale was increased to 1500 and gravity lowered to 0.25. This allowed the nodes to separate, rather than unintelligibly cluster. This iteration can be seen below.

Results

From these graphs we can understand several aspects of this small group of free ranging rhesus macaques. First, based on the statistics generated in Gephi we can see that co-occurences on the Silver River bank is quite uniformly distributed throughout the group. The average degree of each node is 18.2, meaning each individual member of the 20 macaque group was observed at the river bank at least once with almost every other member. Along with this we can see that because the network diameter is only 2, all of the members of the group are connected to all other members by only two steps. Knowing these two aspects of the network, we would expect it to be very dense, which when calculated is found to be true. This network has a density of 95.8%, which means that almost all members of the group have been observed together at the river bank at least once.

In regards to the makeup of the network we can conclude that rank doesn’t have a particular bearing on connectedness as the highest and lowest ranking individuals have equal degree although they do fall in separate modularity classes. We can also see that sex does not seem to impact connectedness for the same reasons. Within the group of 20, there is a core sub-group consisting of 50% of the members, and 5 other sub-groups consisting of 20%, 10%, 10%, 5%, and 5%. Based on the Forced Atlas 2 representation of the network we see that there are several individuals in the group that are not as strongly pulled towards the others based on the weight of there edges. For example, although individual F/15/5 has a degree of 18, it has pulled away from the network because of the lower weight of the edges. From this we can conclude that this individual visited the river bank fewer times than others (or was not observed when she did visit), however did so with an average diversity of other individuals from the group.

Future Directions

There are several things that could be done to expand upon the insight gained from this visualization. If more information was available regarding the makeup of the group, sub-groups based on modularity could not only be identified, but defined to help understand the dynamics of the group. Along with this, specific interactions beyond co-occurence (i.e. grooming, eating, mating, etc.) could be coded and recorded which would give a more robust model of the nature of the group.

Other areas of study could be gathering a more recent dataset (Wolfe’s data were from the mid 1980s) of the same group to provide information on how group dynamics change over time, or to contrast the Silver Springs macaques with macaque groups located in their native habitat. Wolfe mentions that the behavior of the Silver Springs macaques aligns with observations of other macaques observed in their endemic environments, however a more in depth study of these networks could provide information on how non-native environments effect macaque social structure.