Novel Applications of Multivariate Methods for Exploring Personality in African Elephants

Shilo Kimberly Felton, Western Kentucky University


Investigators have shown that elephants exhibit consistent individual differences in behavior by rating elephants using personality adjectives. These adjectives, however, are not based on pre-defined measurements of the behaviors performed. Instead, they are based on the observers’ interpretations of an animal’s behavioral patterns, therefore making them subject to observer bias. Furthermore, elephants have a capacity for learning; thus, they may alter their behavioral patterns over time. This behavioral plasticity in itself might be a way of measuring consistent behavioral differences among individuals. With this in mind, I approached elephant personality as a multivariate problem. I used behavioral observations collected from female elephants in Addo Elephant National Park, South Africa. Instead of grouping behaviors into subjective categories prior to analysis (as is often done in studies of elephant behavior), I used ordination methods to determine which correlations among behaviors were important for defining personality. Ordination methods were performed on matrices of the behavior data set and on subsets of behaviors for each age class. I calculated the angular differences among major axes of covariation from the ordinations of subsets to determine if the behaviors that defined personalities differed by age class. I also defined personalities by centroids (in multidimensional space) for non-metric multidimensional scaling (NMDS) scores of each individual and dispersion of NMDS scores for each individual as a measure of behavioral plasticity. I analyzed the effects of plasticity and age on personality of individual elephants using a non-parametric multivariate analysis of variance. Major axes of covariation were not well defined and therefore not useful in describing differences among groups. The interaction of age and behavioral plasticity did have a significant effect on the personalities of individuals as defined by ordination centroid scores. This suggests that incorporating plasticity may be a helpful measurement in quantifying consistent behavioral differences among individuals.