Submitted to: Journal of Animal Behavior
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: December 8, 2008
Publication Date: March 1, 2009
Citation: Kramer, M., Weldon, P., Carroll, J.F. 2009. Composite scores for concurrent behavious constructed using canonical discriminant analysis. Journal of Animal Behavior. 77(3):763-789.
Interpretive Summary: In discrimination tests of chemical stimuli, the number of concurrently measured behaviors (dependent variables), often correlated, can lead to a complex analysis. Reducing the dimensionality of the data prior to the analysis typically facilitates both the statistical analysis and its interpretation. We propose the use of canonical discriminant analysis to construct a weighting system for concurrently measured behavioral variables. Using canonical discriminant functions as a base, we demonstrate a method of creating a composite score of the measured behaviors, such that each animal’s “score” is reduced to one or two numbers. This method combines the various behaviors measured in a way optimal for discriminating among the stimuli. In addition, it indicates the appropriate dimensionality of the scores, reflecting the number of latent axes represented by the set of stimuli presented. We also demonstrate some graphical tools useful to display these kinds of results.
Behavioral experimentation is a valuable tool to improve our understanding of invertebrate and vertebrate organisms that affect plant, animal and human health. Interpreting and understanding complex behavioral data, where measurements are taken on a large number of variables, can require a complex analysis. A method to combine scores from many variables in a statistically acceptable way has been wanting. Based on data that we collected on the effects of five chemicals on lone star ticks, Amblyomma americanum, a serious medical and veterinary pest, we developed and applied a new method for combining concurrently measured behaviors on each animal into a single score per animal. The method allows the user to greatly reduce the complexity of the statistical analysis. This method will be of immediate value to behavioral scientists and biostatisticians, and ultimately the agricultural and general community who benefit from research.