Which is the best experimental design in animal choice tests?

Abstract

Preference experiments are common in animal behavior research. In most cases, these are approached using completeblock design experiments, after which data are processed with ANOVA/T 2 or similar statistical tests. Using recent developments in statistics, we compared dierent experimental designs for testing animal preferences:a complete block design; and two incomplete block designs: paired (homogeneous), and heterogeneous. We simulatedexperiments and compared the performance of each design under the following scenarios: (i) a simple one withmoderate differences in attractiveness between stimuli, (ii) a glaring case, where one stimulus is much more attractivethan the others, and (iii) a heteroscedasticity case, where stimuli differ in attractiveness. Finally we asked how anincreasing number of stimuli affect the precision of estimations. The paired design was the best performing experimental design overall, whereas the heterogeneous one alsoperformed acceptably, sometimes outperforming the paired. On the other hand, the complete design presented anacceptable performance only when a low number of replicates are considered and when testing a small number ofstimuli. The main disadvantages of complete designs were that they provide limited information about the lowerranked stimuli, and ill manage situations of heterogeneity in the variance of the stimuli. We also note that the complete designs lack flexibility and may be more complex than generally thought. Weconclude that incomplete designs may prove the most reliable option for understanding animal preferences. Suchdesigns offer the important advantage of allowing a sequential addition of new stimuli in the comparisons.