Log Ratio Methodology for Analysis of Compositional Data: A case of Olfactometer Bioassay Data from Insect Behavioural Studies

Abstract

Researchers are often times confronted with compositional data in insect choice studies. The choice of a statistical method to model this type of data is always not obvious. In this study, three approaches for analysis of compositional data from choice tests made by the predatory parasitoid Cotesia sesamaie Cameron in a four-arm olfactometer was explored using centered, additive and isometric log ratio transformations. Oviposition induced plant volatiles (OIPVs), herbivore induced plant volatiles (HIPVs), and two controls were tested in the four-arm olfactometer. The response variable measured was time spent in each field of the olfactometer, when the time to observe the insect was restricted to 12 minutes. The data generated in this study was compositional, thus it conveys exclusively relative information and has a constant sum constraint such that standard statistical methods of analysis (ANOVA, t-test), cannot be used on this data. This study therefore explored the log ratio methodology advocated by Aitchison (1986). CLT, ALT, and ILR log ratio transformations were then performed using CoDaPack statistical software. Using this methodology, mean differences in olfactometer response of female parasitic wasp, Cotesia sesamiae to OIPVs, HIPVs, and control were computed. These findings imply that the CLR transformation is probably the best choice for processing raw compositional data prior to analysis by standard statistical methods. These results revealed that the, parasitic wasps spent much time in olfactometer arm with OIPVs, followed by the olfactometer arm with HIPVs and lastly spent least time in the control arm of the olfactometer. More studies need to be conducted using the log-ratio methodology on olfactometer bioassay data from a different species of parasitic wasps.