Longitudinal data analysis in animal experiments

Abstract

Longitudinal data analysis with repeated measures over time can be done in different ways: in experiments using the split plot design, with the animal as plots and time as subplots; through the analysis of multivariate models; and also by adjusting mixed models, which enables the use of different structures for the covariance matrix. The main problem involved in the analysis of repeated measures over time is related to the lack of randomization, because pairs of measurements taken along the time are correlated, which can invalidate the tests involving the time factor. To discuss the different forms of analysis, data were used from an experiment in which Santa Ines lambs were infected with 6,000 L3 Haemochus contortus and treated with condensed tannin. The source of tannin was the acacia extract obtained from Acacia mearnsii (commercially available). The data were analyzed using the software R 2.11.1 (2010). No significant effect was found (p > 0.05) of supplementation with tannin, acting on the body weight of animals. The methods used for data analysis showed similar results, which does not always occur. And considering the dependence of observations taken over time, recommended the setting of mixed models, which may involve growth curves or polynomial models that include the covariance matrix that best explains the data.