Determination of the Most Appropriate Covariance Structure for Data with Missing Observations in Repeated Measures Design

Sequential measurements taken from same experimental unit at different periods of time for a quantitative trait are named Repeated Measurement. In a repeated measures design with two factors (treatment and time), use of MIXED model offers an opportunity for describing various covariance structures (CS, UN, ANTE (1), AR(1), TOEPLITZ etc.) in analyzing data with/without missing observations instead of a Repeated ANOVA (classical approach) in the event of violation of spherity assumption. In the framework of MIXED modeling used for the repeated measures design, the aims of this study are to evaluate statistical validity of some assumptions relevant to this topic for available data set including missing observations and to get knowledge about selecting the suitable covariance structure for the data set. To achieve these aims, sample data on animal science were provided. The best covariance structure was selected on the basis of goodness of fit criteria such as AIC, BIC, and AICC. In conclusion, present results obviously reflected that in the case of violation of spherity assumption, use of MIXED modeling in repeated measures design was a good choice for defining ideal covariance structures for data set with/without missing observations.