Covariance structure assessment in multi‐level models for the analysis of forests rainfall interception data using repeated measures

Abstract

Repeated measures refer to multiple observations obtained in the same sampling plot or experimental unit. Such observations can be taken in a particular period and the same local space in many cases. In either case, repeated measures lead to correlated data. Statistical analysis using correlated data without an appropriate covariance structure would lead to Type I or Type II errors. This study used rainfall interception data from a coniferous forest in Mexico, May to September 2010, to develop a multi-level linear model by assessing different covariance structures. Our main objectives were to evaluate the implications of these covariance structures in tests of fixed effects over categories of basal area and estimate differences between means and standard errors of each test. Based on the numerical inspection of the estimated correlations, their graphic representation, values of the statistics of fit (AICC and BIC) and considering that it is desirable to model a covariance structure in a parsimonious way, we concluded that the best selection for the structure of covariance was “heterogeneous Toeplitz.” Therefore, F $$ F $$ -values of fixed effects and the estimates of differences between means and standard errors of each test based on the “heterogeneous Toeplitz” covariance structure were considered the most appropriate among the analyzed covariance models.