Comparison of mixed and fixed effects in the analysis of a split plot design in an experiment with Guinea Mombaza (Megathyrsus maximus cv. Mombaza)

Resumen

Models of fixed and mixed effects were compared in the analysis of a split plot design in an experiment with Guinea Mombaza (Megathyrsus maximus cv. Mombaza), to know the effectiveness of the mixed model as an analysis alternative. The experiment was carried out with a split plot design, considering the harvest months of Guinea Mombaza as the main plot and one control and two levels of nitrogen fertilization (0, 50 and 75 %) as subplots. Variance-covariance structures tested were Toeplitz, variance component, compound symmetry, Firts order autoregressive and unstructured variance structure in the mixed model. For a better data fit, Akaike, corrected Akaike and Bayesian information
criteria were taken into account, and the lowest value was considered. Variance-covariance structures with the best fit were Toeplitz, component of variance, and unstructured. Probability values of interaction of the main effects were similar for both models, as well as the mean square of the error and standard errors of mean differences. Selection criteria, which were properly fitted to analyzed variables, allowed to know the variance-covariance structure. Mixed model provided standard errors of appropriate mean differences, so it is proposed as an analysis alternative. The use of the mixed model avoids calculation of standard errors based on the formula established for this purpose.
Key words: variance-covariance, information criteria, random effect, PROC MIXED