Comparación de la potencia y robustez del AMVAR de MR frente al correspondiente modelo mixto del AVAR con dependencia serial en el error

Abstract

Comparison of the power and robustness of AMVAR of repeated measures versus the correspondent mixed model of AVAR with the error including serial dependence. In this research two procedures of analysis for a random experimental desing with several consecutive measures for each experimental unit are analized and composed. To analize this kind of data a traditional mixed model of AVAR with the error structure modeled through ARMA processes can be used. The problem can be worked out as well from a more general perspective, using a multivariate repeated measures approach. Both approaches are compared using simulated data obtaining with Monte Carlo procedures. Special attention was paid to a) the robustness of model parameter estimates when some assumptions are violated, b) the power of statistical test for both procedures.