MIXED MODELS FOR ANALIZING THREE FACTORS AGRICULTURAL EXPERIMENTS WITH A STATISTICAL SOFTWARE

Abstract

PROC GLM of SAS® is the most used procedure for the analysis of variance in agricultural research, but it was designed for fixed models. The objective of this work was to capacitate SAS®´s users on the use of mixed models for the analysis of three factors agricultural experiments. Data came from literature and were analyzed by PROC GLM (fixed and random models) and PROC MIXED. Three factors and their interactions were: Legume (V), Phosphorus (P) and Weed Control (W). The design was CRB (three blocks, BLK) and the response variable was seed-yield (kg/plot). Routines of PROC GLM for fixed and mixed models and PROC MIXED were presented. Under mixed models, V and P were fixed and W was random effect. Degree of freedoms were adjusted by Satterthwaite procedure. With PROC GLM fixed model, BLK, V*W and P*W were p<0,01 and with mixed model only was BLK, but F-Values were minors. W and V*P*W variances were -0,59 and -0,22, respectively. With PROC MIXED, no one source of variation was p<0,05 (F and Z Wald Test). W variance was cero while V*W and P*W variances were minors. V and V*P standard errors (SE) were higher with PROC MIXED and were equal for random effects in all procedures. It was concluded that PROC MIXED gives adequate hypothesis test and SE of mean, so its utilization is recommended.