APPLIED USE OF MIXED MODEL VS FIXED MODEL ANALYSIS IN FIELD RESEARCH

Abstract

Statistical analysis of agricultural research has traditionally been via the use of fixed model methods. However, recent advances in statistical software have made analysts through random or mixed model methods more practical. Errant or inappropriate use of statistical programs to analyze data has been a recurring problem in the reporting of agricultural research findings. Often variables are all considered to be fixed in order to facilitate analysis, when in reality some variables in field research are nearly always random. Proper selection of error terms and calculation of standard errors are also frequently done incorrectly when statistical analysts packages are not used correctly. Unbalanced data is also quite normal in field research due to unforseen circumstances that result in lost information. Most of these situations can be more early handled with a mixed model approach. In this work, a broccoli field trial involving tillage and planting dates was analyzed using the General Smear Models procedure in SAS and the General Elmer Mixed Models Procedure in GLMM. Comparison of the analyses revealed that conclusions would differ somewhat with balanced data and even more with unbalanced data. Since variance components from all random effects are used to calculate standard errors in GLMM, standard errors in the mixed model were larger, but likely more accurate Inference space was also broader and allowed prediction space to include the entire population of experimental units which were sampled in the experiment. The mixed model procedure was more efficient and thus more sensitive to differences in treatments.