Biometric Challenges with Long Term Experiments.

Long-term studies are essential to the charge of agricultural research, but their proper analysis requires proper experimental setup and recognition of inference space and error structure.  In all cases, long-term studies contain both fixed and random terms, are thus mixed models, and must be analyzed as such.  Use of inherently fix model analysis (e.g. PROC GLM) is wrong and may lead to biased estimates and erroneous conclusions for mixed models.  Similarly, when analyzing long-term studies it is common to see a failure of the error structure to meet the prerequisite assumptions of ANOVA.  This includes problem of heteroscedasticity and correlated errors as well as others.  Fortunately, numerous pieces of software now exist and they afford proper analysis of long-term studies.  For this presentation, a preliminary discussion on the establishment of long-term plots will be presented and then an analysis of an example data set will be shown along with SAS code.  The presentation will include a comparison to common, although incorrect, methods of analysis.