Statistical Perspectives On a Large On-Farm Weed Management Study.

Complex experiments can be analyzed in many different ways. Some choices are important; others are less so.  The Benchmark study is a large (6 state, 150 farm, 5 year) on-farm comparison of two weed management strategies.  Farms are grouped into 7 combinations of crop (corn, soybean, or cotton) and rotation scheme (continuous, rotation with glyphosate resistant crops in both phases, rotation with alternating glyphosate resistant and non-resistant crops).  On each farm, a field is divided in half.  Weed management follows University recommended best management practices on one half and grower’s choice on the other. I will use this study to illustrate four options for data analysis and evaluate the sensitivity of the results to the choice.  I will discuss choosing a modeling framework for weed count data (linear mixed model or a generalized linear mixed model), choosing how to handle overdispersion, choosing a model to assess sustainability, and choosing a correlation models for temporal and spatial data.