Options for statistical analysis of unreplicated paired design drainage experiments

Field scale experiments are seldom replicated due to size constraints and limitations of other resources. This presents a difficulty in performing statistical analysis and interpretation of the results collected. In this paper we will evaluate water and nutrient effluent data from two fields, one managed under controlled subsurface drainage (CSSD) compared with an adjacent field under conventional free subsurface drainage (FSSD). The purpose of this paper is to compare approaches for statistical analysis of unreplicated paired design drainage experiments. The methods tested include the paired watershed design, a before-after control-impact paired (BACIP) design, time-series, and analysis using a beta and a gamma distribution. The assumptions used in the paired watershed and BACIP designs require the data to follow a normal distribution with identically and independently distributed errors, while the beta and gamma distributions do not require the assumption of normality. Analysis of the data set using a normal distribution was found to be inappropriate due to the nature of the data set which violates the assumptions required for analysis of variance; whereas, the analysis of the data using a beta distribution was found to produce the most appropriate results. The time series approach mimicked the physical treatments applied to FSSD and CSSD fields and led to a significant effect of CSSD compared to FSSD. Future work should focus on confirming that the beta distribution or time series are the best approaches for statistical analysis of field scale unreplicated paired design drainage experiments.