Improved experimental design and analysis for long-term experiments

This paper addresses inadequacies in the way most long-term experiments (LTEs) are conducted and analyzed. The standard design under which LTEs are usually conducted involves a fired start, establishing all plots in the study in the same year. This design is shown to be inadequate for the purpose of testing and estimating the time x treatment (TAT) interaction, which is generally the primary interest in a LTE. This inadequacy occurs because the repeated measures taken on every plot are all influenced simultaneously by the same random environmental conditions, the effects of which are confounded with the fixed effects of interest. No statistical analysis can completely separate the fixed effects from the random nuisance effects, although added assumptions about the shape of trends across time or covariates to describe the random effects can sometimes be helpful. An alternative experimental design, the staggered-start design, has been used to alleviate this confounding by establishing plots from different blocks in successive years, but proper analysis of this design has not been presented. A correct analysis of the staggered-start design is determined and presented. The analysis is applied to hypothetical data from a staggered-start design whose true means are known, and it is shown to do a much better job of estimating these means than any methods applied to data from the standard design. A staggered start should be considered instead of a fixed start for all future LTEs.