Abstract
Published results of four field experiments testing effects of acidic precipitation in field-grown soybeans have led to contradictory conclusions. This paper examines the experimental procedures and protocols of the four experiments which could have contributed to differences in both the reported crop yields and the interpretations or conclusions drawn from the experiments.
The most important difference among the experiments is in their statistical designs. Two of the field layouts used only one plot per treatment, providing replication only for plants, rows or sectors within plots, but not among plots. By using real data from a highly replicated experiment it is shown that with such deficient statistical designs treatment effects cannot be separated from other microenvironmental variables peculiar to a specific plot location.
The other two experiments were highly replicated. They were designed to detect differences of approximately 10% among treatment means. Type 1 error, a, was predetermined, and replication was sufficient to keep type 2 error low. This design was made possible because a preliminary experiment was available to estimate the expected components of variance. Despite differences in the procedures and protocols among the four experiments, it is primarily the quality of the experimental design which has determined their validity and relative utility for crop loss assessment.