Premium
Comparison of Experimental Designs Used to Detect Changes in Yields of Crops Exposed to Acidic Precipitation 1
Author(s) -
Evans Lance S.,
Thompson Keith H.
Publication year - 1984
Publication title -
agronomy journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.752
H-Index - 131
eISSN - 1435-0645
pISSN - 0002-1962
DOI - 10.2134/agronj1984.00021962007600010021x
Subject(s) - randomized block design , latin square , total sum of squares , mathematics , yield (engineering) , precipitation , field experiment , statistics , row , zoology , chemistry , horticulture , agronomy , explained sum of squares , non linear least squares , residual sum of squares , biology , physics , meteorology , rumen , food science , database , fermentation , computer science , thermodynamics
A comparison of experimental designs used to detect changes in yield of crops exposed to simulated acidic rain was performed. Seed yields were determined from field‐grown soybeans ( Glycine max L .) exposed to simulated rainfalls in which all ambient rainfalls were excluded by automatically moveable rainfall exclusion shelters. When all eight, 4 ✕ 4 , latin squares were used, simulated rainfalls of pH 4.1,3.3, and 2.7 exhibited yields 10.6,16.8, and 23.9%, respectively, below yields of plants exposed to pH 5.6. The linear component of pH treatment differences showed a highly significant decrease in yield (p < 0.001). The eight latin squares analyzed as individual experiments yielded erratic conclusions. However, when all eight latin squares were combined a satisfactory evaluation can be made. The low probability value for differences among latin squares (p < 0.005) indicates that field locations are a legitimate source of stratification. Sources of variation are also highly significant for both pooled among rows within squares (p = 0.0028) and the pooled among columns within squares (p = 0.0168) indicating that the two stratifications within each latin square was doubly effective in reducing the residual error term. The relative efficiency of the latin square to a randomized complete block design, and a completely randomized design was 132‐146 and 162%, respectively.