Biomathematics & Statistics Scotland
In plant variety trials, a complete replicate often occupies a large area of land and so is unlikely to be spatially homogeneous. Explicit inclusion of spatial effects in the analysis can complement the traditional design approach to handing this problem.
The Figure shows a variogram of the grain yield for a wheat trial after allowing for variety effects, indicating both a trend in yield across rows and a zigzag pattern across columns. A check on farm practice showed that the zigzag pattern was due to the direction of harvesting; east-west and west-east on odd and even numbered columns respectively. Two combine harvesters were also used. When these row and column effects were included in the analysis, residual variance was reduced by 40%. Accounting for the spatial effects increased the precision of the experiment and the power to detect true differences among the wheat varieties.
A variogram of the grain yield (t/ha) for a wheat trial, showing mean squared difference in residuals as a function of the number of rows and columns separating plots.
Further details from: Christine Hackett
Article date 2007
