Split-plot designs are a powerful tool in experimental research, combining two factors with different sizes of experimental units. They're perfect when one factor is harder to change or apply than the other, allowing for efficient resource use and precise measurements.
These designs have a unique structure with and subplots, each with their own factors and randomization. This setup leads to two error terms and requires special analysis methods, but it offers valuable insights into and interactions between factors.
Experimental Units and Factors
Experimental Units and Levels
Top images from around the web for Experimental Units and Levels
partR2: partitioning R2 in generalized linear mixed models [PeerJ] View original
Whole plot experimental units are the larger units to which the levels of the whole plot factor are randomly assigned
Subplot experimental units are the smaller units within each whole plot to which the levels of the subplot factor are randomly assigned
Levels of the whole plot factor are randomly assigned to whole plots while levels of the subplot factor are randomly assigned to subplots within each whole plot
Having two sizes of experimental units (whole plots and subplots) is a key characteristic of split-plot designs
Factors and Levels
Whole plot factor is the factor whose levels are randomly assigned to the whole plots
Levels of the whole plot factor are applied to the larger experimental units (whole plots)
Example: In an agricultural experiment, the whole plot factor could be irrigation method (drip or sprinkler)
Subplot factor is the factor whose levels are randomly assigned to the subplots within each whole plot
Levels of the subplot factor are applied to the smaller experimental units (subplots) within each whole plot
Example: In the same agricultural experiment, the subplot factor could be fertilizer type (organic or synthetic)
Design Structure
Restricted Randomization
Split-plot designs involve restricted randomization where the randomization of the subplot factor is restricted within each whole plot
Randomization occurs at two levels:
Whole plot factor levels are randomly assigned to whole plots
Subplot factor levels are randomly assigned to subplots within each whole plot
This restricted randomization structure is a defining feature of split-plot designs and distinguishes them from other designs like randomized complete block designs
Hierarchical Structure and Error Terms
Split-plot designs have a hierarchical structure with subplots nested within whole plots
This hierarchical structure leads to two different error terms:
Whole plot error: Variation among whole plots that have been assigned the same level of the whole plot factor
Subplot error: Variation among subplots within a whole plot that have been assigned the same level of the subplot factor (also called split-plot error)
Having two error terms is another key characteristic of split-plot designs and affects the analysis and interpretation of results
Effects
Main Effects
In split-plot designs, there are two types of main effects:
Main effect of the whole plot factor: Compares the mean responses between levels of the whole plot factor, averaged across all levels of the subplot factor
Main effect of the subplot factor: Compares the mean responses between levels of the subplot factor, averaged across all levels of the whole plot factor
Interpreting main effects in split-plot designs requires considering the hierarchical structure and error terms
Example: The main effect of irrigation method would compare the mean yield between drip and sprinkler irrigation, averaged across both fertilizer types
Interaction Effects
Interaction effect assesses whether the effect of one factor depends on the level of the other factor
In split-plot designs, the interaction of interest is usually between the whole plot factor and the subplot factor
Determines if the effect of the subplot factor is consistent across all levels of the whole plot factor, or if it varies depending on the whole plot factor level
Interpreting requires examining the pattern of mean responses across all combinations of factor levels
Example: A significant interaction between irrigation method and fertilizer type would suggest that the effect of fertilizer type on yield differs depending on whether drip or sprinkler irrigation is used