SPLIT-PLOT DESIGN: ANALYSIS DOLDRUMS COMPARING GENSTAT, SPSS AND R

Abstract

A modelled split plot experiment involving Completely Randomized Design of the main (whole) plots replicated three times was analysed using three statistical software packages namely Genstat, SPSS and R. The results were compared in terms of accuracy of output and ease of analysis. The data set was also analysed semi-manually to validate the results of the software outputs. Genstat and R produced identical result to the semi-manual analysis indicating they were correct. SPSS however, produced erroneous test results even when the correct linear model was specified. The correct output could however be obtained by specifically instructing SPSS on how to carry out the test. This experiment demonstrates the need for experimenters to be in charge of every aspect of an experiment from design to analysis leaving no part to the whims of statistical software. It also demonstrates that SPSS on its own is incapable of producing the correct result based on the linear model for this experiment. SPSS and Genstat are menu-driven and may be easier to learn than R which is mostly command based.