The split-plot design is an experimental design that is used when a factorial treatment structure has two levels of experimental units. In the case of the split-plot design, two levels of randomization are applied to assign experimental units to treatments1. The first level of randomization is applied to the whole plot and is used to assign experimental units to levels of treatment factor A. The whole plot is split into subplots, and the second level of randomization is used to assign the subplot experimental units to levels of treatment factor B.1,2 Since the split-plot design has two levels of experimental units, the whole plot and subplot portions have separate experimental errors2.
The split-plot design is used to analyze descriptive data when applying Analysis of Variance (ANOVA). This design tests significant differences among samples and also estimates variation due to panelist inconsistencies3. Samples evaluated by judges are considered to be the “whole-plot effect” and are placed at the top of the ANOVA table (Table 1). Judges themselves make up the “subplot effect” and are placed at the bottom of the table4,5. Trained descriptive judges are taught to evaluate products in the same way, and therefore, judges are generally treated as a random effect5.
If judges rate their perceptions of attributes differently from sample-to-sample, then there may be a potential for judge-by-sample interaction. If there is potential for this type of interaction, it can be included in the model as a source of variability separate from error. An F-test (F3) can show whether or not this interaction is significant4. If it is assumed that there is no panelist-by-sample interaction, this term can be pooled with error5.
If various judges scale attributes in different ways, then there may be an overall judge main effect. An F-test (F2) can show if the judge main effect is significant4.
Judge-by-sample interactions and judge effects should be minimized by effective panel training and maintenance. If judges do not fully understand how to scale terms, then they may scale in an inconsistent manner across samples. If each judge is using a different portion of the scale to rate attributes, then results will be inconsistent across panelists. Proper training and orientation can help clarify term definitions and evaluation methods in order to avoid these inconsistencies. Data obtained from well-trained panels can accurately show small differences among samples.
1 Ott, R.L. and Longnecker, M. 2010. An introduction to Statistical Methods and Data Analysis, 6th Ed., pp. 1095-1101, Brooks/Cole, Pacific Grove, California.
2 Kuehl, R.O. 2000. Design of Experiments: Statistical Principles of Research Design and Analysis, 2nd Ed., pp. 469-472, Brooks/Cole, Pacific Grove, California.
3 Lawless, H.T. and Heymann, H. 1998. Sensory Evaluation of Food: Principles and Practices, pp. 714-715, Aspen Publishers, Inc., Gaithersburg, Maryland.
4 Stone, H. and Sidel, J.L., 2004. Sensory Evaluation Practices, 3rd edition, pp. 128-133, Elsvier Academic Press, London.
5 Meilgaard, M.C., Civille, G.V. and Carr, B.T. 1991. Sensory Evaluation Techniques, 2nd edition, pp. 266-267, CRC Press, Boca Raton, Florida.
6 Chambers, E. Course notes in Descriptive Sensory Analysis, Kansas State University, Manhattan, KS, 2011.