Split-Plot Experimental Design
Also known as: split-plot ANOVA, whole-plot sub-plot design, Bölünmüş Parsel Deseni (Split-Plot)
The split-plot design is a parametric experimental design that applies one factor to large whole plots and a second factor to subdivisions (sub-plots) within each whole plot. It was introduced by Frank Yates in 1935 to handle agricultural experiments where one factor — such as irrigation or tillage method — is difficult or impractical to change frequently, while a second factor can be varied more easily within the same plot.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use a split-plot design when your experiment has two factors and one of them is genuinely hard or costly to change across experimental runs (e.g., oven temperature, irrigation regime, industrial batch settings, or participant group in an educational study). The design requires at least two blocks (replications of whole plots), the same sub-plot arrangement within every block, and a continuous or ordinal response variable. Normality of residuals and homogeneity of variance within each error stratum should be checked. If too few observations per sub-plot group are available (fewer than about ten), a simpler randomized complete block design is more appropriate.
Strengths & limitations
- Accommodates practical constraints by allowing one factor to be applied at the whole-plot level without sacrificing valid inference on the other factor.
- Higher statistical power for the sub-plot factor and the interaction term compared to a completely randomized design of equivalent size.
- Directly estimates the whole-plot and sub-plot variance components, providing richer information about the sources of variability in the experiment.
- Widely applicable across agriculture, engineering, food science, clinical research, and behavioral experiments wherever a hard-to-change factor is present.
- Inference on the whole-plot factor is based on fewer degrees of freedom (whole-plot error), reducing power for detecting whole-plot effects compared to completely randomized designs.
- The design is invalid if the sub-plot arrangement differs across blocks, because the mixed model assumes a consistent structure.
- Analysis requires a mixed-effects model with two error strata; a standard ANOVA table from single-error software will produce incorrect F-tests for the whole-plot factor.
- Requires careful planning — post-hoc identification that randomization was restricted leads to inflated Type I error if the wrong error term is used.
Frequently asked
How is a split-plot design different from a two-way ANOVA?
A two-way ANOVA assumes complete randomization of all factor combinations and uses a single error term. A split-plot design has restricted randomization — one factor is randomized at the whole-plot level and the other within it — so it requires two separate error terms. Using a standard two-way ANOVA for split-plot data incorrectly pools these errors and produces wrong F-tests, usually anticonservative for the whole-plot factor.
Why does the whole-plot factor have lower power than the sub-plot factor?
Power depends on the degrees of freedom in the denominator of the F-ratio. The whole-plot error has fewer degrees of freedom (based on the number of whole-plot units, i.e., blocks times whole-plot levels minus one) than the sub-plot error. Fewer degrees of freedom means the F-distribution has heavier tails and a higher critical value, making rejection of the null hypothesis harder for the whole-plot factor.
What software can correctly analyze a split-plot design?
Any platform that fits linear mixed models will handle split-plot data correctly, because the design is a special case of the mixed model with two variance components. Commonly used tools are lme4 in R (with emmeans for post-hoc comparisons) and the Mixed Models procedure in SPSS. Standard fixed-effects ANOVA procedures are insufficient unless the software explicitly supports specifying separate error terms.
When should I switch from a split-plot design to a randomized complete block design?
If you have fewer than approximately ten observations per sub-plot treatment group, the split-plot design has insufficient power and the randomized complete block design is a safer choice. Also consider switching if the hard-to-change constraint disappears — for example, if you can fully randomize all factor combinations — because complete randomization is always statistically preferable when feasible.
Sources
- Yates, F. (1935). Complex Experiments. Supplement to the Journal of the Royal Statistical Society, 2(2), 181–247. DOI: 10.2307/2983638 ↗
- Montgomery, D. C. (2017). Design and Analysis of Experiments (9th ed.). Wiley. ISBN: 978-1119492443
How to cite this page
ScholarGate. (2026, June 1). Split-Plot Experimental Design. ScholarGate. https://scholargate.app/en/experimental-design/split-plot-design
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Completely Randomized DesignExperimental design↔ compare
- Hierarchical Linear ModelingStatistics↔ compare
- One-way ANOVAStatistics↔ compare
- Randomized Complete Block DesignExperimental design↔ compare
- Repeated-measures ANOVAStatistics↔ compare
- Two-Way ANOVAStatistics↔ compare