Agriculture & Future Food Systems · Article
Controls, replication and randomisation
Design a comparison of two crop practices so that field position does not quietly become the treatment being tested.
Preparation and related reading
Shared Foundations
Related articles
This article helps you
- Identify a treatment, comparison and experimental unit.
- Distinguish replicated plots from repeated measurements inside one plot.
- Explain the different jobs of blocking and random assignment.
A field comparison has more than one difference
A grower tries a new sowing practice on one half of a vegetable field and keeps the usual practice on the other. At harvest, the halves differ. The observation is useful, but the practice also differs by field position. Soil, shade or drainage could contribute to the result.
A better comparison begins with a specific question and an outcome that can be measured. Here, the question concerns two sowing practices for the same crop. Equipment settings and growing requirements retain their own topics; this article concerns how to compare their results.
Name the comparison
A treatment is the practice or condition deliberately compared. The usual practice can serve as the control: a defined baseline, not necessarily untreated soil. Record what differs and keep other management as comparable as the question permits.
The experimental unit is the unit assigned a treatment. If a practice is assigned to an entire plot, that plot is the unit. Individual plants measured within it help describe its outcome but do not automatically become independently assigned treatment replications.
Replication reveals variation among units
Repeating each treatment across plots provides information about variation under the trial's conditions. One unexpectedly favourable plot is then visible alongside others. Repeated readings of the same plot can improve measurement, but they do not create additional treatment assignments.
Choose replication with the variability, effect of interest, available area and analysis in mind. There is no universally adequate small plot count. The invented three-block diagram explains the structure; it is not a recommended sample size or a power calculation.
Compare within suitable field groups
Blocking groups plots with relevant similarities, such as position along a soil or slope gradient. Both treatments are represented within each block, so the contrast can be made locally rather than placing every new practice at one end of the field.
Blocks should reflect the actual source of variation. They do not make the whole field homogeneous or eliminate every possible difference. The shared estimation and uncertainty Foundations support deeper analysis; field applications still require a design appropriate to the question.
Do not always give one practice the favoured position
Random assignment within each block avoids deliberately or habitually allocating one treatment to the better-looking position. It is an allocation procedure, not a demand that the resulting layout look evenly alternating. The diagram shows one possible realised assignment.
Random assignment and random sampling are different. Assignment chooses which unit receives a treatment; sampling chooses what observations are collected. Both need a documented method. Randomisation reduces systematic assignment bias but does not guarantee perfect balance of every unmeasured feature.
One possible blocked allocation
Read the diagram as text
The invented trial contains three blocks with two assigned plots each. Treatment position is randomised separately within each block.
Parts
- Block 1: usual
- One independently assigned plot
- Block 1: new
- Paired comparison within this block
- Block 2: new
- One possible realised random order
- Block 2: usual
- Both treatments occur in each block
- Block 3: usual
- The plot remains the experimental unit
- Block 3: new
- Plant readings within it are subsamples
Connections
- Block 1: usual → Block 1: new: compare
- Block 2: new → Block 2: usual: compare
- Block 3: usual → Block 3: new: compare
Keep design and results connected
Map plot identities and allocations before implementation, record departures and retain separate plot results. Neighbouring strips can influence each other, so suitable boundaries and buffers require attention. A treatment spilling into another plot changes what was actually compared.
A difference in plot means is a descriptive result. Its uncertainty and causal interpretation depend on assignment, independent units, measurements and appropriate analysis. Do not add a significance label or a universal farming conclusion merely because the new practice's average is higher.
Worked examples
Count assignments, not sampled plants
An invented trial has three blocks, each with one usual-practice plot and one new-practice plot. Two plants are measured per plot. How many plot units and plant observations exist?
- Two plots in each of three blocks gives 3×2=6 assigned plot units.
- Two plants per plot gives 6×2=12 plant observations.
- Each treatment has three assigned plots, not six independent plant-level treatment replications.
Six plot units and twelve within-plot plant observations. The appropriate analysis retains that hierarchy.
Describe the paired differences
Invented matched block outcomes for the new minus usual practice are +0.10,+0.05 and −0.03 kg/m². What is their arithmetic mean?
- The sum is 0.10+0.05−0.03=0.12 kg/m².
- 0.12/3=0.04 kg/m².
- The small teaching set provides no stated statistical model, standard error or confidence interval.
The descriptive mean difference is +0.04 kg/m²; significance and wider applicability are not established.
Check your understanding
Must a control mean no cultivation or no water?
Show answer and reasoning
No.
It is the defined comparison relevant to the question, often the usual practice.
Do twenty plants in one assigned plot make twenty treatment replications?
Show answer and reasoning
No.
Within-plot observations are nested within that assignment.
Does random allocation guarantee identical soil in all plots?
Show answer and reasoning
No.
It addresses assignment bias; residual variation remains.
Key ideas
- Define the experimental unit and comparison before collecting results.
- Use replication, blocking and random allocation for their distinct purposes.
- Keep observed differences separate from supported inference.
References
Sources supporting the explanations and examples in this article.
Sustainable Agriculture Research and Education (SARE) Outreach2 sources
- NIST/SEMATECH e-Handbook: Randomized block designsNational Institute of Standards and Technology / SEMATECHTeaching resource