Designing robust field trials is a cornerstone of sound agricultural research and plant breeding. The choice of design isn’t just a statistical exercise—it directly affects the reliability of your results, the precision of your estimates, and ultimately, the success of your selection decisions.
Want to discuss your project needs? We would love to talk to you. Please Contact us.
This post offers a practical overview of several key experimental designs commonly used in field trials, including when and why you might use each.
Common Pitfalls to Avoid
- Ignoring field variability when assigning plots.
- Forgetting to include check entries in p-rep or augmented trials.
- Choosing complex designs without adequate statistical support.
Why Experimental Design Matters
In real-world field conditions, variation is inevitable. Soil fertility, moisture gradients, and microclimate effects can skew trial results unless controlled with proper design and layout. Experimental designs help control for this variation and ensure that differences observed between treatments (genotypes, hybrids, fertilizers, etc.) are due to real effects—not random noise.
1. COMPLETELY RANDOMIZED DESIGN (CRD)
When to use: In uniform environments, such as greenhouses or growth chambers.
- All treatments are randomly assigned across all plots without blocks.
- Simple to implement, but rarely used in field trials due to lack of spatial control.
2. Randomized Complete Block Design (RCBD)
When to use: Moderate number of treatments and moderate field variability.
- Field is divided into blocks (replications), each containing all treatments.
- Controls for variation across one direction (usually a gradient like soil fertility or slope).
✅ Pros: Easy to analyze; good balance between control and simplicity.
⚠️ Cons: Can become inefficient if you have many genotypes.
3. Alpha Lattice Design
When to use: Large number of entries (e.g., >20) with replication.
- Entries are grouped into incomplete blocks within replications.
- Offers better control of local variability than RCBD.
- Requires more complex randomization and analysis.
✅ Pros: Improved precision for large trials.
⚠️ Cons: Requires statistical software like ASReml, metan, or statgen for analysis.
4. Partially Replicated (p-rep) Design
When to use: Early-generation testing, large nurseries, or trials with limited seed.
- Only a subset of genotypes are replicated; others are unreplicated.
- More genotypes can be screened without increasing trial size.
✅ Pros: Cost-effective and practical under seed-limited conditions.
⚠️ Cons: Unreplicated genotypes have lower precision.
Example: In a 200-entry trial, you may replicate 40 elite checks and early selections twice, while 160 new lines appear once.
5. Augmented Designs
When to use: Preliminary yield trials where most genotypes are unreplicated.
- Control/check genotypes are replicated; test entries appear only once.
- Common in early breeding stages.
✅ Pros: Efficient use of limited space or seed.
⚠️ Cons: Comparisons between test entries can be less reliable.
Choosing the Right Design
Design | Best Use Case | # of Entries | Replications | Software Needed |
CRD | Greenhouse/pots | Few | ≥3 | Basic stats |
RCBD | Field with mild gradient | Moderate | ≥2 | Excel, R |
Alpha | Field trials with many entries | >20 | ≥2 | R, SAS, ASReml |
p-rep | Early-gen testing | >50 | Mixed | R (prDiGGer, DiGGer) |
Augmented | First-year field screening | >100 | 0–2 | R, agricolae |
Common Pitfalls to Avoid
- Ignoring field variability when assigning plots.
- Forgetting to include check entries in p-rep or augmented trials.
- Choosing complex designs without adequate statistical support.
Conclusion
The right experimental design can elevate your trial from a rough estimate to a reliable decision-making tool. Whether you’re running early-generation nurseries with hundreds of entries or focused trials on elite lines, aligning your design with field conditions and objectives is key.
Further Reading and References
- Federer, W. T. (1956). Augmented designs. Biometrics.
- Patterson, H. D., & Williams, E. R. (1976). A new class of resolvable incomplete block designs. Biometrika.
- Piepho, H. P., Williams, E. R., & Michel, V. (2006). Experimental design and data analysis for multi-environment trials. Crop Science.
- R packages: agricolae, metan, prDiGGer
Have questions or want a design suggestion for your trial? Contact us.