agricolae (R) alternative
Browser Alternative to the agricolae R Package
The agricolae package by Felipe de Mendiburu is the de facto R toolkit for agricultural experimental designs. It powers the design generation and ANOVA workflows in countless plant-breeding labs and ICAR theses, and the post-hoc letter display (LSD.test, duncan.test, HSD.test) is the reference output that agronomists know. The package is open source, well documented and free.
agricolae's barrier is R itself. You need a working R install, the package install (which depends on klaR, AlgDesign and a stack of CRAN dependencies), and enough R fluency to write the design call, the model formula, the post-hoc call and the printout. StatVeda gives you the same family of designs as a browser tool with no install and no R: CRD, RBD, factorial RBD, split-plot, split-split-plot, strip-plot, Latin square, balanced lattice, alpha lattice (PBIB), augmented CRD and RBD, plus diallel, line by tester, generation means and path coefficient.
At a glance: agricolae (R) vs StatVeda
| Feature | agricolae (R) | StatVeda |
|---|---|---|
| Cost | Free (open source, GPL) | Free |
| Install | Requires R + agricolae from CRAN | Browser, no install |
| Skill needed | R syntax (design call + model + post-hoc call) | Click and paste |
| CRD / RBD / Latin square / factorial | Yes | Yes |
| Split-plot / split-split-plot / strip-plot | Yes (sp.plot, ssp.plot, strip.plot) | Yes |
| Balanced lattice / alpha lattice (PBIB) | Yes (design.lattice, design.alpha) | Yes |
| Augmented CRD / RBD | Yes (design.dau) | Yes |
| AMMI / stability | Yes (AMMI, stability) | Yes (AMMI, GGE) |
| Diallel / line by tester / generation means | Partial (lineXtester only) | Yes (full Griffing 1 to 4 + 3 / 5 / 6 parameter genmeans) |
| Path coefficient (G / P / E) | Partial (path.analysis is correlation-based) | Yes (G, P and E matrices) |
| Field randomisation | Yes (built into design calls) | Yes (separate Field Layout tool) |
| Post-hoc battery | LSD, Duncan, Tukey, Scheffe, Waller, REGW | LSD, DMRT, Tukey, Dunnett, Tukey-Kramer, Games-Howell |
| Word / PPT export | Via knitr / R Markdown (extra setup) | Yes (one click) |
| Reproducibility | Excellent (R script is the audit trail) | Lower (no script saved) |
Where agricolae (R) wins
- ✓R script is a perfect audit trail. Re-running the analysis a year later gives bit-identical results, which is what reviewers want.
- ✓Field randomisation is part of the design call, with seed control. Replicate a layout exactly by re-running with the same seed.
- ✓agricolae composes with the rest of the R ecosystem: ggplot2 for figures, lme4 for mixed-effects extensions, emmeans for advanced contrasts, knitr for the report.
- ✓The post-hoc battery is broader (Scheffe, Waller, REGW) than what most browser tools expose.
- ✓Open source and on CRAN. The function source is auditable line by line.
Where StatVeda wins
- ✓No R install. The lab Windows machine, the supervisor's laptop and the field tablet all run StatVeda.
- ✓Click and paste. No model formula to write, no post-hoc call to remember.
- ✓Word and PPT export with the F table, post-hoc letters and CV percent already laid out for the thesis chapter.
- ✓Cross-checked against R 4.6.0 (which uses agricolae internally for its agricultural-design fixtures) on FAO and Montgomery datasets to four or more significant figures.
- ✓Goes beyond agricolae on full Griffing diallel (methods 1 to 4) and three- / five- / six-parameter generation means with the Cavalli joint scaling test.
Migration path
- Keep your R script for any analysis you want a reproducible audit trail on. agricolae is the right answer there.
- Use StatVeda for the everyday runs: a one-off ANOVA, a thesis-chapter table, a quick check before you commit to the full R pipeline.
- If you have an existing agricolae model output, paste the same data into StatVeda and confirm F, MS and p match. They will agree to four or more significant figures because both engines use the same closed-form formulae.
- For the post-hoc, pick the same family in StatVeda's selector (LSD, DMRT or Tukey HSD) that you called in agricolae (LSD.test, duncan.test, HSD.test).
Worked example: Same RBD analysis: agricolae vs StatVeda
Five-treatment, four-block randomised block design from Gomez and Gomez (1984) Chapter 2 (rice variety yield trial).
In agricolae (R)
In R: model <- aov(yield ~ block + treatment, data=df); HSD.test(model, 'treatment', console=TRUE) returns the source-of-variation table, treatment means, the honest significant difference and the compact letter display, all printed to the R console.
In StatVeda
In StatVeda: paste the treatment x block grid into the RBD calculator, pick Tukey HSD under the post-hoc selector. The result panel shows the same source table (with eta-squared and omega-squared added), the treatment means with the same compact letter display, and a Word export of the table preformatted for a thesis chapter.
Note: F, MS and the post-hoc letter assignments will match agricolae to four or more significant figures because both engines run the same closed-form RBD ANOVA. The difference is in presentation, not in the numbers.
Honest footnote: what this comparison cannot tell you
- •If you already write R fluently, agricolae plus emmeans is more flexible than any browser tool. The reproducibility argument alone is enough.
- •agricolae composes with mixed-effects (lme4, lmerTest) and with custom contrasts in ways no browser tool can match.
- •StatVeda does not generate or save the analysis script. If your archive policy requires one, agricolae plus knitr is the cleaner answer.
References (agricolae (R))
Try StatVeda
The StatVeda calculator engines run entirely in the browser. No signup, no install, no data sent to a server.
Open the RBD calculator