Experimental Design & Analysis
Comprehensive module for academic researchers — 10 experimental designs with full ANOVA tables, post-hoc tests, effect sizes, APA output, and R verification.
DESIGN SELECTION GUIDE
| Design | When to Use | Controls | Key Output |
|---|---|---|---|
| CRD One-Way ANOVA | Homogeneous units, single factor with k levels | — | F-test, Tukey HSD, η², ω² |
| RCBD Block Design | One nuisance variable (e.g. batch, location, day) | 1 blocking factor | Treatment F, Block F, RE vs CRD |
| Latin Square | Two nuisance variables (e.g. row × column) | 2 blocking factors | Treatment F, row/col F |
| Factorial Two-Way ANOVA | Two (or three) factors; study interactions | — | Main effects, interaction, plots |
| Split-Plot | Hard-to-change factor (whole-plot) + easy factor (subplot) | Blocks | Separate F-tests for WP & SP |
| ANCOVA | Control a continuous covariate (pre-test score, age…) | Covariate | Adjusted means, slopes test |
| Repeated Measures | Same subjects measured at multiple time points | Subjects | Mauchly's W, GG/HF correction |
| Crossover (AB/BA) | Subjects receive all treatments in sequence; clinical trials | Period, carryover | Treatment, period, carryover tests |
| Nested/Hierarchical | Factor B levels differ within each level of factor A | — | Variance components, EMS |
| RSM / CCD | Optimize process; find best factor settings | — | Response surface equation, optimum |
FEATURES
Full ANOVA Tables
All SS, df, MS, F, p-values per source
All SS, df, MS, F, p-values per source
Post-Hoc Tests
Tukey HSD, Bonferroni, LSD, Scheffé, Duncan
Tukey HSD, Bonferroni, LSD, Scheffé, Duncan
Effect Sizes
η², partial η², ω², Cohen's f
η², partial η², ω², Cohen's f
Assumption Tests
Brown-Forsythe, Mauchly's sphericity, normality
Brown-Forsythe, Mauchly's sphericity, normality
Statistical Interpretation
Plain-language results summary
Plain-language results summary
APA Format
Publication-ready results paragraph
Publication-ready results paragraph
R Verification
Cross-check via WebR — real R 4.5.1
Cross-check via WebR — real R 4.5.1
Interaction Plots
Visualize factorial interactions & profiles
Visualize factorial interactions & profiles
📁 Demo Data
One-click realistic datasets for each design
One-click realistic datasets for each design
Which design? Three questions
The choice turns on three things and only three.
THREE QUESTIONS
Design Builder — Randomized Run Sheets
Build the experiment before you run it: a seeded randomization you can report and reproduce, the run order to take to the bench, the degrees of freedom the design will actually leave you, and — for the screening designs — exactly which effects are aliased with which.
DESIGN & RANDOMIZATION
🎲Why the seed matters: the run order below is drawn from it, so the same seed always rebuilds the same sheet. Record it in your methods section and the randomization becomes reproducible by anyone — which is the one part of a randomization that is normally lost. Leave the field blank and one is drawn for you and written back into it.
Power & Sample Size — ANOVA
The F test of each layout under the alternative, read in three directions: the power a sample gives you, the sample a target power needs, and the smallest effect a sample you already have could detect. Non-central F throughout; λ is printed on every answer, because that is where parameterizations differ.
QUESTION & LAYOUT
CRD — Completely Randomized Design
One-Way ANOVA: k treatments, experimental units randomly assigned to treatments. Use when units are homogeneous and no blocking is needed.
DATA & OPTIONS
📋Data format: CSV with one column per treatment group. First row = group names. Each row = one observation per group (unequal sizes: leave blank cells).
Example:
Example:
Control,Drug_A,Drug_B
Editable Data Grid — CRD
Click "Apply to CSV" to transfer grid data to the text area below, then click Run Analysis.
Statistical Interpretation
🟦 RCBD — Randomized Complete Block Design
Two-way ANOVA without interaction: partition block variation to increase precision. Each block contains every treatment exactly once.
DATA & OPTIONS
📋Data format: 3 columns —
Treatment, Block, Response. Each treatment should appear exactly once per block; a missing or replicated plot is analysed by least squares (Type III) with a notice.
Editable Data Grid — RCBD
Columns: Treatment, Block, Response. Click "Apply to CSV" when done.
Statistical Interpretation
🟧 Latin Square Design
Controls two nuisance factors (rows and columns) simultaneously. Requires n treatments, n rows, and n columns (n×n layout).
DATA & OPTIONS
📋Data format: 4 columns —
Row, Col, Trt, Response. Exactly one observation per (row, col) cell.
Editable Data Grid — Latin Square
Columns: Row, Col, Trt, Response. Click "Apply to CSV" when done.
Statistical Interpretation
🟩 Factorial Design (Two-Way ANOVA)
Study main effects and interactions of two (or three) factors simultaneously. Supports balanced and unbalanced designs.
DATA & OPTIONS
📋Data format: CSV columns —
FactorA, FactorB, Response. Each row is one observation. Multiple rows per combination = replicates.
Editable Data Grid — Factorial
Columns: FactorA, FactorB, Response (add FactorC for 3-way). Click "Apply to CSV" when done.
Statistical Interpretation
🔷 Split-Plot Design
Factor A applied to whole plots (blocks), Factor B applied to subplots within each whole plot. Uses two separate error terms.
DATA & OPTIONS
📋Data format: 4 columns —
Block, WholePlotA, SubplotB, Response. Each block contains all combinations of A and B.
Editable Data Grid — Split-Plot
Columns: Block, WholePlotA, SubplotB, Response. Click "Apply to CSV" when done.
Statistical Interpretation
ANCOVA — Analysis of Covariance
Adjusts group means for one or more continuous covariates (pre-test score, age, weight…): Type III tests, adjusted means with their intervals, a joint test of homogeneity of slopes, a Miller–Chapman independence check per covariate, and the Johnson–Neyman region of significance for when the slopes are NOT parallel.
DATA & OPTIONS
📋Data format:
Group, then the covariates (as many as the box above says), then Response last. Every covariate must be measured BEFORE the treatment and be unaffected by it: adjusting for something the treatment changed removes part of the effect being measured (Miller & Chapman 2001), and the panel tests each covariate for that.
Editable Data Grid — ANCOVA
Columns: Group, the covariates, then Response last. Click "Apply to CSV" when done.
Statistical Interpretation
Repeated Measures ANOVA
Within-subjects design: same participants measured across time points or conditions. Mauchly's sphericity test with Greenhouse-Geisser and Huynh-Feldt corrections.
DATA & OPTIONS
📋Data format (wide): First column = Subject ID, remaining columns = measurement at each time point. Column names become condition/time labels.
Editable Data Grid — Repeated Measures (wide format)
Row 1 = header (Subject, T1, T2, …). Each subsequent row = one subject. Click "Apply to CSV" when done.
Statistical Interpretation
Crossover Design (AB/BA)
2×2 crossover trial: subjects receive both treatments in different sequences. Tests treatment, period, and carryover effects (Grizzle 1965).
DATA & OPTIONS
📋Data format: 5 columns —
Subject, Sequence, Period, Treatment, Response. Sequence must be AB or BA. Period must be 1 or 2.
Editable Data Grid — Crossover
Columns: Subject, Sequence (AB/BA), Period (1/2), Treatment, Response. Click "Apply to CSV" when done.
Statistical Interpretation
🏗️ Nested (Hierarchical) Design
Factor B is nested within Factor A — B levels are not the same across A levels (e.g. technicians within labs, students within classes). Estimates variance components.
DATA & OPTIONS
🔀What this changes: with B random (the usual case — technicians, batches, plots sampled from a population) Factor A is tested against MS B(A), and the variance components are reported. With B fixed (every B level is of interest in itself) Factor A is tested against MS Error, and variance components are not defined.
📋Data format: 3 columns —
FactorA, FactorB, Response. Factor B labels are unique within each level of A (e.g. Lab1-Tech1, Lab1-Tech2, Lab2-Tech1 are different technicians).
Editable Data Grid — Nested Design
Columns: FactorA, FactorB (nested in A), Response. Click "Apply to CSV" when done.
Statistical Interpretation
Response Surface Methodology
Fit a second-order model in up to seven factors and any number of responses: ANOVA for regression, inference on every term, pure error and lack of fit, the stationary point with its canonical analysis, contour and surface plots, ridge analysis when the optimum lies outside the region, and desirability optimisation across responses.
DATA & OPTIONS
📋Data format: the first k columns are the FACTORS, in coded units; every column after them is a RESPONSE, so several responses can be fitted at once. Coded means centred at 0 and scaled so the cube is ±1 (the axial points of a rotatable CCD then sit at ±α, e.g. α = 1.414 for two factors). Code the levels: the canonical and ridge analyses are not invariant under rescaling a factor, so running them on raw engineering units describes a different surface — the panel says so when the values do not look coded. Include replicated design points, or lack of fit cannot be tested at all.
Editable Data Grid — RSM / CCD
Columns: the k factors first (coded), then one column per response. Click "Apply to CSV" when done.
Statistical Interpretation