Definition and Purpose
Concept
Blocking: technique to partition experimental units into homogeneous groups called blocks. Purpose: isolate variation due to nuisance factors, increase precision of treatment comparisons.
Experimental Units
Units: smallest entities receiving treatments. Grouped into blocks based on known or suspected sources of variability (e.g., age, location, time).
Primary Goal
Goal: reduce experimental error variance. Improves statistical power and validity of inference about treatment effects.
Rationale for Blocking
Variance Reduction
Blocking controls extraneous variability by grouping similar units. Variance within blocks minimized; between-block variance accounted separately.
Confounding Control
Blocks represent nuisance factors. Separating block effects prevents confounding with treatment effects.
Statistical Efficiency
Efficiency gained by eliminating block-related noise. Smaller error term in analysis of variance (ANOVA). More precise estimates of treatment means.
Types of Blocking
Fixed Blocks
Blocks fixed and known. Effects estimated and interpreted explicitly. Example: different machines, batches.
Random Blocks
Blocks considered random samples from larger population. Effects treated as random variables. Used to generalize findings.
Nested and Crossed Blocks
Nested blocks: blocks within blocks (hierarchical). Crossed blocks: two blocking factors intersect fully. Complex designs use both.
Implementation in Experiments
Identification of Blocking Factors
Choose factors known or suspected to affect response but not of primary interest. Examples: site, operator, time period.
Forming Blocks
Group units so within-block units are homogeneous. Block size depends on experiment size and treatment number.
Randomization Within Blocks
Treatments randomly assigned within each block to avoid bias. Maintains validity of statistical tests.
Block Size Considerations
Block size typically equals number of treatments (complete blocks). Incomplete blocks used when full blocking not feasible.
Statistical Model with Blocks
Linear Model Formulation
Model: \( Y_{ij} = \mu + \tau_i + \beta_j + \epsilon_{ij} \)
Where:
\(Y_{ij}\): response from treatment \(i\) in block \(j\)
\(\mu\): overall mean
\(\tau_i\): treatment effect
\(\beta_j\): block effect
\(\epsilon_{ij}\): random error
Assumptions
Errors independent, normally distributed, equal variance. Block effects additive and fixed or random depending on design.
ANOVA Table Structure
Source of variation: Blocks, Treatments, Error. Degrees of freedom partitioned accordingly.
| Source | Degrees of Freedom | Sum of Squares | Mean Square |
|---|---|---|---|
| Blocks | b - 1 | SS_Blocks | MS_Blocks |
| Treatments | t - 1 | SS_Treatments | MS_Treatments |
| Error | (b - 1)(t - 1) | SS_Error | MS_Error |
Hypothesis Testing
Test treatment effects using F-ratio: \( F = \frac{MS_{Treatments}}{MS_{Error}} \). Blocks tested similarly if fixed.
Advantages and Limitations
Advantages
1. Variance reduction: increases precision.
2. Controls nuisance factors: reduces confounding.
3. Enhances interpretability: separates known sources of variability.
Limitations
1. Requires prior knowledge of nuisance factors.
2. May complicate design and analysis.
3. Ineffective if blocks heterogeneous or misclassified.
Practical Constraints
Small block sizes limit treatment replications. Increased complexity in randomization logistics.
Blocking vs Randomization
Randomization
Purpose: eliminate bias, ensure treatment groups comparable. Random assignment of treatments across all units.
Blocking
Purpose: reduce variability by grouping similar units before randomization. Randomization then performed within blocks.
Complementary Roles
Blocking reduces variance; randomization protects against bias. Combined approach recommended for rigorous designs.
Examples of Blocking
Agricultural Field Trials
Blocks: fields or plots with similar soil type or slope. Treatments: fertilizer types. Blocking controls spatial variability.
Clinical Trials
Blocks: patient age groups or centers. Treatment randomization within blocks balances confounding factors.
Industrial Experiments
Blocks: machines or operators. Treatments: process settings. Controls variability in equipment or personnel.
Table: Blocking Example in Agriculture
| Block | Treatment A | Treatment B | Treatment C |
|---|---|---|---|
| Block 1 | 20.5 | 22.0 | 19.8 |
| Block 2 | 21.7 | 23.1 | 20.3 |
| Block 3 | 19.9 | 21.8 | 20.1 |
Analysis of Blocked Designs
ANOVA Procedure
Partition total sum of squares into block, treatment, and error components. Calculate mean squares and F-tests.
Estimation of Treatment Effects
Estimate treatment means adjusted for block effects. Use least squares or maximum likelihood methods.
Multiple Comparisons
Apply post-hoc tests (e.g., Tukey, Bonferroni) to identify significant treatment differences.
Assumption Diagnostics
Check normality of residuals, homoscedasticity, independence. Use residual plots, tests (Shapiro-Wilk, Levene's).
ANOVA Summary Table:Source DF SS MS FBlocks b-1 SS_Blocks MS_Blocks MS_Blocks/MS_ErrorTreatments t-1 SS_Treatments MS_Treatments MS_Treatments/MS_ErrorError (b-1)(t-1) SS_Error MS_Error -Common Mistakes in Blocking
Improper Block Formation
Blocks heterogeneous or overlapping in nuisance factors reduce effectiveness. Leads to residual confounding.
Ignoring Block Effects
Omitting blocks in analysis inflates error variance. Underestimates precision of treatment effects.
Overblocking
Too many small blocks reduce degrees of freedom, limit treatment replication, and complicate analysis.
Confounding Blocks with Treatments
Blocks correlated with treatments cause confounding, invalidating treatment effect estimates.
Advanced Topics
Incomplete Block Designs
Blocks contain subset of treatments. Used when complete blocking not feasible. Requires specialized analysis.
Blocking in Factorial Experiments
Blocks accommodate multiple factors and interactions. Complex ANOVA models applied.
Random Effects and Mixed Models
Blocks as random effects modeled using mixed-effects models. Allow inference beyond sampled blocks.
Covariate Adjustment and Blocking
Combining blocking with covariate analysis (ANCOVA) to control continuous nuisance variables.
Optimal Blocking Strategies
Algorithmic methods to form blocks minimizing within-block variance. Use clustering or principal components.
References
- Fisher, R. A. "The Design of Experiments." Oliver and Boyd, 1935.
- Kempthorne, O. "The Design and Analysis of Experiments." Wiley, 1952.
- Montgomery, D. C. "Design and Analysis of Experiments." 9th ed., Wiley, 2017.
- Box, G. E. P., Hunter, J. S., & Hunter, W. G. "Statistics for Experimenters." Wiley, 1978.
- Wu, C. F. J., & Hamada, M. S. "Experiments: Planning, Analysis, and Optimization." 2nd ed., Wiley, 2009.