Cluster or Site Blocking¶
Assignment-design method — instantiates Blocking Design
Blocks whole clusters — sites, classrooms, batches, communities — that are the actual unit of assignment, then compares treatments within groups of comparable clusters.
Cluster or Site Blocking applies when the thing you can actually assign is not an individual but a whole group — a clinic, a school, a village, a production line. The intervention reaches everyone in the cluster at once, so the cluster is the experimental unit, and blocking must happen at that level: comparable clusters are grouped, and treatment is randomized within each group of look-alike sites. What makes it this mechanism is its insistence on getting the assignment level right first, and its attention to a consequence peculiar to whole-site blocking — that effects can genuinely differ from one site type to another, and that the blocked set may not transport to sites left out.
Example¶
A public-health team is evaluating a sanitation program that, by its nature, is delivered to an entire village — you cannot give half a village a communal water-treatment system. The village is therefore the unit of assignment. Randomizing villages freely would risk landing all the large, piped-water villages in one arm, so the team blocks villages into groups matched on size, baseline diarrheal rates, and water source, then randomizes treatment within each block of comparable villages.
Because whole communities are the blocks, the team also asks whether the program works differently in large versus small villages, and whether results from the enrolled villages would carry to remote hamlets that were too small to block — treating those as questions to examine, not conclusions to assume. Analysis then respects that outcomes within a village are correlated and that spillover between neighboring villages could blur the contrast.
How it works¶
- Fix the assignment level. Identify that the independently assigned unit is the cluster, with correlation and possible interference within clusters and independence across them; blocking individuals cannot substitute.
- Block comparable clusters. Group sites on cluster-level features that predict outcomes, then randomize treatment within each group of similar clusters.
- Review heterogeneity and transport. Because blocks are whole sites, examine whether the effect varies across site types and whether the enrolled, blockable clusters represent the ones excluded.
Tuning parameters¶
- Cluster definition — what counts as the unit (village vs household, school vs classroom); the choice fixes both assignment and interference structure.
- Clusters per block — pairing sites versus larger blocks, trading tight comparability against flexibility and power.
- Comparability features — which cluster-level attributes define similar sites.
- Interference handling — buffer zones or spacing when treatment can spill between neighboring clusters.
When it helps, and when it misleads¶
Its strength is honesty about the level of action: when an intervention is inherently group-level and sites are heterogeneous, blocking comparable clusters removes the between-site differences that would otherwise dominate a small number of units.
It misleads when someone blocks or analyzes at the individual level while assignment was really cluster-level — no amount of individual blocking repairs that, and it overstates precision. It also misleads when there are simply too few clusters: blocking cannot rescue a design with a handful of sites, where the intracluster correlation and resulting design effect leave little power.[n1] And it misleads when block-level differences in effect are read as clean subgroup findings rather than hypotheses. The discipline is to match the analysis to the cluster level and treat heterogeneity as a question requiring evidence.
How it implements the components¶
experimental_unit_and_assignment_level— its foundational act: it recognizes the cluster as the independently assigned unit and blocks at that level, so randomization and analysis respect clustering and interference.heterogeneity_interaction_and_transport_review— because the blocks are whole sites, it examines whether the effect differs across site types and whether the blocked, enrolled clusters transport to excluded or sparse populations.
It does NOT define the within-block chance mechanism (within_block_assignment_and_probability_rule, Permuted-Block Sequence) or select and contract the prognostic covariates (pretreatment_block_variable_contract, Stratified Randomization Schedule); the cluster-robust uncertainty calculation itself (block_aware_analysis_and_weighting_plan) is Block-Adjusted Effect Estimator's.
Related¶
- Instantiates: Blocking Design — the case where the block, and the unit assigned, is a whole cluster.
- Sibling mechanisms: Block-Adjusted Effect Estimator · Stratified Randomization Schedule · Matched-Pair Randomization · Randomized Complete-Block Design · Incomplete-Block Design · Permuted-Block Sequence · Covariate-Adaptive Randomization · Time, Batch, Run, or Location Block · Within-Block Randomization Inference
Editorial Notes¶
Form Classification¶
Form family: Experiment, Test & Rehearsal
Rationale: Blocks whole clusters — sites, classrooms, batches, communities — that are the actual unit of assignment, then compares treatments within groups of comparable clusters, making its operative form a bounded trial, probe, simulation, or adversarial exercise that generates evidence from performance.
Independent corroboration: The frozen evidence defines Cluster or Site Blocking as 'Blocks whole clusters — sites, classrooms, batches, communities — that are the actual unit of assignment, then compares treatments within groups of comparable clusters', so its operative form is Experiment, Test & Rehearsal.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Single lineage
Present-day reach: Specialized
Rationale: Experimental design established blocking and randomization at the actual assignment unit when treatment is allocated to sites, schools, clinics, or communities.
Review resolution: Both reviewers agree on statistics_experimental_design as primary. Reading the mechanism confirms that its defining operation belongs to that lineage; the final record retains no alternate lineage only as materially formative origin and keeps present-day application breadth separate from provenance.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
The order of operations is non-negotiable: the assignment level is decided before any blocking. Blocking individuals inside a site-assigned intervention is a common and fatal mistake — it manufactures apparent precision from units that were never independently assigned.
[n1] The intracluster correlation coefficient measures how alike units within a cluster are; together with cluster size it drives the design effect, the factor by which effective sample size shrinks under clustering — which is why a cluster study with few sites has little power no matter how carefully the sites are blocked. ↩