Does b at least v guarantee a design exists?

No. It is only one necessary condition.

Is this a test performed on collected responses?

No. It concerns the combinatorial structure of the design.

Can a complete block design have fewer blocks than treatments?

Yes; the nontrivial incomplete-design condition is essential to this statement.

AI Glossary / Statistics and probability / Advanced

Fisher's inequality

Definition

Fisher's inequality is a necessary condition for a nontrivial balanced incomplete block design: the number of blocks b must be at least the number of treatments or points v. Such a design places k distinct treatments in each block, with every treatment occurring equally often and every pair occurring together equally often. Under the usual conditions 1 less than k less than v and positive pair replication, the inequality is b greater than or equal to v. It is a combinatorial theorem, not Fisher's exact test or a significance threshold.

What it means

A design can appear economical on paper while demanding a balance pattern that no arrangement can achieve. Fisher's inequality rules out some such proposals before data collection. Its statement concerns the allocation structure rather than observed outcomes, effect sizes or statistical power. Satisfying b greater than or equal to v is necessary but not sufficient: additional counting conditions and existence constraints remain. A design with enough blocks may still be impossible for the requested block size and replication pattern.

How it works

Let r be the number of blocks containing each treatment and lambda the number containing each pair. Basic counting gives bk = vr and r(k - 1) = lambda(v - 1). Form a v-by-b incidence matrix N whose entries indicate treatment membership in blocks. Then NN-transpose has diagonal r and off-diagonal lambda, so it equals (r - lambda)I + lambda J. In a nontrivial incomplete design r is greater than lambda, making this matrix full rank v. Its rank cannot exceed the number b of columns in N, so b must be at least v. The argument uses exact design balance, not a normality assumption about measured outcomes.

Three practical examples

Experiment planning: reject an impossible request for a balanced incomplete design with fewer blocks than treatments.
Design education: connect pairwise balance to incidence matrices and a rank argument.
Allocation review: distinguish satisfying an arithmetic bound from actually constructing a valid treatment schedule.

Think of it this way

Think of building a perfectly balanced meeting schedule in which every participant attends equally often and each pair meets equally often. Fisher's inequality places a lower limit on the number of meetings when no meeting includes everyone. Meeting that lower limit still does not automatically provide a workable schedule.

Worked example

Consider four treatments A, B, C and D, with blocks ABC, ABD, ACD and BCD. There are v = four treatments and b = four blocks, each of size k = three. Each treatment appears in r = three blocks, and each pair appears in lambda = two blocks. The counting identities hold: bk = 4 x 3 = 12 = vr, and r(k - 1) = 3 x 2 = six = lambda(v - 1). The incidence cross-product has diagonal three and off-diagonal two, so it equals I + 2J. Its eigenvalues are nine in the all-ones direction and one in the three orthogonal directions, establishing rank four. The design attains equality b = v and is called symmetric in this design-theoretic sense. Reducing the number of blocks below four cannot preserve a nontrivial balanced incomplete design on these four treatments.

Illustrative case study

Fictional case: Magnolia Experiment Planning receives a request to compare eight treatments using only six balanced incomplete blocks. Before collecting outcomes, a statistician explains that the requested exact design cannot satisfy Fisher's inequality. The team revisits the number of blocks and the desired balance structure, then separately checks statistical precision and operational constraints. This invented case illustrates a feasibility check. It does not claim that every experiment with fewer blocks than treatments is impossible; the restriction applies to the specified balanced incomplete design conditions.

Common misconceptions and mistakes

Do not apply the theorem indiscriminately to complete blocks, arbitrary incomplete designs or every experimental layout. Its assumptions matter. Do not confuse blocks with individual observations: a block contains several treatment placements. Equality b = v has a specific symmetric-design meaning and does not mean a drawing must look geometrically symmetric. The inequality does not guarantee that all treatment effects are estimable under every analysis model, nor does it provide adequate power. Fisher's inequality also differs from Fischer's inequality in matrix analysis, which has a different spelling and statement.

Business use

Businesses planning constrained comparisons can use this result as an early structural check before investing in scheduling or data collection. Translate operational requirements into explicit treatment, block-size, replication and pair-balance conditions. Confirm all necessary arithmetic relationships, then construct and verify the actual incidence schedule. If the requested design is impossible, reconsider the balance requirements or resources with a statistician rather than quietly relaxing them after collection. Feasibility is only the first step; measurement quality, randomization and sufficient information for the intended comparisons still need attention.

Check your understanding

In the worked example, what does lambda = two mean?

Answer: Every distinct pair of treatments appears together in exactly two blocks, which is separate from each individual treatment appearing three times.

Related terms

Replication (statistics)
balanced incomplete block design
incidence matrix
replication
pairwise balance
symmetric design
matrix rank
experimental design

Learn next

verify the incidence counts of a proposed schedule, then study why necessary parameter conditions do not always imply the existence of a balanced incomplete block design.

Block design

Sources

Sources: Indian Statistical Institute lecture notes, Fisher's inequality for designs: https://www.isibang.ac.in/~d.yogesh/Course_Notes/DM1/Ch9.S2.html ; University of Lethbridge combinatorics text, Fisher's Inequality: https://opentext.uleth.ca/Combinatorics/sect_designs-FisherIneq.html ; Elkies, Harvard combinatorics lecture notes, incidence-matrix proof: https://people.math.harvard.edu/~elkies/M155.15/notes.html .

Previous: Fisher exact test Next: Fisher's least significant difference