Block design

A block design is an incidence structure consisting of a finite set of points and a collection of subsets called blocks. The points may represent treatments, people, locations, or abstract symbols; the blocks specify which points are grouped together. In a 2-design, every pair of distinct points occurs together in exactly λ blocks. A balanced incomplete block design, or BIBD, has v points, b blocks, block size k, and replication number r, meaning that each point occurs in r blocks. Its parameters satisfy bk = vr and r(k−1) = λ(v−1).

The word “incomplete” means that k is smaller than v: no block contains every point. A design is simple when repeated blocks are forbidden, and resolvable when its blocks can be partitioned into parallel classes, each class partitioning the points. These conditions distinguish designs that have identical numerical parameters but different structural behavior.

Block designs turn a comparison problem into a controlled pattern of incidences. In the design of experiments, a block often groups experimental units that are similar in a nuisance factor, such as a field strip, laboratory batch, or testing day; treatments are then assigned within blocks so that treatment contrasts are estimated with less unwanted variation.¹

Designs can be constructed from finite fields, difference sets, recursive procedures, or algebraic objects such as orthogonal arrays. The Fano plane, with seven points and seven three-point blocks, is a 2-(7,3,1) design: every pair lies in exactly one block. More generally, a Steiner system S(t,k,v) requires every t-subset of the v points to occur in exactly one k-subset. Thus a Steiner triple system is precisely a 2-(v,3,1) design, and it exists exactly when v is congruent to 1 or 3 modulo 6.

The incidence matrix records a block design as a zero–one matrix whose rows represent points and whose columns represent blocks. Row sums are r, column sums are k, and the scalar product of two distinct rows is λ in a BIBD. Consequently, the matrix identity NN^T = (r−λ)I + λJ exposes both the design’s regularity and restrictions on its possible parameters.

Important families include finite projective planes, symmetric designs with b=v, and designs derived from Hadamard matrices. Duality interchanges points and blocks in an incidence structure; for a symmetric design, the dual has the same parameters. Existence is not guaranteed merely because the parameter equations have integral solutions. Congruence conditions, eigenvalue arguments, and the Bruck–Ryser–Chowla theorem can rule out designs, while isomorphism testing asks whether two apparently different descriptions are actually the same structure.

Block designs also connect finite geometry, coding theory, and statistical computation. A projective plane of order n would have n²+n+1 points and the same number of lines, with n+1 points on every line; planes are known for every prime-power order, while the general existence question remains unresolved. The orders 6 and 10 are notable nonexistence cases, established by different methods, showing that numerical plausibility can conceal deep obstructions.

Not every useful design is balanced: incomplete or partially balanced designs relax equal co-occurrence when practical constraints make a BIBD impossible. In modern applications, computational searches enumerate designs, test automorphism groups, and optimize criteria such as variance or robustness. The original statistical motivation remains visible in the work of Ronald Fisher, whose use of randomization, replication, and blocking helped establish experimental design as a formal discipline.

Point

An element of the finite set being arranged; in an experiment it may represent a treatment or experimental unit.

Block

A selected subset of points, usually required to have a prescribed size.

BIBD

A balanced incomplete block design in which every point occurs equally often and every pair occurs together equally often.

Replication number

The number r of blocks containing any given point in a balanced design.

Incidence matrix

A zero–one matrix indicating whether each point belongs to each block.