Balanced Incomplete Block Design — Definition, Formula & Example
A balanced incomplete block design (BIBD) is an experimental layout for when each block is too small to hold every treatment. Each block gets only some of the treatments, but every pair of treatments appears together in the same number of blocks, so all pairs can be compared equally well.
A BIBD with parameters is a set of treatments (points) together with a collection of subsets of called blocks, each of size with , such that every pair of distinct treatments lies in exactly blocks. It follows that every treatment lies in exactly blocks. Fisher's inequality states that .
Key Formula
Where:
- = number of treatments
- = number of blocks
- = number of blocks containing each treatment (replications)
- = number of treatments per block (block size), with k < v
- = number of blocks containing each pair of distinct treatments
Worked Example
Problem: A food scientist wants to compare 7 salsa recipes. A judge's palate tires after 3 tastings, so each judge (block) can taste only 3 recipes. She wants every pair of recipes tasted together by exactly 1 judge. How many judges does she need, how many judges taste each recipe, and what is one valid assignment?
Step 1: Use the pair-counting equation with v = 7, k = 3, and λ = 1 to find the replication r.
Step 2: Use bk = vr to find the number of blocks (judges).
Step 3: Build the blocks by starting from {1, 2, 4} and adding 1 to each recipe number, wrapping 8 back to 1, 9 to 2, and so on. Check balance: each block holds 3 pairs, and 7 blocks hold 21 pairs, which matches the 21 distinct pairs of 7 recipes. For instance, recipe 1 appears in blocks 1, 5, and 7, and the pair (1, 2) appears only in block 1.
Answer: She needs 7 judges, and each recipe is tasted by 3 judges. The seven blocks above form a BIBD, also known as the Fano plane.
Why It Matters
Taste tests, wine judging, and field trials often limit how many treatments one block can hold. A BIBD keeps every treatment comparison equally precise despite that limit. The analysis uses an ANOVA with treatment means adjusted for block effects. The same structures appear in coding theory and in tournament scheduling, as Steiner triple systems.
Common Mistakes
Mistake: Assuming a design exists whenever the two parameter equations give whole numbers.
Correction: Those equations are necessary but not sufficient. For v = 16, k = 6, λ = 1 you get r = 3 and b = 8, which are both integers. But b = 8 < 16 violates Fisher's inequality, so no such design exists. Even parameters that pass all three checks can fail to have a design.
