Incomplete Block Designs (BIBD, PBIB, Alpha and Test-versus-Control Designs)

Description

Designs for when each block (often one participant) can take only some of the treatments, allocating subsets so comparisons stay balanced or nearly so, with block-adjusted analysis.

When a block holds fewer units than there are treatments, as when a participant can evaluate only three of nine prototypes, each block receives a subset. In a balanced incomplete block design every pair of treatments appears together in the same number of blocks, lambda, so all comparisons are equally precise; parameters must satisfy rt = bk and lambda(t - 1) = r(k - 1), and designs exist only for some combinations (for example t = 6, b = 10, k = 3, r = 5, lambda = 2). Partially balanced designs relax this into a few association classes when no practical BIBD exists; alpha designs give resolvable layouts for any block size; reinforced and BTIB designs maximise precision of test-versus-control comparisons. Intrablock analysis estimates treatment effects within blocks; treating blocks as random lets interblock information be recovered with REML. The design must be connected so every difference is estimable.

Attributes

Loss type
Siloed
Effort
Days

Steps

  1. Fix the number of treatments t and the block size k that a participant, session or unit can tolerate.
  2. Choose a BIBD from published tables or software; if none of practical size exists, use a PBIB or alpha design with high efficiency factor, or a BTIB or reinforced design if comparisons with a control matter most.
  3. Check the design is connected so every treatment difference is estimable.
  4. Randomise the block labels, the allocation of treatment sets to blocks and the order of positions within each block.
  5. Analyse with a block-plus-treatment model, reporting block-adjusted treatment means and contrasts.
  6. Where blocks can be treated as random and there are enough treatments, estimate block and error variances by REML and use the combined intra- and interblock estimator.

Sources

  • Design and Analysis of Experiments, Vol. 1 (Hinkelmann & Kempthorne, 2008)
  • Design and Analysis of Experiments, Vol. 2 (Hinkelmann & Kempthorne, 2005)
  • Statistical Design and Analysis of Experiments (Mason et al., 2003)

Also known as

  • Balanced incomplete block design
  • BIBD
  • Balanced treatment incomplete block design
  • BTIB
  • Partially balanced incomplete block design
  • PBIBD
  • Alpha designs
  • Resolvable incomplete block design
  • Reinforced BIBD
  • Intrablock and interblock analysis

Categories

  • Analytics & Behaviour
  • Experimentation & Causal Inference