This article was accepted into the corpus but its outbound wikilinks were never NER-processed — typical at the deepest BFS hop or when the run's entity cap was reached. No expansion funnel to show.
| Latin square (design of experiments) | |
|---|---|
| Name | Latin square |
| Field | Design of experiments |
| Introduced | 18th century |
| Notable users | Ronald Fisher, William Gosset, Fisher's Exact Test, Fisher–Yates shuffle |
Latin square (design of experiments) A Latin square is an experimental layout that arranges treatments in a square so each treatment appears exactly once in each row and each column. Developed in the tradition of systematic designs used by Ronald Fisher, Sir Ronald A. Fisher-era statisticians, and applied by researchers at institutions such as the Royal Society, Imperial College London, and Wye College, the Latin square controls two orthogonal nuisance factors while estimating treatment effects. It has been used historically in agricultural trials by practitioners linked to John Bennet Lawes, Joseph Henry Gilbert, and later applied in industrial studies associated with George E. P. Box and Walter A. Shewhart.
A Latin square is an n × n arrangement of n treatments where each treatment appears once per row and once per column, aiming to remove the confounding influence of two blocking factors while assessing treatment differences; its use was popularized by Ronald Fisher and adopted by researchers at University of Cambridge and Rothamsted Experimental Station. The purpose is to increase precision in experiments influenced by two systematic sources of variation—examples include trials at Wye College comparing crop varieties across soil gradients and early industrial trials at Bell Labs examining process settings. The design underpins analyses used by statisticians associated with Fisher's Exact Test, Neyman–Pearson lemma contexts, and subsequent methodology advanced at Harvard University and University of Chicago.
Construction typically starts from a reduced Latin square or a cyclic generator: for prime n a canonical construction uses modular arithmetic tied to methods employed by mathematicians like G. H. Hardy and John Littlewood; for composite n orthogonal pairings reference finite field techniques developed by researchers at École Normale Supérieure and University of Göttingen. Key properties include orthogonality to two blocking factors, balance in treatment representation akin to properties studied by Leonard Euler and later by combinatorialists at Princeton University and MIT, and the availability of mutually orthogonal Latin squares (MOLS) which relates to work by R. C. Bose and E. T. Parker. Existence results connect to finite projective planes investigated by scholars at University of Michigan and University of California, Berkeley; non-existence for certain orders references the Euler conjecture history and contributions from R. C. Bose and S. S. Shrikhande.
The standard linear model for a Latin square includes additive fixed effects for treatments and the two blocking factors; estimation and hypothesis testing follow classical analyses as formulated by Ronald Fisher and extended in textbooks from Cornell University and Stanford University. Analysis of variance (ANOVA) partitions total variability into treatment, row, column, and residual components in the spirit of work by William Sealy Gosset and later formalized by researchers at Iowa State University. Assumptions—additivity, homoscedasticity, independence—are discussed in the same methodological lineage as Student's t-test and methods refined at Johns Hopkins University. Extensions to mixed models and random block effects reflect developments from George Box, David R. Cox, and statisticians at University of Warwick and North Carolina State University.
Extensions include Graeco-Latin squares (pairing two orthogonal Latin squares) inspired by Leonard Euler and advanced by R. C. Bose and S. S. Shrikhande; Youden squares (rectangular form) used in trials at Rothamsted Experimental Station; and hyper-Graeco-Latin structures studied by combinatorialists at Princeton University and University of Waterloo. Related designs comprise randomized block designs promoted by Fisher and developed further by William Cochran and Gertrude Cox, balanced incomplete block designs (BIBD) researched at Bell Labs and Bell Telephone Laboratories, and factorial experiments championed at DuPont and analyzed at General Electric under guidance from Ronald Fisher-influenced statisticians. Orthogonal arrays and confounding structures connect to coding theory developments from Claude Shannon and combinatorial research at AT&T Bell Laboratories.
Classic agricultural applications include variety trials at Rothamsted Experimental Station, fertilizer experiments influenced by John Bennet Lawes, and irrigation trials at Wye College; industrial examples include process parameter studies at Bell Labs, DuPont, and General Motors. Clinical and pharmaceutical applications have appeared in trials coordinated by NIH and analyzed by teams at Mayo Clinic and Cleveland Clinic; psychological and educational testing align with work conducted at Stanford University and University of Oxford. Computer experiments and simulation studies using Latin squares have been used at Los Alamos National Laboratory and Sandia National Laboratories for sensitivity analyses and screening.
Implementing a Latin square requires randomization of treatment labels, verification of blocking factor structure (spatial, temporal, or batch effects), and assessment of assumptions via residual diagnostics practiced in statistical groups at Imperial College London and London School of Hygiene and Tropical Medicine. Missing plots and unequal variances lead to modifications recommended by practitioners at University of Minnesota and North Carolina State University, including use of mixed-effects modeling and restricted maximum likelihood as implemented in software from teams at SAS Institute, R Project, and Python Software Foundation. Practical constraints—availability of r×c layouts, field heterogeneity, and logistical block sizes—mirror considerations in multi-factorial studies at University of California, Davis and Cornell University.
Category:Design of experiments