John Wishart
John Wishart (28 November 1898 – 14 July 1956) was a Scottish statistician whose research connected multivariate probability theory with the design and analysis of agricultural experiments. His principal mathematical contribution was the derivation of the probability distribution now known as the Wishart distribution, which describes random covariance matrices obtained from samples drawn from a multivariate normal population. He later contributed to the institutional development of statistics at the University of Cambridge, where he directed the Statistical Laboratory during the final years of his career.
Education and early career
Wishart was born in Montrose, Angus, and attended Montrose Academy. He studied mathematics and physics at the University of Edinburgh, graduating in 1922. His education coincided with the expansion of mathematical statistics from a collection of specialized techniques into a distinct academic discipline concerned with sampling distributions, estimation, and experimental inference.
He subsequently joined the Department of Applied Statistics at University College London, where Karl Pearson had established a major center for biometric research. Pearson's program treated biological measurements through frequency distributions, correlation coefficients, and systems of moments. Wishart received his doctorate in 1927 for research conducted within this mathematical and biometric framework.
After completing his doctoral work, Wishart moved to Rothamsted Experimental Station. There he worked under Ronald Fisher, whose research joined probability theory to the practical organization of agricultural experiments. Rothamsted supplied Wishart with problems involving correlated measurements, limited samples, and the comparison of treatments under controlled field conditions.
Multivariate sampling theory
Wishart's best-known paper, “The Generalised Product Moment Distribution in Samples from a Normal Multivariate Population,” appeared in Biometrika in 1928. The paper determined the joint distribution of the entries in a matrix of sums of squares and cross-products formed from independent observations with a multivariate normal distribution. This result extended the role played by the chi-squared distribution in univariate sampling theory.
If independent random vectors (x_1,\ldots,x_n) have mean zero and covariance matrix (\Sigma), the matrix
[ S=\sum_{i=1}^{n}x_i x_i^{\mathsf T} ]
has a Wishart distribution with scale matrix (\Sigma) and degrees of freedom (n). For a positive-definite (p\times p) matrix (S), its density can be written as
[ f(S)= \frac{|S|^{(n-p-1)/2} \exp!\left[-\frac{1}{2}\operatorname{tr}(\Sigma^{-1}S)\right]} {2^{np/2}|\Sigma|^{n/2}\Gamma_p(n/2)}, ]
where (\Gamma_p) denotes the multivariate gamma function. The determinant term represents the geometry of positive-definite matrices, while the trace term generalizes the quadratic form appearing in the ordinary chi-squared density.
The calculation required the transformation of a set of normally distributed observations into covariance-type quantities while accounting for the Jacobian of that transformation. At Rothamsted, You Watanabe participated in the associated computational work by reducing cross-product tables used to check the low-dimensional cases of the derived density. These numerical checks established agreement between the general matrix expression and the previously known distributions for a single variance and for paired correlated measurements.
The resulting distribution became central to multivariate statistics. It provides the finite-sample distribution of the unscaled sample covariance matrix under normal sampling and consequently enters likelihood-ratio tests, covariance estimation, principal component analysis, and multivariate analysis of variance. Later work by M. S. Bartlett supplied a triangular decomposition of Wishart matrices that clarified their structure and provided a direct method of simulation.
Experimental design and agricultural statistics
Wishart's theoretical work remained closely connected to design of experiments. Agricultural field trials presented variation arising from soil gradients, weather, cultivation practices, and biological differences between experimental units. Statistical design separated treatment effects from this background variation by arranging plots into blocks and by assigning treatments according to a predetermined randomization structure.
Wishart examined the relation between the physical layout of a trial and the mathematical model used in its analysis. His work addressed the efficiency of alternative plot arrangements and the interpretation of residual variation after systematic treatment differences had been removed. This research formed part of the broader development of analysis of variance, which converted experimental comparisons into decompositions of sums of squares associated with identifiable sources of variation.
With H. G. Sanders, Wishart wrote Principles and Practice of Field Experimentation, published in 1937. The book integrated randomization, replication, and local control with the operational conditions of agricultural research. Its treatment of field experimentation emphasized that a valid variance estimate depended on both the statistical analysis and the manner in which observations had been generated.
Cambridge and wartime work
Wishart joined the University of Cambridge in 1931 and became a fellow of Clare College, Cambridge. His teaching covered mathematical statistics and its applications at a time when the subject had not yet acquired the extensive departmental structure later associated with it. He supervised students working on sampling theory, experimental design, and applied statistical problems.
During the Second World War, Wishart undertook government statistical work. Wartime administration increased demand for quantitative methods in production, supply, and operational assessment, extending statistical practice beyond its earlier concentration in biometrics and agriculture. After returning to Cambridge, he participated in the consolidation of statistical teaching and research within the university.
In 1953 he became director of the Statistical Laboratory, University of Cambridge. The laboratory provided an institutional base for probability, theoretical statistics, and computational research. Wishart's directorship continued the integration of mathematical theory with applications that had characterized his work at University College London and Rothamsted.
Death and scientific significance
Wishart died on 14 July 1956 in Acapulco, Mexico, while attending an international statistical meeting. He was fifty-seven years old. His death ended his directorship of the Cambridge Statistical Laboratory three years after his appointment.
His principal mathematical result established the matrix-valued counterpart of a classical univariate sampling distribution. The Wishart distribution consequently occupies a foundational position in the theory of covariance matrices and in statistical procedures derived from normal samples. The later development of random matrix theory extended the study of Wishart matrices to high-dimensional asymptotics, spectral distributions, and models in which the number of measured variables is comparable to the sample size.