Numerical tables in early twentieth-century statistics: accuracy, usage, and computational practices

Journal of Dynamical and Control Systems · Springer Nature · 2026-09-11 · DOI: 10.1007/s00407-026-00386-8

Аннотация

Before electronic computers, statistical analysis depended on numerical tables that provided discretized approximations to continuous functions. These tables were essential for the biometric and agricultural studies of Karl Pearson and Ronald A. Fisher. Yet, their accuracy has never been systematically examined. This paper compares Pearson’s (1914) and Fisher’s (1925) tables with modern computations. Pearson’s normal tables achieved remarkable accuracy (mean absolute error $$4 \times 10^{-5}$$), while errors for the t and Chi-square distributions were larger (up to $$5 \times 10^{-4}$$). We examine the computational methods, economic constraints, and contemporary awareness of numerical errors. Discretization errors were small relative to other sources of uncertainty, and the tables were accurate enough for their purposes. This study contributes to the history of statistical computation by providing a quantitative assessment of early twentieth-century table accuracy. The findings reveal the priorities and constraints that shaped table construction, and provide a foundation for understanding the reliability of historical analyses.

Читать у издателя →

Похожие статьи