The Origins of Logistic Regression
| Author | J. S. Cramer |
|---|---|
| Published | 2002 |
| Format | Research paper |
| Original | Read the original |
TL;DR — J. S. Cramer’s 2002 history of the logistic function, from Verhulst’s 1830s population curves to Berkson’s ‘logit’ and McFadden’s discrete-choice models.
J. S. Cramer of the University of Amsterdam and the Tinbergen Institute wrote this 16-page discussion paper in November 2002. It is an expanded version of a chapter from his book Logit Models from Economics and Other Fields, then about to be published by Cambridge University Press. His thesis is that the survival of the word “logistic” and the wide use of the method came down to the personal histories and actions of a handful of scholars.
The story starts with population growth. Quetelet saw that exponential growth could not continue forever, and his student Pierre-François Verhulst added a brake term and named the result the logistic function, in papers published between 1838 and 1847. Pearl and Reed rediscovered the curve in 1920 for US census data without knowing of Verhulst. A second root runs through chemistry, where the same curve describes autocatalytic reactions. In bioassay, Bliss’s probit, built on the normal distribution, dominated from the 1930s. Joseph Berkson proposed the logistic as an alternative in 1944, coined “logit”, and fought a long campaign for it against the statistical establishment. A table counting journal articles shows logit drawing level with probit by about 1970 and then pulling ahead. The paper ends with logistic regression spreading to economics and the social sciences through computer packages, Cox’s and Theil’s multinomial versions, and McFadden’s link to discrete-choice theory, which earned him the 2000 Nobel Prize.
The paper also includes small biographical details, such as Verhulst’s political misadventure in Rome in 1830 and Quetelet’s suggestion that his early death was due partly to his great height.
What it is: A working paper in the history of statistics by an econometrician who worked on logit models. It is careful scholarship with full references, but it is a discussion paper rather than a peer-reviewed article. The text layer of this PDF is garbled by a font-encoding error, so it has to be read visually.
Where this came from
- Will Manidis’s “most underrated PDF” thread, July 2023 — routing record: The Most Underrated PDF on the Internet (July 2023)
16 pages. A copy is archived locally against link rot; the header links the original source.