Gary King
Harvard University
- Research field
- Political methodology, statistics for the social sciences
Gary King is the Albert J. Weatherhead III University Professor at Harvard University and the founding director of its Institute for Quantitative Social Science. With Langche Zeng he showed that logistic regression underestimates the probability of rare events such as wars and epidemics, even in samples of thousands of observations, because the maximum likelihood estimate of the intercept is biased downwards, and the pair derived a correction for the coefficients and a better way of computing the probabilities. The same paper shows that a researcher may collect all the events and only a random sample of the non-events, as in a case-control study, and still estimate the odds ratios without bias after a known correction of the intercept. The unit presents this work after the section on fitting by maximum likelihood, whose large-sample standard errors it qualifies.
In the masterclass
Section 3, after the fitting of a logistic regression by Newton's method, presents his paper with Langche Zeng on what the method does when events are rare.
The research is presented in Unit 11: Logistic Regression and Generalized Linear Models, in a research spotlight box that explains the result at the level of the unit and links it to what you have just learned.