Approximate Binomial Confidence Limits
Colin R. Blyth · Journal of the American Statistical Association · 1986
This article examines the accuracy of normal approximations to confidence limits for the binomial (n, p) parameter p. The method of getting “exact” values from an F table is too slow for practical use and useless for large n. The two usual approximations have unsatisfactory accuracy both for small n and asymptotically, and the simpler one lacks an invariance property of the confidence limit. Hall's (1982) approximation also lacks invariance and, even when symmetrized, has unsatisfactory accuracy. Pratt's (1968) approximation has excellent accuracy for all n. Inversion of a Molenaar (1973) normal approximation to the binomial gives better asymptotic accuracy than Pratt's approximation, but poor values for observations close to n.
Cited by 181 works. doi:10.1080/01621459.1986.10478343
Authors
References
- Journal of The American Statistical Association
- Introduction to Statistical Analysis.
- Statistical Theory with Engineering Applications
- The American Statistician
- Biometrika Tables for Statisticians
- Biometrika Tables for Statisticians.
- The Distribution of Chi-Square
- Statistical Theory and Methodology in Science and Engineering
- Statistical Theory with Engineering Applications.
- Applied Statistics