A. Agresti et Ba. Coull, APPROXIMATE IS BETTER THAN EXACT FOR INTERVAL ESTIMATION OF BINOMIAL PROPORTIONS, The American statistician, 52(2), 1998, pp. 119-126
For interval estimation of a proportion, coverage probabilities tend t
o be too large for ''exact'' confidence intervals based on inverting t
he binomial test and too small for the interval based on inverting the
Wald large-sample normal test (i.e., sample proportion +/- z-score x
estimated standard error). Wilson's suggestion of inverting the relate
d score test with null rather than estimated standard error yields cov
erage probabilities close to nominal confidence levels, even for very
small sample sizes. The 95% score interval has similar behavior as the
adjusted Wald interval obtained after adding two ''successes'' and tw
o ''failures'' to the sample. In elementary courses, with the score an
d adjusted Wald methods it is unnecessary to provide students with awk
ward sample size guidelines.