What is the difference between BINOMIAL and POISSON Confidence Intervals?
The 1994 PHIP recommends that if there are 100 or fewer events or the rate is less than 10%, 95% Poisson confidence intervals should be calculated. If there are more than 100 events and the rate is greater than 10%, then the normal approximation to the binomial is recommended (in VistaPHw this is labeled “binomial”). In reality, the appropriate distribution that describes these outcomes for people (i.e., death/no death, birth/no birth, low birthweight/normal birthweight, etc.) is the binomial distribution. Calculation of exact binomial confidence intervals involves complex mathematics. However, conveniently, for rare events, the Poisson distribution (by Ury & Wiggins) approximates the binomial distribution, and for non-rare events the normal distribution approximates the binomial distribution. It is relatively straightforward to calculate confidence intervals for Poisson and normal variables, so these methods are in VistaPHw. Thus, in VistaPHw, when “binomial” confidence intervals are
