I'm using a simulation that includes reliability models of different types of entities. I want to estimate the proportion of failures among those entities under certain conditions. The textbook formulas for developing a CI on a population proportion assume that each observation is a binomial random variable and that if the sample is large enough the binomial approximation to the normal suffices to build a CI with a z score and standard deviation based on the binomial distribution. My question is: since i have a simulation model and can perform N independent replications each of n observations, and then have N estimates for the population proportion, would the standard CI formula using the student's t distribution, mean over all N, and empirical estimate of s be more appropriate? or should i stick with the text book method? Thanks, Jon
asked
Jon Davis |