Comparison of confidence intervals for fraction in finite populations

Consider a finite population. Let qÎ(0,1) denotes the fraction of units with a given property. The problem is in interval estimation of q on the basis of a sample drawn due to the simple random sampling without replacement. In the paper three confidence intervals are compared: exact based on hypergeometric distribution and two other based on approximations to hypergeometric distribution: Binomial and Normal. It appeared that Binomial based confidence interval is too conservative while the Normal based one does not keep the prescribed confidence level.