Generating correlated binary variables with complete specification of the joint distribution

Most statistical methods for the analysis of correlated binary data are based on asymptotic theory. Therefore it is important to generate correlated binary data efficiently for Monte Carlo simulation studies to investigate the finite sample performance of these methods. This article provides a simple method for generating correlated binary data with a given joint distribution. The key idea is to consider k-variate binary data as a multinomial distribution with 2^k possible outcomes.