Maximum Likelihood Estimation of Models with Beta-Distributed Dependent Variables

Research in political science is often concerned with modelingdependent variables that are proportions. Proportions are relevantin a wide variety of substantive areas, including elections,the bureaucracy, and interest groups. Yet because most researchersrely upon an approach, OLS, that does not recognize key aspectsof proportions, the conclusions we reach from normal modelsmay not provide the best understanding of phenomena of interestin these areas. In this paper, I use Monte Carlo simulationsto show that maximum likelihood estimation of these data usingthe beta distribution may provide more accurate and more preciseresults. I then present empirical analyses illustrating someof these differences.