Extended empirical likelihood for estimating equations

We derive an extended empirical likelihood for parameters defined by estimating equations which generalizes the original empirical likelihood to the full parameter space. Under mild conditions, the extended empirical likelihood has all the asymptotic properties of the original empirical likelihood. The first-order extended empirical likelihood is easy to use and substantially more accurate than the original empirical likelihood.