Abstract
A maximum likelihood estimate exists for both parameters of the binomial distribution if and only if the sample mean exceeds the sample variance. The derivative of the log-likelihood function has at most one finite root; consequently the likelihood function is unimodular.