(Received April 3, 1989; revised February 14, 1990)
Abstract. Let random samples of equal sizes be drawn from two exponential distributions with ordered means lambdai. The maximum likelihood estimator lambdai* of lambdai is shown to have a smaller mean square error than that of the usual estimator \bar Xi, for each i = 1, 2. The asymptotic efficiency of lambdai* relative to \bar Xi has also been found.
Key words and phrases: Asymptotic efficiency, exponential distribution, isotonic regression, maximum likelihood estimation, mean square error.