The choice of sample size in estimating mutual information

摘要

When a population is too large to be censused, the mutual information concerning two random variables in it must be estimated in a simple. In previous papers we have obtained unbaised estimators of the quadratic index of mutual information in random samplings with and without replacement and confirmed the advantages of this index over the Shannon one. In the present paper we first analyze the problem of choosing a suitable sample size on the basis of the information supplied by a pilot survey or a previous sampling of the population. Then, the results in this analysis are illustrated by applying them to an ecological example. Finally, we examine the goodness of such results by means of Monte Carlo simulation.