Estimating the Size of a Truncated Sample

Abstract

Consider N (unknown) values of a variable X (discrete or continuous) independently sampled from a distribution with density f(x, θ), where θ is an unknown vector parameter. Suppose that the values of X belonging to a certain region R are not observable. This article deals with the problem of estimating N and θ in such situations which arise, for instance, in life testing and capture-recapture census. Asymptotic theory for maximum likelihood estimation of N and θ is presented here and is shown to yield as corollaries both some existing results and a new result pertaining to the truncated negative binomial distribution.

Key Words:

• Truncated sampling
• Truncated negative binomial
• Sample size estimation
• Maximum likelihood
• Conditional maximum likelihood
• Asymptotic theory