Tests of the Assumptions Underlying Life Table Methods for Estimating Parameters from Cohort Data

Since the late 1940's, life table methods have been used to estimate annual, age-specific mortality rates from tagged or banded animal populations. Deterministic methods in the form of a dynamic or composite dynamic life table were developed and used by Bellrose and Chase (1950) and Hickey (1952), and discussed by Geis and Tabor (1963). Seber (1971) used the same basic model, formulated it in the proper stochastic framework, and provided closed form estimators of the survival parameters based on approximations to the maximum likelihood (ML) estimators. Cormack (1 970) found the ML estimates from the same stochastic model by numerical methods; however, identifiability was not explicitly addressed. Recently, North and Morgan (1979) proposed an extension of the same model by allowing the age-specific annual mortality rate to be a function of weather (i.e., they recognized time-dependent variation). Seber (1972, 1981) and Eberhardt (1972) provided further information on this basic model and its assumptions. Finally, we note that these same assumptions are often applied to capture-recapture data, which can then be analyzed by exactly these same life table methods applied to the final capture only (see, e.g., Mardekian and McDonald 1981). Given this long-used model (i.e., the assumptions), there are a wide variety of ways to attempt to estimate its parameters. However, we question the fundamental assumptions that the model rests upon. In the general context here, there are two assumptions of primary concern: