Estimation, inference, and data analysis for log-linear regression models in tagging studies

In this paper we investigate the use of log-linear regression models for the estimation of natural and fishing (hunting) mortality in a tag-recapture experiment.The efficiencies of several methods of estimation are considered and, when computational ease is also considered, weighted least squares is the recommended technique.Choosing an inappropriate model could have serious ramifications, e.g. for management, in this type of experiment analysis, and the properties of a class of alternative models are studied. The alternative models include age-specific mortality and changing fishing intensity. The examination of residuals, after fitting the simplest model, is considered as a technique for detecting when an alternative model is appropriate.A joint confidence region for F and M is obtained by transforming the confidence region for the slope and intercept in the log-linear model. This is illustrated by a orked example.Because F and M are estimated it is necessary to consider F and M jointly via a confidence region rather than via separate confidence intervals for F and M individually. The need for this joint region is illustrated in a management-oriented example.