Abstract: In this article, we analyze data obtained in a study
involving the tagging and release of red drum fish. Four groups of fish are
considered; nylon dart tags released in the wild, nylon dart tags released in a
tank, stainless steel tags released in the wild, and stainless steel tags
released in a tank. All fish have an additional subcutaneous embedded tag so
that those caught without external tags can still be identified as having been
tagged and released previously. Tag loss is assumed to have an exponential
distribution with rate depending on the group. At
non-random time points, fish are caught and observed for tags. Data is censored
since the exact time of tag loss is unknown for fish caught without tags. We
construct logistic regression models to predict the probability that a fish will
have retained its tag based on the number of days at large (since release). We
also use Bayesian methods for estimating the tag loss parameters lfor each group, and
use these to determine if differences exist among the groups with regards to tag
loss rates. Model predictions are compared to actual data values to determine
the goodness of fit.
Keywords and phrases: tag loss times, Bayesian estimation methods, logistic regression modeling, censored data.
