ANALYSIS OF CODED WIRE TAG RETURNS FROM COMMERCIAL CATCHES

Three alternative but equivalent approaches to the analysis of coded wire tag (CWT) data using log-linear models are presented. All three use iteratively weighted least squares to estimate treatment effects in hatchery releases under the assumption that the variance of a count is proportional to its expected value. The commonly made assumption of normal distributions with constant variance for recovery rates is inefficient. Analysis of tag recovery at the most disaggregated level (i.e. the level at which the sample fraction f is measured) is found necessary for valid inferences. Failure to include zero counts in analyses of recovery data is also shown to induce or mask interactions among CWT recoveries. Recoveries of CWT from coho salmon (Oncorhynchus kisutch) are used to illustrate the method of analysis. Coordinated CWT releases to facilitate mixing of stocks is recommended in the design of hatchery studies.