RELIABILITY OF SELECTION CURVES OBTAINED FROM TROUSER TRAWL OR ALTERNATE HAUL EXPERIMENTS

This study presents simulation results from a comparison of three methods for estimating selection curves from selectivity trials where the experimental codend of a trawl is fished in conjunction with a "control" (small mesh) codend. Methods 1 and 2 are based on the traditional approach of fitting selection curves to the so-called "proportion retained," the ratio of experimental codend to control codend catch, at each length class. Method 3 fits to the proportion of the total (control codend plus experimental codend) catch of each length class that is from the experimental codend. The traditional approaches exhibited considerable bias and consistently overestimated the 50% retention length and selection range. This partly explains why, in the published literature, trouser trawl experiments typically give higher estimates of the 50% retention length than covered codend experiments. Method 3 displayed little or no bias and always had lowest mean square error of the three methods. It is shown that the 50% retention length and selection range estimated by method 3 are approximately normally distributed for modest sample sizes.