Reliability of size-selectivity estimates from paired-trawl and covered-codend experiments

To determine the size selectivity of a trawl codend from sea trials, the researcher must choose between using paired (experimental and control) and covered-codend gears. The paired-gear method has the advantage that the experimental codend can be deployed without modification, but has the disadvantage of a more complex statistical analysis. It has previously been claimed that this analysis is inherently biased regardless of sampling effort. This claim is shown to be erroneous, and it is seen that both methods produce estimators that have negligible bias under typical experimental conditions. However, paired-gear analyses were seen to produce estimators having greater statistical variability than the covered-codend analyses. Over the range of simulations employed herein, it was found that the paired-gear method generally had moderately higher (ratio of 1.3-1.9) statistical error for estimation of the length of 50% retention, when fished with equal sampling effort as the covered-codend gear. Paired-gear estimates of the selection range could have considerably higher standard error than covered-codend estimates (ratio of 1.6-4.2), with the highest ratio being for a scenario where the selection curve was asymmetric and sampling effort consisted of ten replicate deployments where the number of measured fish was restricted to 250.