Benefits of statistical blocking techniques in the design of gear evaluation trials: introducing the Latin Square design

Abstract

We evaluated the use of a Latin Square design for assessing the performance of bycatch reduction devices (BRDs). We wanted to ensure the most precise result possible given limited resources, high variability and the need to detect small differences. The Latin Square design has two features to reduce variation: removing the tow-to-tow variation by deploying the experimental and standard nets simultaneously; and removing the different efficiencies of the nets by swapping codends and averaging consecutive pairs of tows on the same net. The design, which was relatively easy to implement, was used for evaluating the AusTED II in three locations in northern Australia. We reviewed a number of similar studies to see what efficiency they achieved with simpler statistical designs and found that trials based on less than 100 tows had low power (at most 31%) to detect a 5% fall in prawn catch. The Latin Square has the potential to produce more powerful results with some extra effort in swapping codends between nets. The best of our trials (60% power to detect a 5% fall in prawn catch) out-performed similar-sized trials. We recommend this design for gear trials in variable environments, if the codends can be swapped quickly and especially if the trial involves hundreds of tows. We also demonstrate the importance of assessing the statistical power of such trials to avoid over-interpretation of non-significant differences between fishing gear.