Abstract
Comparative fishing experiments may include experimental mesh sizes that capture relatively few individuals, resulting in sparsely represented treatments whose influence on selectivity estimation is not well understood. This study evaluated the analytical influence of a sparsely represented experimental mesh on trammel-net selectivity estimates for blackfin flounder (Glyptocephalus stelleri) using complementary parametric (SELECT) and non-parametric generalized additive model (GAM) approaches. Experimental fishing was conducted using five mesh sizes (69, 76, 90, 97, and 121 mm). A complete dataset containing all mesh sizes was compared with a reduced dataset excluding the sparsely represented 121 mm mesh. Model selection, selectivity-curve characteristics, prediction performance, and parameter uncertainty were evaluated using both analytical frameworks. Exclusion of the 121 mm mesh did not change the best-supported model within either framework, and the principal selectivity pattern remained comparatively stable between datasets. Prediction errors increased moderately following exclusion of the sparse observations, while bootstrap analysis indicated substantially greater uncertainty in the secondary binormal component than in the primary component. Differences between datasets were more apparent along the descending limb and upper tail of the selectivity curve, suggesting that the sparsely represented mesh may provide additional information in these less well-represented regions while having little influence on the principal retention region. These findings suggest that the analytical value of sparsely represented experimental treatments should not be assessed solely according to sample size, but also according to their influence on different regions of the selectivity curve. The hierarchical sensitivity framework applied in this study provides a structured approach for evaluating such treatments and may help inform the design and interpretation of future comparative fishing experiments.