Omission rates
We found all selection approaches chose optimal models with percentile ORs equal to or more than theoretically expected 10% for both fish (Figure 3a) and odonates (Figure 3b). On the other hand, except for four fish and 10 odonates, all had optimal models with 0 balance ORs (Figures 3a-b). Generally, there was a very high and statistically significant correlation for the percentile ORs of the optimal models chosen by the EXP approach to the optimal models chosen by all four sequential approaches for fish, while for the odonate there was statistically significant but very low correlations only with ORTEST_PER, AUCDIFF_PER and ORDIFF_BAL (Table S4). However, for the balance OR EXP approach had statistically significant correlations only with ORTEST_PER and AUCDIFF_PER for fish and with AUCDIFF_PER for odonates (Table S4).