TABLE 11.

Cross-testing: the power of detecting a sweep within a bottleneck if training and testing parameters do not coincide

	Testing data (%)
Training data	b_s1	b_s2	b_s3	b_s4	b_s5	b_s6	b_s7	b_s8	bot1	bot2
bot1 + b_s1	96	85	99	15	77	74	98	16	5	2
bot1 + b_s2	94	85	99	13	81	77	97	10	5	2
bot1 + b_s3	84	70	99	49	62	60	98	68	5	6
bot1 + b_s4	73	59	99	60	53	53	96	81	5	10
bot1 + b_s5	99	95	99	23	95	94	99	14	17	5
bot1 + b_s6	99	95	99	22	95	94	99	14	16	5
bot1 + b_s7	99	94	100	33	93	91	100	41	14	5
bot1 + b_s8	71	54	99	46	45	45	95	69	3	5

Please refer to Table 10 for the definition of the scenarios bot1, bot2, and b_s1, …, b_s8. The FP rates have been adjusted to 0.05 under the training null scenario. The percentages should therefore be compared with the Acc* column in Table 10.