TABLE I.

Comparison of model validation performance using the bubbles-derived decision variable versus the STMI. Percent agreement between predicted and true responses displayed at left, binned by true number of keywords correct (as in Fig. 11). Kendall's tau computed between predicted and true responses displayed at right. Table entries reflect the group mean with standard error of the mean in parentheses below. UC = uncompressed speech group, 2× = time-compressed speech group, ORIG = model performance determined using the bubbles-based decision variable to generate predicted responses, STMI = model performance determined using the STMI to generate predicted responses.

	Percent agreement
	Number of keywords correct
	0	1	2	3	4	5	Kendall's τ
UC-ORIG	66.2 (2.0)	12.5 (1.9)	10.4 (1.9)	13.2 (2.2)	18.3 (2.0)	61.8 (2.0)	0.53 (0.01)
UC-STMI	75.6 (2.2)	15.6 (2.9)	15.3 (2.6)	19.1 (2.3)	18.7 (1.8)	65.3 (2.8)	0.63 (0.02)
2×-ORIG	59.4 (1.6)	13.6 (1.8)	15.4 (1.3)	14.4 (1.5)	20.3 (3.4)	45.6 (2.7)	0.45 (0.01)
2×-STMI	67.0 (1.6)	18.5 (1.9)	14.7 (1.0)	14.3 (1.9)	20.1 (2.6)	44.3 (2.4)	0.48 (0.02)