Table 1.

Model parameters.

	Model parameters
Model	Gain of sensory response (g)	Response to stimulus noise (kn)	Strength of divisive normalization (ks)	(Population) Fano factor (kv)	Fixed variance	Offset	(Effective) integration time constant (ms; τ or τeff)	Decision threshold	Residual time (ms; tres)	Remaining error after mean RT fit	Percentage of predicted choice data points within 95% confidence intervals	Quality of RT distribution match: mean intersection/mean fidelity
Slightly leaky feedforward inhibition model (Figure 1)	0.0101	0.120	2.01	0.332	N/A	N/A	714 (imposed; chosen to match feedback inhibition model)	1 (imposed)	353	97 (49)	87	0.858/0.984
Very leaky feedforward inhibition model (Figure 2)	0.0315	0.135	1.83	0.675	N/A	N/A	50 (imposed)	1 (imposed)	477	80 (50)	83	0.822/0.926
Feedback inhibition (LCA) model with scaling variance (Figure 4)	0.0117	0.0928	2.78	0.417	N/A	N/A	714(b = 0.00140 for a time unit of 1 ms)	1 (imposed)	404	84 (75)	65	0.829/0.935
Feedback inhibition (LCA) model with fixed variance (Figure 5)	0.0131	0.0967	2.29	N/A	0.000407	N/A	694(b = 0.00144 for a time unit of 1 ms)	1 (imposed)	392	61 (44)	74	0.751/0.920
Feedforward MSPRT model (Figure 6)	0.0192	0.0975	2.08	0.434	N/A	N/A	∞ (imposed)	−0.383	390	92 (65)	61	0.773/0.928
Feedback MSPRT model (Figure 8)	0.0174	0.101	2.19	0.696	N/A	1.47	∞ (imposed)	1 (imposed)	447	81 (69)	57	0.841/0.926

Parameters in bold have been determined through optimization. The first number in the “Remaining error” column indicates the remaining error after a high-resolution simulation (50,000 trials per experimental condition) of the model with the indicated parameter set. The second number in brackets indicates the smallest error that was seen during the optimization process based on a lower-resolution simulation (10,000 trials per condition).