Table 2. Goodness-of-fit metrics.

Model	Stimulus-evoked bias	Evidence	k	G 2	AIC	W
BurstIE	Burst	Increasing	14	43	71	0.47
SustIE	Sustained	Increasing	14	60	88	0.0001
BurstSE	Burst	Stationary	13	69	95	0
SustSE	Sustained	Stationary	13	89	115	0
Unbiased urgency slopes	Burst	Increasing	17	54	88	0.0001
Urgency-only bias	None	Increasing	11	362	384	0
DDM	None	Stationary	14	606	634	0
Constraints-Swap 1	Burst	Increasing	14	272	300	0
Constraints-Swap 2	Burst	Increasing	14	122	150	0
BurstIE + drift boost	Burst	Increasing	15	42	72	0.22
BurstIE + sZ	Burst	Increasing	15	42	72	0.31
SustIE + sZ	Sustained	Increasing	15	59	89	0
BurstSE + sZ	Burst	Stationary	14	69	97	0
SustSE + sZ	Sustained	Stationary	14	92	120	0
BurstSE + sTev	Burst	Stationary	14	64	92	0
SustSE + sTev	Sustained	Stationary	14	92	120	0

Goodness-of-fit quantified by chi-squared statistic, G². Model comparison was performed using AIC, which penalises for the number of free parameters (k). The Akaike weights (W) shown, which can be cautiously interpreted as the probability that each model is the best in the set, are calculated here based on the set of models in this table. The probability mass is shared between the different versions of the BurstIE model. In the two Constraints-Swap models, the constrained parameters for (A) high- coherence, (B) low- coherence, and (C) interleaved blocks were taken from the neural signals corresponding to [B,C,A] (Swap 1) and [C,A,B] (Swap 2), respectively.