Table 3.

The main results were replicated using four different sets of models that included different sets of predictors

Model	Error	Difference	Balanced error	Balanced difference	Congruent balanced error and difference	Interaction of error and difficulty	Slope (β_e/β_d)
0–300 ms
Full	28	34	22	22	10	16	1.43
Full	24%	30%	19%	19%	9%	14%	p < 10⁻⁵
Drop	32	38	27	31	13	18	1.44
Drop	28%	33%	23%	27%	11%	16%	p < 10⁻⁵
Base	33	37	21	28	10	18	1.49
Base	29%	32%	18%	24%	9%	15%	p < 10⁻⁶
Reduced	35	37	19	33	18	20	1.53
Reduced	30%	32%	17%	29%	16%	17%	p < 10⁻⁵
0–200 ms
Full	27	36	18	24	7	13	1.63
Full	23%	31%	16%	21%	6%	11%	p < 10⁻⁴
Drop	28	38	19	28	8	16	1.62
Drop	24%	33%	17%	24%	7%	14%	p < 10⁻⁵
Base	29	38	20	28	8	21	1.61
Base	25%	33%	17%	24%	7%	18%	p < 10⁻⁵
Reduced	29	36	18	24	7	16	1.74
Reduced	25%	31%	16%	21%	6%	14%	p < 10⁻⁵
−400 to −200 ms
Full	6	7	3	8	0	8	−1.73
Full	5%	6%	3%	7%	0%	7%	Ns
Drop	7	4	4	8	0	9	−1.68
Drop	6%	3%	3%	7%	0%	8%	Ns
Base	6	5	4	7	0	8	−0.90
Base	5%	4%	3%	6%	0%	7%	p < 0.05
Reduced	4	7	5	6	0	7	−1.05
Reduced	3%	6%	4%	5%	0%	6%	p < 0.05

All numbers indicate how many cells passed a corresponding statistical test at α = 0.05. Bold indicate that this value is significantly higher than expected by chance (binomial test, p < 0.01, p0 = α; df = 114; p0 = α2/2 for the conjunction analysis of error and difficulty). The same analyses were conducted using two additional time epochs from 0 to 200 ms and from −400 to −200 ms. The 0–200 ms epoch replicates the main findings and shows that similar signals can also be found closer to saccade onset. The −400 to −200 ms epoch provides a negative control and shows that, as expected, none of the findings can be explained by elevated false-positive rate of any of the tests. Ns, Not significant.