Table 2. Binary logistic GEE results for the Goal ROI (per time window of interest).

	By-participants (N = 43)							By-items (N = 36)
Window	Effect	Est.	SE		p	ΔQIC	Est.	SE		p	ΔQIC
Verb	Speed	−.017	.138	0.02	.90		−.021	.132	0.03	.87
	Time	+.009	.004	5.50	.02	−34.9	+.010	.004	6.78	.01	−176.7
	S×T	+.017	.008	3.99	.05		+.018	.006	5.64	.02
Path	Speed	+.424	.118	10.37	.001		+.422	.110	10.48	.001
	Time	+.033	.005	23.90	.001	−192.3	+.036	.005	25.87	.001	−325.1
	S×T	−.004	.006	0.35	.55		−.004	.006	0.44	.52
Goal	Speed	+.380	.133	6.97	.01		+.377	.104	9.32	.002
	Time	+.029	.010	7.91	.005	−66.5	+.030	.012	7.24	.01	−91.1
	S×T	−.018	.012	2.09	.15		−.017	.011	1.92	.17
Post-sent.	Speed	+.130	.092	1.91	.17		+.128	.098	1.63	.20
	Time	−.006	.004	3.38	.07	−2.9	−.007	.003	3.98	.05	−93.8
	S×T	−.001	.007	0.03	.88		−.001	.006	0.03	.87

Table shows GEE modelling results for occurrences of looks to the Goal ROI as a function of Verb Speed (fast vs. slow) and Time (continuous predictor referring to the 50 ms time-bins per window). Shown are the by-participant/by-item GEE parameter estimates and corresponding SEs (both in logit units) along with generalised score chi square statistics and related p-values (at df = 1). As for interactions (S×T), a positive parameter estimate indicates a more positive slope for the Time predictor in the fast than in the slow Verb Speed condition. ΔQIC refers to the goodness of fit of the Verb Speed×Time model in relation to a competitor Agent-Verb Suitability×Lexical Frequency×Time model (negative ΔQICs indicate superior fit of the Verb Speed×Time model).