Table 5. Fixed effects of optimal GLMM (Model 2’) for the probability of word skipping.

	Estimate	Std. Error	z value	Pr(>|z|)
(Intercept)	-0.69956	0.13511	-5.17778	< 0.00001
FREQ	0.06766	0.01676	4.03729	0.00005
LENGTH	-0.69301	0.04295	-16.13351	< 0.00001
LAUNCH	0.61995	0.04408	14.06404	< 0.00001
FREQ:LENGTH	-0.06234	0.01041	-5.98644	< 0.00001
FREQ:LAUNCH	-0.00768	0.01026	-0.74814	0.45438
LENGTH:LAUNCH	0.01414	0.01788	0.79096	0.42897
FREQ:LENGTH:LAUNCH	0.01349	0.00608	2.21757	0.02658

This analysis was conducted across all words in the sentences that responded to our selection criteria (see Materials and Methods). The fixed structure included the effects of word length (“LENGTH”; 4–8 letters), word frequency (“FREQ”; between 0.01 and 9.02 log units), and saccadic launch-site distance (“LAUNCH”; between -6.00 and -0.002 letters from the space in front of the test words), as well as all interactions; the random structure included a random intercept by participant, sentence pair, and word, as well as by-participant random effects of word length and saccadic launch-site distance (see S1 Table). The model's estimates and standard errors are expressed in logit units. The intercept estimate (logit: -0.69956) indicates that the words were skipped in about 33% of the cases, when all variables were at their reference, mean, value (Word Length: 5.60 letters; Launch Site: -2.40 letters; Word Frequency: 4.33 log units). Colon stands for interaction. See S4 Table for the corresponding minimalist optimal GLMM.