Scanpath Regularity as an Index of Reading Comprehension

Figures & data

Figure 1. Example scanpaths for sentence and passage reading.

Figure 1 shows examples of highly regular and less regular scanpaths when reading single sentences (a) and passages of text (b). The Figure illustrates the overall fixation patterns of readers, and that scanpaths can be irregular in different ways, such as multiple re-readings of the sentences (a) or zig-zagging patterns in the reading of passages, with varying fixation durations (b).

Figure 2. Examples of scanpath variance maps.

Figure 2 shows examples of two types of umaps illustrating scanpath dissimilarity from the multi-dimensional maps of scanpath space, which showed either clear cluster separation (a) or a cloud-like distribution of scanpaths (b). Only the first two components of the maps, which explain the most amount of variance, are shown in the figure. V1 = first component; V2 = second component.

Table 1. Summary statistics of comprehension scores.

Test	Mean (SD)	Range
YARC	90.25 (13.5)	70–127
GORT	90.35 (9.43)	70–110
WRAT	105.78 (12.07)	82–131

Figure 3. Examples of scanpaths across tests and comprehension levels.

Figure 3 shows example scanpaths of three participants for each test, including above average (A), average (B), and below average (C) comprehenders.

Figure 4. Examples of typical and atypical scanpaths.

Figure 4 shows the most typical (a) and most atypical (b) scanpaths for each test and for clusters 1 (red) and 2 (blue).

Figure 5. Stability of predictors across measures.

Figure 5 shows the model estimates from single regression models and illustrate the stability of the relationship between individual predictors and comprehension scores across tests. The dotted line equals 0, indicating no relationship between the predictor and comprehension score.

Table 2. Estimates from single-predictor regressions.

	YARC			GORT			WRAT			Average
Predictor	$\hat{\beta }$	R^2	95% CrI	$\hat{\beta }$	R^2	95% CrI	$\hat{\beta }$	R^2	95% CrI	$\hat{\beta }$	R^2	95% CrI
Density	3.6*	0.07	[0.6, 6.6]	1.8	0.04	[−0.5, 4.1]	4.1*	0.12	[1.5, 6.8]	3.8*	0.15	[1.7, 5.9]
Dissimilarity	−2.2	0.03	[−5.3, 1.0]	−2.8*	0.02	[−4.9, −0.7]	−5.8*	0.23	[−8.2, −3.4]	−3.1*	0.10	[−5.3, −1.0]
Cluster Membership	−2.4	0.03	[−5.5, 0.8]	−2.5*	0.07	[−4.7, −0.4]	−5.6*	0.21	[−8.2, −3.2]	−3.5*	0.13	[−1.5, −6.8]
Average Fixation Duration	−2.3	0.03	[−5.3, 0.8]	−1.0	0.01	[−3.2, 1.2]	−3.3*	0.07	[−5.9, −0.7]	−2.1	0.04	[−4.3, 0.1]
Forward Saccade Length	2.1	0.02	[−1.0, 5.1]	3.9*	0.17	[1.8, 6.0]	1.2	0.07	[−1.5, 3.9]	2.4*	0.06	[0.3, 4.5]
Skipping	−0.1	0.01	[−3.4, 3.1]	0.8	0.01	[−1.4, 3.1]	1.9	0.02	[−0.9, 4.8]	0.7	0.01	[−1.6, 2.9]
First-Fixation Duration	−2.5	0.03	[−5.4, 0.5]	−0.7	0.01	[−2.9, 1.6]	−3.1*	0.07	[−5.7, −0.5]	−2.1	0.04	[−4.3, 0.1]
Gaze Duration	−3.4*	0.06	[−6.4, −0.5]	−3.3*	0.12	[−5.4, 1.2]	−3.4*	0.08	[−6.1, −0.7]	−3.2*	0.11	[−5.3, −1.2]
Regressions	−0.5	0.01	[−3.6, 2.5]	1.1	0.01	[−1.2, 3.4]	−1.1	0.01	[−3.7, 1.7]	−0.3	0.01	[−2.6, 2.0]
Go-Past Time	−1.1	0.01	[−4.1, 2.0]	−2.8*	0.09	[−4.9, −0.7]	−5.1*	0.18	[−7.6, −2.5]	−3.3*	0.11	[−5.1, −1.1]
Total Reading Time	−0.8	0.01	[−4.0, 2.4]	−3.4*	0.13	[−5.4, −1.3]	−5.9*	0.24	[−8.3, −3.5]	−4.0*	0.17	[−6.0, −2.0]

This table shows the model estimate (), r-squared (R2), and 95% credible interval (95% CrI) of models with single predictors. * = 95% interval does not include 0. The best predictor for each test is indicated in bold.

Table 3. YARC: intercepts and estimates of the best ten models and full model.

Predictor	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8	Model 9	Model 10	Full Model
Intercept	90.7**	90.7**	90.6**	90.7**	90.6**	90.6**	90.6**	90.7**	90.6**	90.7**	90.7**
Density	3.4**	4.0**		3.0*	3.1*	3.4**	2.7	3.3**	3.4**		2.2
Dissimilarity											3.3
Cluster Membership			−3.6				−2.0			−4.6*	−4.2
Average Fixation Duration	8.8**	9.0**	7.7*	9.3**		9.9	8.0**	8.9**	8.8**		6.8
Forward Saccade Length								0.1		5.8**	3.4
Skipping									0.4	−6.0**	−4.8*
First-Fixation Duration					8.7**	−1.3					0.2
Gaze-Duration	−15.4**	−16.2**	−14.9**	−16.5**	−15.3**	−15.1**	−14.8**	−15.5**	−15.5**	−9.1**	−16.5**
Regressions				−1.5						−5.6**	−4.6*
Go-Past		2.4								5.4*	4.8
Total Time	7.1**	6.1**	8.6**	7.7**	6.9**	7.0**	8.3**	7.1**	7.2**	7.8**	6.5*
R2 Bayes	0.23	0.24	0.21	0.24	0.22	0.23	0.24	0.23	0.23	0.27	0.31
ELPD-LOO	−305.58	−306.13	−306.29	−306.33	−306.50	−306.51	−306.51	−306.56	−306.56	−306.75	−310.75

Models 1–10 are ordered according to their elpd-LOO (descending). ** and bold = 95% credibility interval does not include 0; * = 90% credibility interval does not include 0; blank: predictor not included in the model.

Table 4. GORT: intercepts and estimates of the best ten models and full model.

Predictor	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8	Model 9	Model 10	Full Model
Intercept	90.5**	90.5**	90.5**	90.5**	90.5**	90.6**	90.5**	90.5**	90.5**	90.5**	90.5**
Density		−0.3									−0.2
Dissimilarity	5.7**	5.4*	5.5**	5.6**	6.2**	6.3**		5.7**			5.8
Cluster Membership						−0.5	2.5				−0.3
Average Fixation Duration	−10.9**	−11.0**	−10.4**	−10.6**	−10.4**	−11.0**	−10.6**	−10.1**	−11.6**	−10.6**	−10.0
Forward Saccade Length	4.0**	3.9**	3.7**	3.7**	4.2**	3.9**	4.1**	4.6**	3.9**	2.6**	4.4**
Skipping								−1.1			−1.1
First-Fixation Duration	13.8**	13.9**	13.6**	13.8**	13.5**	14.0**	12.6**	13.1**	12.5**	13.7**	13.5**
Gaze-Duration				−0.6						−4.8**	−1.5
Regressions			0.4								−0.4
Go-Past					−1.8						−0.5
Total Time	−9.3**	−9.3**	−9.4**	−9.0**	8.3**	−9.5**	−5.6**	−9.1**	−3.1**		−7.6
R2 Bayes	0.39	0.40	0.39	0.40	0.40	0.40	0.37	0.40	0.35	0.34	0.41
ELPD-LOO	−241.76	−242.59	−242.65	−242.67	−242.73	−242.81	−242.84	−242.89	−243.00	−243.51	−248.60

Models 1–10 are ordered according to their elpd-LOO (descending). ** and bold = 95% credibility interval does not include 0; * = 90% credibility interval does not include 0; blank: predictor not included in the model.

Table 5. WRAT: intercepts and estimates of the best ten models and full model.

Predictors	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8	Model 9	Model 10	Full Model
Intercept	105.9**	105.9**	105.9**	105.9**	105.9**	105.9**	105.9**	105.9**	105.9**	105.9**	105.9**
Density											0.0
Dissimilarity	−5.2**			−4.2**			−4.1**				−1.1
Cluster Membership											−3.5
Average Fixation Duration								−4.4			−1.5
Forward Saccade Length								2.1			3.1
Skipping											−1.9
First-Fixation Duration							2.2		2.8		−4.0
Gaze Duration	3.1	3.7						7.4		3.1	9.6
Regressions	4.5**	4.0**	2.1	3.4**		2.9*	4.6**		4.2**	2.7	2.5
Go-Past Time	−6.0**	−4.1**		−3.9		−2.5	−6.2**		−4.8		−3.5
Total Reading Time		−6.9**	−6.9**		−5.9**	−5.1		−8.6**	−5.4**	−9.5**	−3.9
R2 Bayes	0.31	0.32	0.21	0.29	0.24	0.28	0.31	0.31	0.31	0.30	0.36
ELPD-LOO	−289.22	−289.69	−289.70	−289.86	−289.87	−290.08	−290.14	−290.15	−290.24	−290.24	−296.48

Models 1–10 are ordered according to their elpd-LOO (descending). ** and bold = 95% credibility interval does not include 0; * = 90% credibility interval does not include 0; blank: predictor not included in the model.

Table 6. Average: intercepts and estimates of the best ten models and full model.

Predictor	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8	Model 9	Model 10	Full Model
Intercept	95.6**	95.6**	95.6**	95.6**	95.6**	95.6**	95.6**	95.6**	95.6**	95.6**	95.6**
Density	3.4**	3.2**	3.1**	3.5**	3.0**	3.4**	2.8*	3.4**	3.2**	3.2**	3.5**
Dissimilarity	15.6**	10.7**	10.8**	10.0**	14.5**	13.5**	15.8**	10.9**	13.2**	10.8**	15.9**
Cluster	−6.7**				−4.3	−4.1	−6.9**		−6.1**		−5.7
Average Fixation Duration								2.7*		3.0*	−1.8
Forward Saccade Length	5.7**	2.1**	3.3**	3.6**	3.1**	3.5**	4.4**	2.1*		3.1**	4.9**
Skipping	−5.1**		−1.7	−3.3*	−2.2	−3.6**	−2.6*			−1.6	−4.8**
First-Fixation Duration		3.0**	3.3**	3.0*	2.9*	2.5*					5.0
Gaze-Duration											−3.0
Regressions	−3.0*						−2.2				−2.4
Go-Past	6.7**			4.8		4.7					6.2*
Total Time	−15.1**	−13.4**	−13.8**	−16.9**	−13.3**	−16.3**	−9.5**	−13.1**	−8.7**	−13.4**	−15.5**
R2 Bayes	0.39	0.33	0.35	0.36	0.36	0.38	0.35	0.33	0.29	0.34	0.40
ELPD-LOO	−277.49	−277.87	−278.04	−278.07	−278.10	−278.18	−278.19	−278.20	−278.22	−278.32	−280.32

Models 1–10 are ordered according to their elpd-LOO (descending). ** and bold = 95% credibility interval does not include 0; * = 90% credibility interval does not include 0; blank: predictor not included in the model.