Key words
IReST - eye tracker - eye movements - reading
Schlüsselwörter
IReST - Eye Tracker - Augenbewegungen - Lesen
Background
Accurate reading as well as reading comprehension are fundamental in todayʼs society. Reading is an acquired and complex task that requires a wide range of capabilities, strategies, knowledge
bases, and cognitive processes [1].
While reading a text, the eyes make a series of fast movements (saccades) separated by pauses (fixations). The saccades are typically forward movements of 7 – 9 characters where the eyes
fixate from one point to the next [2], but they occasionally move backwards (backward saccade) to refixate a previous point [3] – [7]. The assessment and process of reading has been evaluated extensively in competent adult readers, as well in readers with visual disabilities [5], [8]. Numerous investigators who evaluated reading skills have focused on reading speed as a measure of performance. The results of early studies have
shown that competent fast readers produce fewer and larger saccades, fewer fixations, fixations of shorter duration, and fewer regressions [9].
Only a few studies have evaluated the developmental aspects of eye movements during reading [5], [8]. Compared with adult readers,
children need more and smaller saccades, more and prolonged fixations and thus read slower [10] – [13]. There is also a big difference
in the recognition and perceptual span, which is the area where readers can acquire useful information around each fixation. Competent adult English readers typically have a perceptual span
ranging from 3 to 4 letter spaces to the left and 14 to 15 letter spaces to the right of the fixation [14], [15]. Several authors have
shown that the perceptual span is smaller in people learning to read and in dyslexic readers compared to proficient readers [11], [16] – [18].
Eye movements are affected in many diseases, which may influence reading ability. One of the most common learning disabilities is dyslexia, which is a phonetic disorder where, as a result,
individuals have difficulties with accurate or fluent reading. The prevalence of dyslexia is estimated to be 5 – 17% of school-aged children [19] and there is a
significantly higher male-to-female prevalence ratio (2 – 4 : 1) [20]. The eye movement behaviour in dyslexic children differs from competent readers in that they
need more and longer fixations as well as more saccades with shorter amplitudes and more backward saccades and thus need 2.7 – 2.9 times as long to read a given text [21], [22].
The International Reading Speed Texts (IReST) [23] test was used to assess the reading speed of normally sighted young children in this study. Ten texts were
designed for 17 languages and are matched in content, length, and difficulty. The IReST allows a standardised assessment of mean reading speed and variance that allows for the comparison of an
individualʼs reading performance, especially for comparing results while reading before and after interventions with different texts.
Abnormal eye movements while reading may be influenced by other factors such as age, gender, grade of school, or mother tongue [24]. In an earlier study, we showed
the feasibility of recording eye movements in 10- and 11-year-old children (N = 27) while reading [25]. In this current study, we recruited a large cohort of young
primary school children and analysed their eye movements across different primary schools and four different school grades. Furthermore, we investigated the influence of age, gender, school
level, and mother tongue on reading performance. Finally, we took all these factors into account and included data obtained from the initial feasibility study [25].
Methods and Material
Children were recruited from grades 2 – 5 from 4 primary schools of the German-speaking part of Switzerland. The study was approved by the Swiss Ethics Committees on research involving humans
and conducted in accordance with the Declaration of Helsinki. The class teachers were informed about the aim of the study and its purpose, and permission was obtained. The parents of the
participants or the legal guardian signed the written consent, and the children gave their informed consent orally. The children had the possibility of terminating the measurement at any
time.
Recruitment criteria for this study were: binocular visual acuity, if applicable with their current correction, > 80 Snellen, no known reading difficulties, no history of difficulty with
near vision, no history of vision impairment, and attending primary school. To screen visual impairments, all participants underwent an ophthalmological examination including binocular visual
acuity with the eye chart or with the Freiburg Vision Acuity Test (FrACT) [26], cover-uncover test at near to rule out manifest strabismus, and assessment of
stereopsis with the Lang II test.
Children were asked to read two texts that were presented in a randomised order in black letters on a white background (New Times Roman, font size of 26) on a laptop screen. These
standardised texts were developed by the IReST Study Group and are matched in content, length, difficulty, and linguistic complexity for 19 languages to assess reading speed in normally
sighted people [23]. We chose two texts of similar difficulty:
-
Text 1: “Biber” with 137 words, 224 syllables, and 681 characters
-
Text 2: “Beute” with 138 words, 232 syllables, and 671 characters
The examination was performed in a separate dark room of the primary school as described previously [25]. Briefly, the children sat comfortably at a desk on which
a laptop was placed (15-inch HP notebook, resolution: 1920 × 1080 pixels) at a distance of 50 to 80 cm. The children were instructed to read a text aloud as soon as it appeared on the laptop
screen and to not correct any mistakes nor go back in the text.
Eye movements were recorded with the SMI eye tracker RED 250 m (SensoMotoric Instruments, Teltow, Germany; http://www.smivision.com). The tracker bar was attached below the screen of the
laptop. All recordings were obtained binocularly with a sampling rate of 250 Hz and with a gaze position accuracy of 0.4° (tracking resolution: < 0.01 degrees of visual angle). Following a
five-point calibration routine [27], two IReST texts were presented in randomised order with a short break of 0.1 – 1 minute in between.
Analysis of the raw data were processed using the SMI software BeGaze, provided with the SMI RED 250 eye tracker (SensoMotoric Instruments, Teltow, Germany). We evaluated reading speed
(words/minute), total number of saccades, number of saccades per word, total number of fixations, number of fixations per word, total reading time, and reading errors. Gender, age, school
grade, and native language were also examined as possible factors that could influence outcome parameters. The number of mistakes made (words misread) were noted by the examinator and also
evaluated.
Descriptive statistics are shown as counts and frequencies for categorial data, and mean (standard deviation) and median (interquartile range) for metric variables. We employed the
Kruskal-Wallis test for means, and chi-squared or exact Fisher test when the expected frequencies were less than 5 in some cell. Overall p values correlate to the t-test for means and a p
value < 0.05 was considered significant. All statistical analysis described above were performed by using the statistical software program R version 3.1.1 (R Core Team, R Foundation for
Statistical Computing, Vienna, Austria, URL https://www.R-project.org/).
Results
One hundred and twenty-seven children volunteered to take part in this study. The age of the children ranged from 7 to 12 years, and all were recruited from the 2nd–5th grade of 4 different
primary schools in Switzerland (Zufikon AG, Pratteln BL, Birsfelden BL, and Rudolf Steiner Schule Basel). Nine children had to be excluded as the eye tracker could not measure the eye
movements due to reflections of the corrective glasses they wore. Thus, 118 children (mean age 9.8 SD 1.24, m = 58, f = 60) were included in the data analysis whose demographics are given in
[Table 1]. Overall, there was an equal distribution of gender with 60 female and 58 male children. Sixteen children were in 2nd grade, 29 in 3rd grade, 27 in 4th
grade, and 46 in 5th grade. Most children were aged 9 – 11, but only 6 children volunteered to participate at ages 7 and 12. Sixty-three children were German native speakers and 55 had another
mother tongue ([Fig. 1], p 0.15). Ophthalmic screening showed no visual impairments in any of the participants. The best-corrected Snellen distance visual acuity
measured using the Snellen visual acuity was over 1.0 in all but one child, who had an acuity of 0.80. In all individuals, the Lang II was positive and there was no manifest strabismus. No
significant difference was found between the parameters of the “Biber” and the “Beute” text. Therefore, we pooled the data from both texts and present our results in the form of one boxplot
for each parameter analysed.
Table 1 The table presents age and gender distribution of our children and shows the number of children excluded due to refractive errors per age (worn
glasses).
|
Age
|
|
7
|
8
|
9
|
10
|
11
|
12
|
|
|
N = 118
|
N = 6
|
N = 11
|
N = 26
|
N = 33
|
N = 36
|
N = 6
|
|
M
|
58 (49.2%)
|
5 (83.3%)
|
6 (54.5%)
|
14 (53.8%)
|
18 (54.5%)
|
12 (33.3%)
|
3 (50.0%)
|
|
F
|
60 (50.8%)
|
1 (16.7%)
|
5 (45.5%)
|
12 (46.2%)
|
15 (45.5%)
|
24 (66.7%)
|
3 (50.0%)
|
|
Excluded
|
9
|
1
|
1
|
|
4
|
3
|
|
Fig. 1 Distribution of non-native German speakers.
[Table 2] depicts an overview of the results obtained for children attending the 2nd–5th school grade. [Fig. 2] depicts the increase in
reading speed according to age ([Fig. 2 a]) and school grade ([Fig. 2 b]), and the influence of mother tongue ([Fig. 2 c]) and gender ([Fig. 2 d]) on the number of words read per minute.
Table 2 The table shows median normative data and 10th and 90th percentiles for each school grade tested.
|
School level
|
Percentiles
|
Reading speed (words/minute)
|
Number of saccades (saccades/word)
|
Number of fixations (fixations/word)
|
Reading errors (reading errors/word)
|
|
2nd class
|
10th
median
90th
|
29.92
50.13
74.27
|
0.88
2.52
10.35
|
0.82
2.21
4.74
|
0.02
0.05
0.10
|
|
3rd class
|
10th
median
90th
|
52.69
74.06
111.2
|
0.72
1.05
1.43
|
0.89
1.29
1.70
|
0.01
0.04
0.07
|
|
4th class
|
10th
median
90th
|
67.15
93.41
133.4
|
0.88
1.16
2.79
|
1.12
1.44
2.21
|
0.01
0.03
0.06
|
|
5th class
|
10th
median
90th
|
68.55
102.2
129.2
|
0.85
1.19
2.05
|
0.97
1.36
2.03
|
0.01
0.03
0.05
|
Fig. 2 Boxplots of reading speed in words/minute according to (a) age, (b) school level, (c) mother tongue, and (d) gender. For each boxplot, the bottom
of the rectangle indicates the first quartile, the top indicates the third quartile, and the horizontal line in the middle of the rectangle marks the median.
The higher the school grade, the faster the reading speed ([Fig. 2 b]). While 7- and 8-year-old children averaged 57.1 (SD 17.27) vs. 57.49 (SD 29.27) words per
minute, mean reading speed increased to 72.11 (SD 24.62) at 9 years of age, 92 (SD 26.24) at 10 years, and to 102 (SD 25.41) words/minute at 11 years of age. At 12 years of age, this was
slightly slower with 87.01 (SD 22.99) words/minute; however, this age group consisted of only 6 children ([Fig. 2 a]). The influence of mother tongue was not
statistically significant. However, [Fig. 2 c] shows that German native speakers read much faster in 2nd and 3rd grade but the reading speed of children with a
non-German mother tongue increased faster with school grade and they had caught up to native speakers by 5th grade. Overall, reading speed did not differ significantly between genders except
for a slower reading speed in girls aged 7 and 12; however, there were only 6 children in these age groups, and this was not statistically significant ([Fig. 2 d]).
[Fig. 3] shows boxplots of the number of saccades/word and [Fig. 4] of fixations/word according to ([Fig. 4 a]) age, ([Fig. 4 b]) school level, ([Fig. 4 c]) mother tongue, and ([Fig. 4 d]) gender.
Younger children aged 7 and 8 ([Fig. 3 a]) that is, the children in grade 2 ([Fig. 3 b]), needed significantly more saccades/word and
also fixations/word ([Fig. 4 a] and [b]) than the older children. From the age of 9 years and grade 3 onwards, the eye movement
parameters did not change significantly. Native German speakers who attend the 2nd grade needed significantly less saccades/word (p = 0.008) and fixations/word (p = 0.03; [Fig. 3 c] + [Fig. 4 c]). There was no gender difference in the number of saccades/word ([Fig. 3 d]) and
fixations/word ([Fig. 4 d]).
Fig. 3 Saccades/word according to (a) age, (b) school level, (c) mother tongue, and (d) gender. For each boxplot, the bottom of the rectangle indicates
the first quartile, the top indicates the third quartile, and the horizontal line in the middle of the rectangle marks the median.
Fig. 4 Fixations/word according to (a) age, (b) school level, (c) mother tongue, and (d) gender. For each boxplot, the bottom of the rectangle indicates
the first quartile, the top indicates the third quartile, and the horizontal line in the middle of the rectangle marks the median.
[Fig. 5] indicates the total number of errors, that is, mistakes made while reading out aloud. Children aged 7 and 8 made the most errors/word of 0.06 (SD 0.03;
[Fig. 5 a]). The total number of errors decreased with increasing age, apart from the 12-year-old children, who showed slightly more reading errors ([Fig. 5 a]). Consistently, children made significantly more mistakes in grade 2 compared to higher grades. Thereafter, the number of mistakes decreased slowly but not
significantly through to grade 5 ([Fig. 5 b]).
Fig. 5 Reading errors according to (a) age, (b) school level, (c) mother tongue, and (d) gender. For each boxplot, the bottom of the rectangle indicates
the first quartile, the top indicates the third quartile, and the horizontal line in the middle of the rectangle marks the median.
The relationship between reading errors/word and mother tongue is shown in [Fig. 5 c]. Second grade children with a foreign native language made significantly more
reading errors compared to German native speakers. Thereafter, a significant difference was no longer observed, as reading errors improved rapidly until 5th grade, where they made less
mistakes. Overall, there was no significant influence of gender ([Fig. 5 d]).
[Fig. 6] demonstrates the correlation between saccades and fixations per word, and between reading speed and reading errors per word. Children who made more
saccades also made more fixations. Finally, the faster the children read, the fewer mistakes they made.
Fig. 6 Correlation between saccades and fixations per word (a), as well the reading speed and reading errors per word (b).
Interestingly, we did not find any significant difference in reading parameters between the different schools, three of which were public schools and one a private school with an
anthroposophical teaching concept.
Discussion
Our main objective was to analyse the eye movement patterns during reading in a large cohort of 118 young children of primary school age (grades 2 – 5) in order to establish normative data
and assess the influence of age, school level, sex, and mother tongue. Although the IReST contains reading texts of a difficulty level for 6th grade, we intentionally chose to apply the same
reading text for all age groups to assess the reading performance of the same texts.
The main finding of this study is an increasing reading speed with age, which confirms the need for normative values in childhood dependent on age and school grade. Reading speed is
influenced by several factors, such as age, cognitive ability, reading skill, difficulty of the reading task, and reading aloud or silently. In our study, children of grade 2, who were
beginners in reading, showed the slowest reading with most saccades, fixations/word, and reading errors. With increased schooling, reading speed increased, the saccades and fixations needed
per word decreased, and errors diminished. The fastest mean reading speed was seen in children in 5th grade, with a median of 102.2 words/minute. This is still 1.75 times slower than in young
German adults (18 – 35) whose mean reading speed was 179 words/minute while reading an IReST text of the same difficulty (6th grade) [23]. Healthy German-speaking
children aged 9.6 ± 0.4 years (grades 4 and 5) performed at a mean reading speed of 119.56 (SD 17.60) words/minute [22] for an easier reading task from the
Salzburger Lese- und Rechtschreibtest (SLRT) [28], which contains material of third and fourth grade. The slight difference is likely due to the more difficult
reading task of IReST (6th grade) but may also be influenced by the different ages at which schooling begins in different countries (6 years in Germany compared to, at the time, 7 years for
average first graders in Switzerland).
The same study showed 9.6-year-old children to need between 1.37 (SD 0.20) saccades/word and 1.58 (SD 0.23) fixations/word in the 1st text and 1.21 (SD 0.90) saccades/word and 1.38 (SD 0.26)
fixations/word in the 2nd text [22]. The first more difficult text was age appropriate and designed for children in 3rd and 4th grade, while the 2nd text was easier
as it was designed for children in 1st and 2nd school grade. This compares to a mean number of 1.79 (SD 2.33) saccades/word and 1.62 (SD 0.97) fixations/word in our children (mean age 9.8
years, SD 1.24). Thus, even though the IReST we applied was designed for 6th grade, our children aged 9.6 (3rd grade) only needed 0.04 fixations/word more compared to an age appropriate SLRT
text used in the same age group. In this context, it is important to note that our saccades include both forward and backward saccades.
In the US, English-speaking students in the 4th grade showed a mean reading speed of 119 words/minute [29] for the reading task from the National Assessment of
Educational Progress (NAEP, 4th grade difficulty), which was faster than the reading speed in our 4th grade children (median 93.41 words/minute). Another study by Vagge et. al. [30] showed competent Italian readers aged 8 – 13 to use 1.51 (SD 0.50) saccades/word when reading a text silently. In this context, one has to mention that our children
read the IReST text aloud, which may have contributed to them needing more time. These results correspond with systematic reviews, which investigated eye movements in silent and oral reading.
In three studies, authors have reported significant differences between these two reading modes in adults. They have shown that fixation duration is about 50 ms longer, saccadic amplitude is
shorter, and the number of regressions is higher when reading out aloud [5], [8], [31]. As part of the initial
diagnostics to determine dyslexia, a reading test is performed. Various reading tests have been designed to make a statement of the quality of reading, reading comprehension, reading speed,
and reading errors. For example, the SLRT (Salzburger Lese- und Rechtschreibtest [28]) and the ZLT-II (Zürcher Lesetest [32]) are
reading tasks that the participants read aloud. Other reading tests such as Würzburger Leise Leseprobe [33] are performed silently. Although reading silently is
more in line with the natural reading situation, the advantage of reading aloud is that the investigator can evaluate the reading errors.
Overall, German native speakers read faster (94 s vs. 99.25 s) and needed fewer saccades/word (1.09 vs. 1.29) and fewer fixations/word (1.3 vs. 1.48) but made similar reading errors/word of
0.03. Non-native German-speaking children differed significantly from native speakers only in 2nd grade. They quickly improved their skills and there was no more difference by grade 5. An
explanation might be that these children attend additional German courses and therefore have more practice in reading.
A gender gap in reading abilities has been recognised and researched over the past couple of decades. A large study by Hedges and Nowell [34] found that girls
perform significantly better in reading than boys in each year of assessment (1971 – 1992). This gender gap has been confirmed in 31 countries and thus seems to be a cross-culture phenomenon
[35]. However, in our study, we found no significant influence of gender on reading performance.
Conclusion
This study was performed to establish normative data of eye movements in young children (grades 2 – 5, aged 7 – 12) while reading an IReST text. The number of saccades/word, fixations/word,
and reading errors decrease with age and school grade in children attending primary school. Thus, the reading speed is higher in older children compared to young beginners in reading. There
was no significant influence of gender. While native German speakers had an initial better performance, no significant difference was observed by grade 3. The results of our study indicate the
need for an age-appropriate normative database for eye movements in reading.
Conclusion Box
Already known:
Newly described:
-
We established normative data of eye movements in young children.
-
Native German speakers had an initial better performance, but no significant difference was observed by grade 3.
Data availability statement
Data availability statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.
