Eye movements during oral and silent reading

(1)

Eye Movements during Oral and Silent Reading

Merel C. Wolf

University of Amsterdam

Faculty of Social and Behavioural Sciences Graduate School of Child Development and Education

Research Master Child Development and Education Research Master Thesis

Supervisors: dr. Madelon van den Boer & prof. dr. Peter de Jong

(2)

Preface

This master thesis was written from September 2016 to June 2017 as part of the Research Master Child Development and Education at the University of Amsterdam. This personal project, an extension of my bachelor’s thesis, was set up under supervision of Madelon van den Boer and Peter de Jong.

I would like to thank Peter de Jong, who sparked my enthusiasm for research on reading and language during the UvA Master’s Day in November 2014 and participated in the writing process of my bachelor’s thesis as well as my two master theses. Your knowledge about reading and experience with language experiments helped me design the experiment and your analytical and observant nature aided me in looking objectively at the project and reading my own writing with a critical eye.

I also want to thank Madelon van den Boer, who helped me set up this project to find answers to the questions that we could not yet answer after finishing my bachelor’s thesis. Your quick thinking, your knowledge on the subject, your crystal-clear writing style and especially your enthusiasm regarding research inspired me to pursue an academic career by (successfully) applying to a PhD position at the Max Planck Institute for Psycholinguistics.

I thank you both for supporting my academic and personal growth and I look forward to meeting again, but next time as fellow researchers!

Merel Wolf

(3)

Abstract

Children’s reading fluency and comprehension are often measured orally, but in class and during academic achievement tests they are required to read silently. Previous studies suggest oral reading to be slower, but yield better comprehension than silent reading. Oral and silent reading processes are, however, not yet understood to the extent that speed and comprehension differences can be explained. This study aims to meticulously investigate the differences between oral and silent reading processes in children using eye tracking methodology. Thirty-six third grade children’s eye movements were registered during a sentence-verification task. Word frequency and sentence congruency were manipulated to examine differences in word decoding and sentence comprehension between oral and silent reading. Moreover, individual differences in oral and silent reading were explored by examining the relation of various eye movement variables with several underlying skills of reading, namely phonological awareness (PA), rapid naming (RAN) and visual attention span (VAS). It was found that silent reading was characterized by fewer fixations and regressions. Silent reading was found to be the faster reading mode. Regarding comprehension, there was no effect of reading mode, but an interaction was found with sentence congruency. Concerning the contributions of PA, RAN and VAS, trends were similar to previous studies in that PA contributed to both reading modes, RAN contributed stronger to oral reading and VAS contributed stronger to silent reading. The present study implies that generalizing oral reading performance to silent reading ability and vice versa is not advisable.

Keywords: oral reading, silent reading, eye movements, decoding, reading comprehension, phonological awareness, rapid naming, visual attention span

(4)

Eye Movements during Oral and Silent Reading

Literacy may be the most important skill that allows individuals to function in our modern, information-driven, digital society. Therefore, learning to read is one of the most important cornerstones of education worldwide. Silent reading is the predominant reading mode to practice reading and to use reading during other learning tasks. Moreover, it is the most natural reading mode for proficient readers. Remarkably, reading development is almost always assessed orally, be it in research, diagnostics or education. However, the generalizability of oral reading performance to silent reading ability and vice versa cannot be assessed as of yet, because it is still unclear to what extent oral and silent reading differ due to the limited number of studies that have investigated this subject. Therefore, the present study examines differences between oral and silent reading in children.

Previous research has shown that oral and silent reading have different reading outcomes. With respect to reading speed, it has been found that children read faster silently (McCallum, Sharp, Bell & George, 2004; van den Boer, van Bergen & de Jong, 2014). Regarding reading comprehension, children up to fifth grade seem to comprehend texts better while reading orally (Kragler, 1995; de Jong & Share, 2007), but from sixth grade up to adulthood, silent reading yields the best comprehension (Prior et al., 2011). Thus, oral reading tests of reading fluency might give a distorted image of a young child’s functional (silent) reading speed. Moreover, silent reading comprehension tests might not be optimal to reflect or foster a child’s comprehension abilities. It is therefore necessary to meticulously investigate the different processes of oral and silent reading, which can aid in explaining the reading speed and comprehension outcomes. Previous studies’ methodologies did now allow to carefully investigate the processes of oral and silent reading. The present study, however, uses eye tracking methodology to investigate differences in the reading processes of oral and silent reading in detail.

(5)

A body of literature is dedicated to the difference in reading outcomes (e.g., reading speed and comprehension) between oral and silent reading, but the difference in reading processes is relatively unexplored. This is a result of studies using offline measures, such as the time needed to read texts orally or silently or the number of correctly answered comprehension questions. Such measures describe the outcome of the reading process, but not the process itself. Therefore, online reading measures, that measure reading as the reader progresses through text, might provide a more detailed description of differential reading processes between the two modes. Ultimately, differential reading processes could explain the differences in reading outcomes (speed and comprehension).

An online method to investigate the process of reading as it occurs is eye tracking. Examining eye movements during reading provides detailed information about the way readers process text (Rayner, 1998). During reading, the eyes fixate on some words in the text (fixations) and sometimes look back at words read earlier (regressions). Only two studies investigated eye movements during oral and silent reading in children or adolescents (Vorstius, Radach & Lonigan, 2014; Krieber et al., 2017). They both found a very distinctive difference in eye movement patterns between oral and silent reading. Oral readin g is characterized by more fixations, longer first fixation durations, longer average fixation durations, smaller saccadic amplitudes, fewer regressions and individual words having higher probabilities to be fixated upon (Vorstius et al., 2014; Krieber et al., 2017). In contrast, the opposite pattern with fewer and shorter fixations, larger saccadic amplitudes and more regressions are present during silent reading. Altogether, these findings indicate a distinct eye movement pattern for each reading modes (Figure 1). Oral reading seems to happen word-by-word. Most words are read and read thoroughly, hence the increase in fixations and fixation durations, higher fixation probability and smaller saccadic amplitudes. In contrast, the silent reading eye movement pattern seems “jumpy”: more words are skipped but also more

(6)

regressions are made. Thus, there is some evidence from eye tracking studies indicating that oral and silent reading might have different eye movement patterns, which could mean that the reading process of the two modes differs. Further investigation is, however, necessary, due to the limited number of studies.

Figure 1. The difference in eye movement patterns between oral (upper sentence) and silent reading (lower sentence) as found by Vorstius et al. (2014) and Krieber et al. (2017).

Difference in Reading Speed

The differential eye movement patterns of silent and oral reading shed light on the different processes of oral and silent reading. These different processes might explain differences in reading speed between the two reading modes. The slower reading speed during oral reading might be explained by the reader carefully reading (almost) every word because the words have to be pronounced. Eye tracking studies confirm that individual words are more focused upon in oral reading (i.e., more and longer fixations (Vorstius et al., 2014; Krieber et al., 2017)). Whether the longer word reading times are caused by the pronunciation of the words has not been investigated yet. Nonetheless, there is evidence that the preparation of pronunciation and the pronunciation itself are important aspects of oral reading and that this focus on pronunciations might slow down reading. For example, it has been found that rapid naming, i.e. the ability to quickly name familiar stimuli such as letters and digits, contributes more strongly to oral reading than silent reading (van den Boer et al., 2014). This might

(7)

indicate that the retrieval of orthographical and phonological characteristics of words is more important for oral reading, as well as the computation of a phonetic code and the pronunciation of this code. There is also evidence that pronunciation decreases response speed. In a study by Huestegge, Pieczykolan and Koch (2014) participants were told to respond to one particular rectangle out of two rectangles on a screen. They found that articulatory (vocal) responses were slower than oculomotor (eye movement) responses and that a dual (both articulatory and oculomotor) response was even slower. Thus, pronunciation could slow down reading and reading might even further slowed down because of the dual task (articulatory and oculomotor) being performed during oral reading.

For silent reading the eye movement pattern also sheds light on the underlying reading process. The faster reading speed during silent reading might be explained by the reader spending less time on individual words because they do not have to be pronounced. Eye movement studies confirm that during silent reading more words are skipped (i.e., less and shorter fixations, larger saccadic amplitude (Vorstius et al., 2014; Krieber et al., 2017)). Because words do not have to be pronounced, there is no delay in reading speed due to pronunciation itself or the cognitive load of a dual (articulatory and oculomotor) response (Huestegge et al., 2014). Reading speed might even more sped up because silent readers can process upcoming information more effectively. When fixating on a word, the reader can see up to 5° of the visual angle (15 – 20 characters) to the left and right of the fixation point. This area is known as the parafoveal region (Engbert, Longtinn & Kliegl, 2002). The availability of parafoveal information increases reading speed and reduces the number and duration of fixations. This effect is even stronger for silent reading than for oral reading (Ashby, Yang, Evans & Rayner, 2012). The authors argue that the additional phonological working memory processes needed for pronunciation during oral reading restricts the cognitive resources available for parafoveal processing. Thus, during silent reading parafoveal information is

(8)

processed more effectively, presumably due to the lack of the cognitive restraint produced by the pronunciation process, allowing a more “jumpy” eye movement pattern and ultimately speeding up reading time. Another finding that confirms this idea is that visual attention span, the number of orthographic units that can be processed at a glance, seems to contribute more strongly to silent reading than oral reading (van den Boer et al., 2014). It makes sense that having a larger visual attention span fosters silent reading more as it aids in parafoveal processing, which is an important aspect of silent reading.

It should be noted that the fact that articulatory processes are not present during silent reading does not mean that phonological characteristics of words are not activated. For example, several studies show that phonological awareness, the skill to identify and manipulate phonemes in words, contributes to oral (Vellutino, Fletcher, Snowling, & Scanlon, 2004) and silent reading (Bar-Kochva, 2013) and a comparison between the reading modes found equal contributions (van den Boer et al., 2014). Moreover, silent reading speed is affected by the phonological complexity of the words (Haber & Haber, 1982) in such a way that phonological complex words (e.g., tongue-twisters) are read slower than words matched on syntactic complexity, syllable count, and sentential stress pattern. These findings can be explained by strong phonological theories of reading (Frost, 1998; Perfetti & Hart, 2002). Such theories hypothesize that phonological, orthographical and semantic features of words are integrated so closely that upon the encounter of a word during reading all these features are activated automatically. For oral reading, these phonological representations are further used to compute the phonetic code which is used in the process of speech production (Levelt, 1992). For silent reading, the phonological representations are activated, but there is no further processing towards phonological codes. Thus, in both reading modes phonological representations of words might be activated, but these only cause cognitive load during oral reading because then they are used to prepare speech.

(9)

Difference in Reading Comprehension

It is not clear how different reading processes, indicated by different eye movement patterns, can explain why oral reading is the most effective reading mode for comprehension for children up to sixth grade. Silent reading is characterized by higher reading speed, less fixations and more regressions (Vorstius et al., 2014; Krieber et al., 2017). As a result of the lower number of fixations and lower fixation durations, important words could be skipped or not read long enough to fully process their meaning. So at the end of a sentence, the reader missed important information which explains why comprehension deteriorates during silent reading. However, silent reading is also characterized by more regressions, which could be used to reread information missed at first pass reading (Reichle, Warren, & Connell, 2009). Since silent reading is the less effective reading mode for comprehension, it appears that these regressions are not effective. A characteristic of oral reading also seems to contradict the finding that oral reading is the most effective reading mode for comprehension for children. It was mentioned before that oral reading is probably slowed down partly due to the cognitive load of a dual (articulatory and oculomotor) task that is being performed during oral reading. It would make sense that this cognitive load would also interfere with comprehension processes and in effect would deteriorate comprehension. Nonetheless, oral reading is the most effective reading mode comprehension for young children. Prior et al. (2011) argue that oral reading ensures that readers do not skip words important for comprehension and that vocalization helps to keep information in memory while creating a coherent mental picture of the text. As of yet, it seems that the comprehension processes of oral and silent reading are not yet understood to the extent that the comprehension difference can be explained.

Present Study

Previous studies have not yet been able to fully explain why in children oral reading is slower, but yields better text comprehension than silent reading. Until very recently, oral and

(10)

silent reading have only been studied with offline measures of reading, which do not provide information about differential processes during oral and silent reading. Investigating the process of oral and silent reading is very valuable as it can provide important insights into the differences in reading speed and comprehension between the two reading modes. Eye tracking is a method to explore reading processes as they occur, but this is has only been used twice in research comparing oral and silent reading. Moreover, only once eye movement measures have been used to investigate reading processes in children who are not yet fluent readers (Vorstius et al., 2014). Therefore, the present study uses eye-track methodology to investigate the difference in reading outcomes between the two modes and to explore differential eye movement patterns of oral and silent reading to uncover differences in reading processes. Moreover, through an experimental design, two aspects of the reading process, namely decoding processes and comprehension processes, are explored separately. The present study focuses on sentence reading. Decoding processes are looked at on the word level and comprehension processes on the sentence level. Differential eye movements during the word decoding process are investigated by manipulating word frequency of a target word at the beginning of the sentence. Differential eye movements during the comprehension process are investigated using a sentence verification task and manipulating congruency of the sentences. Decoding and comprehension processes during oral and silent reading are studied by comparing several eye movement parameters across reading mode, word decoding and comprehension conditions.

In addition, the present study aims to extend research on individual differences in oral and silent reading. Therefore, it is investigated to what extent underlying cognitive skills such as phonological awareness, rapid naming and visual attention span contribute to these eye movement measures of oral and silent reading. The contribution of vocabulary is also

(11)

examined, as vocabulary is known to be a strong contributor to reading comprehension ( Swart et al., 2016).

It is firstly hypothesized that silent reading is the faster reading mode (McCallum et al., 2004; van den Boer et al., 2014; Vorstius et al., 2014; Krieber et al., 2017), shown by higher sentence reading speed. Also, infrequent target words during silent reading are expected to be read with a higher reading speed than during oral reading. These words require more decoding in order to be pronounced and reading speed will thus be slowed down during oral reading, but less so during silent reading.

Second, in line with Kragler (1995) and de Jong and Share (2007), it is hypothesised that oral reading yields the best comprehension. This is indicated by a higher percentage correct on the sentence verification task during oral reading. Also, higher reading speed on the target word that manipulates sentence congruency during oral reading than during silent reading is expected: since oral reading yields the best comprehension, the words that determine comprehension will be processed faster.

Third, it is hypothesized that, in line with Vorstius et al. (2014) and Krieber et al. (2017), eye movements differ during oral and silent reading to the extent that oral reading is associated with more fixations and fewer regressions and silent reading is characterized by the opposite pattern with fewer fixations and more regressions.

Fourth, it is hypothesized that phonological awareness contributes to both oral and silent reading (Vellutino et al., 2004; Bar-Kochva, 2013; van den Boer et al., 2014), that rapid naming contributes more strongly to oral reading and that visual attention span contributes more strongly to silent reading (van den Boer et al., 2014).

(12)

Methods Participants

Thirty-six third graders (20 boys) with a mean age of 8 years and 7 months (SD = 4.51 months) participated in this study. Two children spoke another language aside from Dutch at home (Chinese or Bulgarian). The sample included children with a representative range in reading abilities, as they scored 10.29 (SD = 2.66) on average on a standardized word reading fluency task (Brus & Voeten, 1979), with an average of 10 and standard deviation of 3.

Design

A repeated measures design was used in this study. Children were required to read sentences both orally and silently. Therefore, two sets of stimuli were created. The children read one set orally and one set silently. To control for order effects, set order and reading mode were counterbalanced. This resulted in four order conditions. Group 1 read set 1 orally and then set 2 silently. Group 2 read set 1 silently and then set 2 aloud. Group 3 first read set 2 orally and then set 1 silently. Group 4 read set 2 silently first and then read set 1 aloud. The four order condition groups were randomly matched on reading ability, to ensure that the participants in each order condition had similar reading skills. The four groups did not significantly differ regarding gender (χ2(3) = 0.90, p = .83), age (F(3, 35) = 0.52, p = .67) or reading ability (F(3, 35) = 0.08, p = .97). Table 1 displays descriptive statistics of the four order conditions.

Table 1

Number of Boys, Age and Reading Ability of the four Order Conditions

Group 1 Group 2 Group 3 Group 4

# Boys 6 5 5 4

Agea (SD) 8;7 (5.55) 8;10 (3.16) 8;10 (5.76) 8;8 (2.96) RA (SD) 10.00 (3.12) 10.56 (2.30) 10.17 (3.30) 10.44 (2.19) Note. RA = Reading Ability norm score.

a

(13)

Material

Oral and Silent Reading.

Material. Word decoding and sentence comprehension during oral and silent reading were measured with an experimental task. The stimuli consisted of 40 sentence pairs. Each sentence pair consisted of fifteen words: six in the first and seven words in the second sentence. An example is displayed in Figure 2. Two variables were manipulated. Word decoding was manipulated through the frequency of a target word in the first sentence (Target Word 1). Frequency was determined by a database of words from Dutch child literature (Schrooten & Vermeer, 1994). Target Word 1 was frequent (e.g., man [man]) or infrequent (e.g., tandarts [dentist]). Reading comprehension was manipulated through the last word in the second sentence (Target Word 2), which caused the sentences to be congruent or incongruent. Figure 2 displays an example of two sentence pairs. The first sentence pair has an infrequent Target Word 1 and congruent Target Word 2. The second sentence pair has a frequent Target Word 1 and an incongruent Target Word 2.

Figure 2. Examples of two sentence pairs. Target word 1 is italicized and Target word 2 is underlined.

The children were asked to read the sentence pairs both orally and silently. Therefore, two sets of sentences were created. Each set consisted of 20 sentence pairs. Set 1 contained ten congruent and ten incongruent sentence pairs. Of the ten congruent sentence pairs five sentence pairs had a frequent Target Word 1 and five had an infrequent Target Word 1. Of the ten incongruent sentences again five sentence pair had a frequent Target Word 1 and five had an infrequent Target Word 1. This pattern was the same for set 2. Target Word 1 and Target

1. The toddler falls down the stairs. His feet slide from the slippery steps. 2. The cat is glaring at the bird. With her claws she catches the mouse.

(14)

Word 2 of the two sets were matched on word frequency (Schrooten & Vermeer, 1994). Table 2 displays the average frequency of Target word 1 and Target word 2 in both sets. Because the sentence pairs had a very strict word order due to the positions of the target words, eight filler sentence pairs with a different word order were created to ensure that the children did not notice the strict word order.

Table 2

Average Frequency of the Target Words

Target Word 2

Congruent T2 Incongruent T2

Set 1 Set 2 Set 1 Set 2

Target Word 1 T1 T2 T1 T2 T1 T2 T1 T2

Frequent T1 271 33 268 39 254 32 251 31

Frequent T1 5 13 11 27 10 15 (37a) 10 33

Note. T1 = Target Word 1; T2 = Target Word 2

a

For two plural words (pears and elephants) used in the sentences the plural form of the word was not included in the frequency database. The number between brackets represents the average frequency using the frequency of the singular form of the word (e.g., pear).

Experiment procedure. Children were tested individually in an empty room. The experiment lasted for twenty minutes. The testing area could be closed off with curtains. Children were asked to sit down and the chair and forehead rest were adjusted to their height. Children were asked to place their hands on a green stickered and red stickered key on a keyboard in front of them. The experiment started with a calibration and verification. When successful, the task was explained both visually on screen and verbally by the test assistant. Children were instructed to look at a fixation cross and to start reading the sentence pair when it appeared. After reading carefully, the children were instructed to decide whether a sentence pair was congruent (pressing the green key) or incongruent (pressing the red key). After the explanation, there was a practice session of eight trials. After each practice trial the children

(15)

were informed whether they solved the verification correctly or incorrectly. The experiment was divided in four blocks of twelve sentence pairs, so that the children could move during the breaks. Before each block children were both visually and verbally instructed whether they had to read the sentence pairs orally or silently. Each new block started with a gaze -check. The first sentence of the next block was always a filler.

Apparatus. The sentence pairs were displayed one-by-one vertically and horizontally centred on a single line in a sans-serif font. Each sentence pair was preceded by a fixation cross (+) displayed for 25 milliseconds at the centre of the screen. Participants’ heads were stabilized with a forehead rest situated at a distance of 90 cm from the monitor. Eye movements of the left eye were recorded using the EyeLink1000 with a 35 mm lens, sampling at 1000 Hz with the illumination power set at 100%. A three-point calibration and verification was performed at the beginning of the task and a drift check at the beginning of a new set after each break. If the gaze position deviated more than 30 pixels compared to the previous calibration, gaze position was recalibrated.

Eye movement measures. Several eye movement measures were recorded during reading. Sentence Reading Speed (Sentence RS) was the total reading time from appearance of the sentence pair until the moment the child pressed either the green or red key to indicate the congruency of the sentence pair. Proportion correctly solved verifications was also recorded. Reading Speed of Target word 1 (T1 RS) and Reading Speed of Target word 2 (T2 RS) were the total duration of the fixations within the coordinates of the target words. The coordinates differed for each target word and sentence pair and had an error margin of ± 25 pixels. A fixation was registered when the participant looked at a particular coordinate with a margin of 25 pixels for more than 50 milliseconds. When a child made a fixation to the left of a previous fixation, this was registered as a regression. Because the fixation cross that preceded the sentences was displayed in the middle of the screen, sometimes children had

(16)

several ‘pre-regressions’ before they fixated on the first word of the sentence and started reading to the right. Therefore, the regressions preceding the first fixation within the coordinates of the first target word, or, if the child did not read the first target word, the number of regressions made before the child made two consecutive fixations to the right were disregarded.

Reading ability. Reading ability was measured with the Eén Minuut Taak [One

Minute Task], a standardized word reading fluency task (Brus & Voeten, 1979). The test consisted of 116 words of increasing difficulty presented in four columns of 29 words each. Participants were asked to correctly read aloud as many words as possible in one minute. The score was the number of items read correctly. The test has a test–retest reliability of .87 (Brus & Voeten, 1979).

Phonological awareness. A deletion task (de Jong & van der Leij, 2003) was used to

measure phonological awareness. Participants were first asked to repeat a Dutch pseudoword (sep) and were then asked to delete a designated sound from the word (sep without /s/). The words and deletions became increasingly difficult, starting with monosyllabic words, then bisyllabic words and finally two deletions of the same sound within one bisyllabic word. There were four items per level of difficulty. There were three monosyllabic practice items and two items to practice double deletions in bisyllabic words. Each correct answer was rewarded one point. The final score was the total amount of correct answers with a maximum of twelve. Previous studies have reported a test–retest reliability of .75 in third grade (van Bergen, Bishop, van Zuijen & de Jong, 2015).

Rapid naming. A digit and letter naming task (van den Bos & Lutje Spelberg, 2007)

were used to measure rapid naming. For the digit naming task, children received a sheet with 50 digits (2, 4, 5, 8 and 9) presented ten times each in five columns of ten digits. The children were asked to name the digits as quickly as possible. The letter naming task was similarly

(17)

presented, but with 50 letters (a, d, o, p and s) instead of numbers. There were no practice items. The score was the average number of correct responses per second of the digit and letter condition. Split-half reliability coefficients in third grade were .93 and .87 for digits and letters respectively (Evers et al., 2011).

Visual attention span. Visual attention span was measured with the Visual Attention

Span task (Valdois et al., 2003). In this computer task, participants were shown 20 five-letter strings (e.g., B P T F L) for 200 milliseconds each. The strings were made from 10 consonants (B, D, F, H, L, M, P, R, S and T). Each consonant was used in ten strings and was placed twice in each letter position. Participants were asked to name the letters after disappearance. The task had five practice trials. The final score was the number of correctly named letters in the right order, with a maximum of 100. Previous studies have found a test–retest reliability of .88 in third grade (van Bergen et al., 2015).

Vocabulary. Vocabulary was measured with the vocabulary subtest from the

WISC-III-NL (Wechsler et al., 2002). Participants were asked to verbally define words of increasing difficulty. The test assistant always presented the word in the “What is...” question format, for example: “What is a diamond?” When children’s answers were partially correct, the test assistant was allowed to ask whether the child could explain the answer. There were no practice items. Answers were scored 0, 1 or 2 points depending on their level of correctness according to the manual. The task was stopped if the participants obtained 0 points four times in a row. For the final score all given points were added up with a maximum of 70. The split-half reliability is .88 according to the manual for children aged 8.5 years (Wechsler et al., 2002).

(18)

Procedure

This study was approved by the institutional Ethics Committee (project number: 2016-CDE-7531). Passive consent was obtained from parents after receiving an information letter on paper from the teacher and digitally through email.

The tests were administered in two sessions in empty class rooms. During the first session, reading ability, rapid naming, phonological awareness, vocabulary and visual attention span were administered in that order in approximately 20 minutes. The session started with several background questions. During the second session, which again took about 20 minutes, the eye tracking experiment was administered.

Analyses

The differences between oral and silent reading were tested with 2x2 repeated measures ANOVAs. The within-subjects factors were reading mode (oral or silent), the frequency condition (frequent or infrequent Target Word 1 (T1)) or the congruency condition (congruent or incongruent Target Word 2 (T2)). Order effects were controlled for by adding the order condition as a between-subjects factor in all ANOVA’s. This factor did not have a main effect in any analysis, nor interacted with reading mode, the frequency condition or the congruency condition. Therefore, order was not included in any of the final analyses unless otherwise stated. The dependent variables were Sentence Reading Speed (Sentence RS), Reading Speed of Target Word 1 (T1 RS) and Target Word 2 (T2 RS), the proportion correctly solved verifications (proportion correct); the number of fixations (fixations); and the number of regressions (regressions). For all ANOVAs the first within-subjects factor was always reading mode. The second within-subject factor depended on the hypothesis being tested. For hypotheses regarding reading speed (hypothesis 1) frequency is the second within-subjects factor, because it is investigated whether decoding differences affect reading speed. For hypotheses concerning comprehension or eye movement patterns on the sentence level

(19)

(hypotheses 2 and 3) the congruency condition is the second within-subject factor, because the congruency of the sentence might influence reading behaviour of the whole sentence. For each ANOVA four variables per child were calculated: four average scores of the dependent variable (Sentence RS, T1 RS, T2 RS, proportion correct, fixations or regressions), one for each condition of the first factor (always reading mode) and the second factor.

To investigate individual differences in oral and silent reading, correlations were calculated between Sentence RS, T1 RS, T2 RS, proportion correct, fixations, regressions, reading ability, phonological awareness (PA), rapid naming (RAN), visual attention span (VAS) and vocabulary, separately for oral and silent reading. Furthermore, regression analyses were carried out with Sentence RS, T1 RS, T2 RS, proportion correct, fixations and regressions as outcome variables and four cognitive skills (PA, RAN, VAS and vocabulary) as predictors.

(20)

Results Data cleaning

One child correctly answered only 55% of the verification judgements during the sentence verification task. Because this percentage is not significantly above .50, i.e., chance level (z-test of proportion: p = 0.26), it cannot be assumed that this child performed the task as was intended. Therefore, it was decided to remove the child from the sample.

Then, data were checked for missing data. Out of the 8400 data points (35 children that had read 40 sentence pairs each during which 6 eye movement measures were registered) 360 data points (4.29%) were missing.

Secondly, the data were checked for within-subject outliers. For all five eye movement variables (Sentence RS, T1 RS, T2 RS, proportion correct, fixations and regressions) scores that deviated more than three standard deviations from the individual’s mean were recoded as missing values (67 data points; 0.80%). These deviating scores might have been caused by disruptions during the testing session and therefore it was decided to recode them as missing.

Thirdly, it was decided to remove eye movement measurements of sentence pairs that were incorrectly judged during the sentence verification task. This is common practice in the sentence verification paradigm (e.g., Gick, Craik & Morris, 1988; Goolkasian, 1996; Lauro, Reis, Cohen, Cecchetto & Papagno, 2010). In total 15.89% (192 sentences) of all the sentences read by the children (1400) were incorrectly judged and 883 data points (10.51%) were recoded as missing.

After cleaning, 1310 (16.17%) data points were missing. For Sentence RS, 206 (2.45%) data points were missing, for RS T1 282 (3.36%) data points, for RS T2 367 (4.37%), for fixations 227 (2,70%) data points, for regressions 228 (2.71%) and for proportion correct there were no missing data points.

(21)

Next, average scores of Sentence RS, T1 RS, T2 RS, proportion correct, fixations and regressions were calculated. The average scores were checked for between-subject outliers. Twelve scores that deviated more than three standard deviations from the group mean were recoded into a score that deviated exactly three standard deviations from the group mean. Moreover, one child deviated more than 3 standard deviations from the group mean on the reading ability test and this score was also recoded.

For the correlations and regression analyses averages of the outcome variables were calculated per child, separately for oral and silent reading. Three scores deviated more than three standard deviations from the group mean and were recoded to a score deviating exactly three standard deviations from the group mean.

Regarding normality of the data, the variables proportion correct, fixations and regressions were normally distributed according to Kline (2011). RS of Sentences, T1 and T2 was measured in reading time in milliseconds, but these scores were not normally distributed. Due to the extremeness of these values, RS of Sentences, T1 and T2 were transformed into words read per second, which normalized the data. The cognitive skills reading ability, phonological awareness (PA), rapid naming (RAN) visual attention span (VAS) and vocabulary were normally distributed.

For the regression analyses investigating the contribution of PA, RAN, VAS and vocabulary to eye-track measures or oral and silent reading, assumptions were checked. Residual plots showed no evidence against the assumption of homoscedasticity. Moreover, multicollinearity was not an issue (all VIF < 2).

Reading Speed

It was first investigated whether reading speed differed between reading orally and silently. This was analyzed with two repeated measures ANOVAs. The first within-subjects factor was reading mode (oral or silent). The second within-subject factor was the frequency

(22)

condition (frequent or infrequent T1), because it was investigated whether decoding affected reading speed and the frequency condition manipulated decoding. The dependent variables were Sentence RS and RS T1.

Sentence RS. Table 3 displays the descriptive statistics of Sentence RS. It seems that

oral reading speed is somewhat lower than silent reading speed. The difference between frequent and infrequent words is not large.

Table 3

Descriptive Statistics of Sentences RS measured in Items Read per Second

Oral Silent

M (SD) Range M (SD) Range

Frequent 1.88 (0.51) 0.79 – 3.45 1.91 (0.56) 0.51 – 3.44 Infrequent 1.74 (0.47) 0.55 – 2.56 1.94 (0.60) 0.48 – 3.11

The ANOVA revealed a significant main effect for reading mode, F(1, 34) = 5.86, p = .02, = .15. Sentences were read significantly faster silently than orally. There was no main effect of frequency (F(1, 34) = 3.30, p = .08). There was a significant interaction effect (Figure 3) between reading mode and frequency, F(1, 34) = 6.51, p = .02, = .16. The difference between oral and silent reading was larger for infrequent T1 than for frequent T1.

Figure 3. Interaction effect of reading mode and frequency on Sentence RS. 1,6 1,65 1,7 1,75 1,8 1,85 1,9 1,95 2 Oral Silent Wor d s re ad p er s ec on d Reading mode

Frequent target word Infrequent target word

(23)

T1 RS. Concerning RS T1, the descriptive statistics are displayed in Table 4. It seems

that there is no difference between oral or silent reading. Frequent words seem to be read faster than infrequent words. There was no significant main effect of reading mode (F(1, 34) = 0.30, p = .86). RS T1 did not differ between oral and silent reading. There was a significant main effect of frequency, F(1, 34) = 63.70, p < .001, = .65. Frequent target words were read faster than infrequent target words. The interaction effect between reading mode and frequency was not significant (F(1, 34) = 1.48, p = .23).

Table 4

Descriptive Statistics of T1 RS measured in Items Read per Second

Oral Silent

Frequent 1.28 (0.48) 0.38 – 2.54 1.25 (0.54) 0.30 – 3.16 Infrequent 0.87 (0.38) 0.23 – 2.05 0.93 (0.38) 0.21 – 2.28

Reading Comprehension

Next, it was investigated which reading mode yielded the best comprehension. This was analyzed with three repeated measures ANOVAs. The first within-subjects factor was reading mode (oral or silent). The second factor was the congruency condition (congruent or incongruent T2) as congruency might influence the comprehension process. The dependent variables were proportion correct, Sentence RS and T2 RS.

Proportion Correct. Table 5 depicts the descriptive statistics of proportion correct.

There seems to be a difference between congruent and incongruent sentence pairs, but the difference between oral and silent reading seems small. Moreover, average proportions correct for incongruent sentences were near 1.00, indicating a ceiling effect. Therefore, these results should be interpreted cautiously.

(24)

Table 5

Descriptive Statistics of Proportion Correctly Solved Verifications

Oral Silent

Congruent 0.79 (0.13) 0.50 – 1.00 0.76 (0.15) 0.50 – 1.00 Incongruent 0.93 (0.06) 0.80 – 1.00 0.96 (0.07) 0.71 – 1.00

For proportion correct there was indeed no main effect of reading mode (F(1, 31) = 0.02, p = .88). Congruency had a significant main effect, F(1, 31) = 72.91, p < .001, = .70. Incongruent sentence pairs were solved correctly more often than congruent sentence pairs. There was also a significant interaction effect (Figure 4) between reading mode and congruency, F(1, 31) = 4.65, p = .04, = .13. The difference between congruent and incongruent sentences was larger during silent reading than during oral reading.

When controlling for order effects for this hypothesis, a three-way interaction was found between reading mode, congruency and order, F(3, 31) = 3.55, p = .03, = .26. Interaction plots showed that three of the four groups had similar interaction patterns, but that the interaction effect for Group 2 deviated from the rest. However, none of the post-hoc tests found a significant difference between the groups (p-values ranging from .996 to 1.00).

Figure 4. Interaction effect of reading mode and congruency on proportion correctly solved verifications. 0,7 0,75 0,8 0,85 0,9 0,95 1 Oral Silent Proportion correctly resolved verifications Reading mode

Congruent sentence pair Incongruent sentence pair

(25)

Sentence RS. For Sentence RS, the descriptive statistics are displayed in Table 6. The

statistics indicate that silent reading might be faster than oral reading, but the difference between congruent and incongruent sentence pairs seems small. A significant main effect was found for reading mode, F(1, 34) = 5.94, p = .02, = .15. Sentences were read significantly faster silently than orally. There was no significant main effect of congruency (F(1, 34) = 2.66, p = .11). The interaction between reading mode and congruency on sentence reading speed approached significance (F(1, 34) = 0.46, p = .05). There seemed to be a trend that the difference between oral and silent reading was larger for incongruent sentence pairs than for congruent sentence pairs.

Table 6

Descriptive Statistics of Sentence RS measured in Items Read per Second

Oral Silent

T2 RS. Next, T2 RS was analyzed. The descriptive statistics (Table 7) indicate little

difference between the reading modes, but there seems to be an effect of congruency. No main effect was found of reading mode (F(1, 33) = 0.09, p = .77). There was a main effect of congruency, F(1, 33) = 18.00, p < .001, = .35: congruent T2 were read faster than incongruent T2. The interaction effect between reading mode and congruency was not significant (F(1, 33) = 0.26, p = .61).

(26)

Table 7

Descriptive Statistics of T2 RS measured in Items Read per Second

Oral Silent

Eye Movements

Next, it was investigated whether oral and silent reading were characterized by differential eye movements. Two repeated measures ANOVAs were performed. The first within-subjects factor was reading mode (oral or silent). The second factor was the congruency condition (congruent or incongruent T2), because this hypothesis is on the sentence level and the congruency condition influences reading at the sentence level. The dependent variables were fixations and regressions.

Fixations. In Table 8 the descriptive statistics of fixations are presented. Oral reading

seems to require more fixations, as well as congruent sentence pairs. A significant main effect was found for reading mode, F(1, 34) = 4.74, p = .04, = .12. More fixations were made when reading orally than silently. Moreover, a main effect of congruency was found, F(1, 34) = 15.71, p < .001, = .32. Congruent sentences were characterized by more fixations than incongruent sentences. No interaction effect between reading mode and congruency was found (F(1, 34) = 0.12, p = .73).

Table 8

Descriptive Statistics of Fixations

Oral Silent

(27)

Regressions. Concerning regressions, of which the descriptive statistics are displayed

in Table 9, more regressions were made when reading orally or congruent sentence pairs. A significant main effect was found, F(1, 34) = 4.31, p = .046, = .11. More regressions were made during oral reading than during silent reading. There was also a significant main effect of congruency, F(1, 34) = 7.38, p = .01, = .18. Congruent sentence pairs required more regressions than incongruent sentence pairs. The interaction effect between reading mode and congruency approached significance (F(1, 34) = 3.66, p = .06). A trend was visible in which the difference between oral and silent reading was larger for congruent sentence pairs than for incongruent sentence pairs.

Table 9

Descriptive Statistics of Regressions

Oral Silent

Individual Differences and Oral and Silent Reading

Lastly, it was investigated how individual differences in reading and underlying skills of reading influence online measures of oral and silent reading. Considering the small sample (N = 35), these results should be seen as exploratory.

First, separate correlations for oral and silent reading were calculated to investigate the associations between reading ability, PA, RAN, VAS and vocabulary and the six eye movement measures. Table 10 displays the descriptive statistics of the cognitive skills.

(28)

Table 10

Descriptive Statistics of the Raw Scores of the Cognitive Skills.

M (SD) Range Reading ability 56.69 (12.22) 24.00 – 79.00 PA 6.51 (2.39) 2.00 – 12.00 RAN 1.72 (0.30) 1.09 – 2.33 VAS 64.09 (13.32) 39.00 – 93.00 Vocabulary 25.51 (5.43) 11.00 – 38.00

Note. PA = Phonological awareness; RAN = Rapid naming; VAS = Visual attention span.

Table 11 display the correlations between the eye movement measures, as well as the associations between eye movement measures and reading ability (RA), PA, RAN, VAS and vocabulary, separately for oral and silent reading. The eye movement measures of oral and silent reading correlated highly, except for T2 RS. Moreover, reading ability (RA) correlated highly with the eye movement measures of reading Sentence RS and T1 RS, but not RS T2. This indicates that the variables Sentence RS and T1 RS measured the construct of reading.

PA is associated with T1 RS and proportion correct, but only for oral reading. RAN is related to Sentence RS of oral reading, but not silent reading. VAS is equally related to oral and silent Sentence RS and T1 RS. Moreover, it is significantly related to proportion correct during oral reading, but not silent reading. Also, larger VAS is associated with fewer fixations during silent reading, but this pattern is weaker for oral reading. Lastly, vocabulary is more strongly associated with proportion correct for silent than for oral reading.

(29)

Table 11

Correlation Table of Outcome and Predictor variables, Separated for Reading Mode

RA PA RAN VAS Vocabulary

ORM Oral Silent Oral Silent Oral Silent Oral Silent Oral Silent Sentence RS .86** .85** .72** .28 .22 .42* .29† .56** .59** -.03 -.07 T1 RS .54* .74** .52* .35* .29† .29† .06 .46* .51* -.06 .05 T2 RS .32† -.14 -.25 -.03 -.16 -.17 -.29† -.22 -.14 -.10 -.11 Prop. Correct .46* .14 .21 .36* .16 -.06 .22 .35* .22 .32† .38* Fixations .75** -.03 -.06 -.08 .00 .05 .06 -.29† -.39* .03 .03 Regressions .75** .05 .14 -.03 .17 .04 .11 -.01 .03 .10 .20 Note. ORM = Opposite Reading Mode; Prop. Correct = Proportion Correct; RA = reading

ability.

†

< .10, * < .05, ** < .001

Regression analyses were performed to investigate the contribution of cognitive skills (PA, RAN, VAS and vocabulary) to online measures of oral and silent reading. Table 12 displays these regressions. PA did not contribute significantly to any eye movement measure, but it came close to contributing to proportion correct during oral reading (p = .08). RAN did not contribute significantly to any eye movement measure either, but came close to contributing to oral Sentence RS (p = .07). VAS contributed significantly to Sentence RS and T1 RS. Contributions were slightly higher for silent reading. VAS also contributed negatively to fixations, which was significant for silent reading and close to significance for oral reading (p = .06). Vocabulary contributed significantly to Sentence RS. The contributions are negative, which is likely due to a suppression effect, as there are no high negative correlations between vocabulary and Sentence RS. Vocabulary came close to contributing to proportion correct during silent reading (p = .07).

(30)

Table 12

Contributions of Cognitive Skills to Outcome Variables

PA RAN VAS VOC

R2 β Β β β

Oral Silent Oral Silent Oral Silent Oral Silent Oral Silent Sentence RS .45* .44* .12 .07 .27† .12 .52* .62** -.27* -.30* T1 RS .24* .32* .24 .19 .14 -.13 .43* .54* -.27 -.14 T2 RS .06 .09 .06 -.06 -.11 -.26 -.19 -.01 -.03 -.04 Proportion Correct .29* .18 .29† .05 -.27 .13 .28 .06 .23 .33† Fixations .13 .22† -.03 .08 .16 .20 -.38† -.53* .13 .14 Regressions .06 .07 .05 .14 -.11 .08 -.19 -.10 .03 .19 Note. PA = phonological awareness; RAN = rapid naming; VAS = visual attention span.

†

(31)

Discussion

In the present study the differences in reading speed and comprehension between oral and silent reading in third graders was examined. Importantly, eye tracking methodology was used to investigate eye movement patterns during oral and silent reading to shed light on the different processes of oral and silent reading, which could help explaining the differences in reading outcomes between the two reading modes. In addition, this study aimed to extend research on individual differences in oral and silent reading by examining the relation of underlying skills of reading with various eye movement variables.

Reading Speed

Sentence reading speed was higher for silent reading than for oral reading. This is consistent with previous studies which found higher reading speed for words (van den Boer et al., 2014; Vorstius et al., 2014; Krieber et al., 2017) and texts (McCallum et al., 2004; van den Boer et al., 2014). Van den Boer (2014) did not find higher reading speed for sentences, but their sentences were shorter (three to seven words). In the present study the material consisted of two sentences of six to seven words, and thus the task had more resemblance with short text passages tasks than sentence reading tasks.

Interestingly, the effect on reading speed was affected by the frequency of one target word (T1) in the sentence: especially when sentences contained an infrequent word, sentence reading was faster for silent reading. According to Share’s (1995) self-teaching hypothesis, the more frequent a word is read, the stronger the orthographical representation of that word is in the mental lexicon. At a certain point, the representation is strong enough to allow the reader to use the lexical route (sight word reading) instead of the nonlexical route that requires decoding (dual-route cascaded model of Coltheart, Rastle, Perry, Langdon & Ziegler, 2001). The present study showed that for frequent words, that are most likely read through sight word reading, there is little speed difference between oral and silent reading, despite the

(32)

articulation process present in oral reading. For infrequent words, that require more decoding, oral reading is slowed down. This deceleration cannot solely be the result of the articulation process, because if it were, the decrease in reading speed for oral reading should also be apparent for frequent words. Thus, another process causes the delay in oral reading. A candidate might be the decoding process, as the delay is found to be stronger when more decoding is required (infrequent target words). It is possible that silent reading is faster because difficult words are not decoded, but simply skipped, as suggested by Prior et al. (2011). The finding that silent reading is characterized by fewer fixations also indicates that more words are skipped during silent reading. It could also be that the decoding process is slowed down during oral reading because cognitive capacity is reduced as a result of the dual (articulatory and oculomotor) task that is being performed (in line with Huestegge et al., 2014). Further research could investigate these suggestions.

Regarding reading speed of the target word itself, no difference was found between oral and silent reading. This might indicate spill-over effects: processing time of (infrequent) words are known to “spill over” onto the next fixation (Rayner & Duffy, 1986; Rayner, Sereno, Morris, Schmauder & Clifton, 1989). In the present study only reading speed of the target word was registered and not that of the subsequent word, so total processing duration of the target word (including spill-over effects) was not determined.

Reading Comprehension

No difference between oral and silent reading was found in proportion correctly solved verifications. This is inconsistent with previous literature (Kragler, 1995; de Jong & Share, 2007; Prior et al., 2011), but not entirely unexpected explainable. Oral reading is thought to facilitate reading comprehension in children because the reader is forced to read more thoroughly due to subsequent pronunciation (Prior et al., 2011). Previous studies (de Jong & Share, 2007; Prior et al., 2011) used materials in which children were required to answer

(33)

comprehension questions after reading a text. In the present study, however children had to decide immediately on the congruency of the sentences. It is therefore possible that the children read the sentences with the same thoroughness regardless of reading mode. Thorough reading probably does not mean that every word in the sentence is read, but that words important for comprehension are read. This is supported by the finding that there was no difference between reading speed between the reading modes on the target word that determined congruency. Moreover, silent reading was not characterized by more regressions, thus words were not read shallowly during first pass reading during silent reading, since superficially read words are often reread (Reichle et al., 2009). These results indicate that the sentence pairs were read equally intensive in both modes. To conclude, the immediate nature of the comprehension task in the present study might have caused the children to read exhaustively, regardless of reading mode, which in effect decreased comprehension differences between the reading modes.

In the present study it was also found that incongruent sentence pairs were solved correctly more often than congruent ones. Apparently the difficulty of the task also influences which reading mode is more efficient. Detecting congruent sentence pairs was more difficult, as indicated by lower proportion correct, than detecting incongruent sentence pairs. It seems that incongruent sentences were false more obviously and that congruent sentences were less obviously correct.

An interaction effect revealed that incongruent sentence pairs were correctly verified more often, especially when reading silently. As said before, congruent sentences pairs were found to be more difficult than incongruent pairs. Congruent sentences were therefore probably read more thoroughly, which is also indicated by the fact that congruent sentence pairs required more fixations and more regressions. This exhaustive reading process for congruent sentence pairs might have diminished comprehension differences between the two

(34)

modes. Incongruent sentences were read less thorough (less fixations and regressions), because probably ‘the catch’ was very obvious, as indicated by lower reading speed on the word that determined congruency (T2). The decision process for incongruent sentences was more efficient during silent reading, shown by the higher proportion correct and higher reading speed. This increased effectivity might have been caused by more cognitive capacity being available for the comprehension and decision process, because cognitive resources were not restricted by articulatory processes.

To conclude, this study indicates that oral reading does not always result in better comprehension in children. There seems to be no difference in comprehension between the reading modes when the comprehension task requires an immediate comprehension response. Task difficulty might have an effect on the preferred reading mode when using immediate response comprehension tasks. For difficult trials text appears to be read equally exhaustive during oral and silent reading, which may decrease comprehension differences between the reading modes. For easy trials oral reading appears to be less efficient, because the comprehension or decision process may be slowed down by additional articulatory processes. A difference in comprehension between oral and silent reading does seem to occur when a delayed response comprehension task is used, in which case oral reading has been found to be the preferred reading mode (Kragler, 1995; de Jong & Share, 2007; Prior et al., 2011).

Thoroughness of reading may be the factor that enhances oral reading comprehension in delayed comprehension tasks. This is indicated by the finding of the present study that silent reading comprehension can be enhanced when the reader is forced to read intensively (by using an immediate comprehension task). This ‘natural’ thoroughness of oral reading, also showed by an increased number of fixations during oral reading (also Vorstius et al., 2014; Krieber et al., 2017), might be caused by pronunciational demands of oral reading. Oral reading requires every word to be pronounced, which might cause more words to be fixated

(35)

upon and more processing time of each individual word: thus a natural thoroughness. Why it is only beneficial for children, but not for more experienced readers (Prior et al., 2011) is not clear yet, but following Prior et al. (2011) it could be possible that young children cannot yet reach this level of thoroughness when reading silently for themselves because they have not mastered reading and might be more tempted to skip difficult words. Once silent reading is mastered, there are probably less words that could cause reading difficulties. This might make it possible to achieve the same level of thoroughness. Prior et al. (2011) suggest that then oral reading comprehension declines, because articulation distracts the reader from the content of the text.

Eye Movements

Regarding fixations, oral reading was indeed characterized by a higher number of fixations, consistent with previous literature (Vorstius et al., 2014; Krieber et al., 2017). This can explain why oral reading is slower. What exactly causes the necessity for more fixations during oral reading is not clear yet, but it is probably a combination of articulation, decoding and less effective parafoveal processes. Since all words have to be pronounced, an exhaustive decoding process is probably necessary (especially with infrequent words) to compute phonetic codes for pronunciation. Because of this decoding process and subsequent articulation processes, parafoveal processing of the next few words may be less effective (Ashby et al., 2012), resulting in smaller saccadic amplitudes and more fixations.

Concerning regressions, a pattern contrary to previous studies was found. Oral reading was characterized by more regressions, especially in the congruent condition. This seems to contrast with previous literature that found that silent reading required more regressions (Vorstius et al., 2014; Krieber et al., 2017). It is thought that during silent reading more information is missed at first pass reading, which is solved by rereading that information with a regression (Reichle et al., 2009). However, it was discussed earlier that in the present study

(36)

the sentence pairs were probably read thoroughly regardless of reading mode, because after reading an immediate comprehension decision had to be made. The cognitive constraints imposed by the articulatory process in oral reading during first pass reading might then explain the need for more regressions, especially in the congruent condition, which seemed to be the more difficult condition.

Individual Differences

Lastly, individual differences in oral and silent reading were investigated. Phonological awareness (PA) did not contribute significantly to any eye movement measure in the present study. Low power, caused by a small sample size, seems to be an issue, as some regression coefficients were of similar size compared to coefficients in previous studies. Taking power issues into account, the overall trend in the present study seems similar to that of previous studies, namely that PA contributes equally to both reading modes. This fits with strong phonological theories of reading that suggest that phonological, orthographical and semantic features are all automatically activated upon reading a word (Frost, 1998; Perfetti & Hart, 2002).

In the present study rapid naming (RAN), like PA, did not contribute significantly to any eye movement measure. The regression coefficients, however, show somewhat similar trends comparing the present study to previous studies in that the contribution of RAN in oral reading was twice the size of the contribution in silent reading. Effect sizes were, however, much smaller in the present study. The similarity in trend could indicate that indeed the computation of a phonetic code and the pronunciation of this code are more important for oral reading.

Visual attention span (VAS) contributed slightly stronger to silent reading speed than oral reading speed, which is consistent with previous studies (Van den Boer et al., 2014). This might indicate that VAS, which might influence the size of the parafoveal region, is utilized

(37)

more effectively during silent reading (in line with Ashby et al., 2012). Also, VAS contributed negatively to number of fixations, especially for silent reading. Having a larger visual attention span may mean that the reader has a larger parafoveal region. Therefore, the reader has a larger preview benefit (Rayner, 1975) and can plan the next fixation further away from the current fixation, ultimately resulting in fewer fixations.

Limitations

The present study had several limitations. First, the sample size was small, which decreased the power to find effects. Some ANOVA effects were large enough to be immune to power issues (for = .18 to .70 the power was respectively 0.91 and 1.00 using G*Power (Faul, Erdfelder, Lang & Buchner, 2007)). Other ANOVA effects, however, had low power because of the small sample size (for = .11 to .16 the power was respectively 0.52 and 0.83). For the regression analyses power was also decreased. For small effects (R2 = .24) power was low (0.55) and even for large effects (R2 = .45) power was not optimal (0.85). These power issues might have caused small effects to be non-significant. For further research, therefore, it is recommended to have a larger sample, so that power is no longer an issue in interpreting the findings.

Another limitation was that several eye movement measures could not be recorded due to programming limitations which did not allow several eye tracking variables to be extracted from the eye movement data. One example discussed before is reading speed of words after the first target word, which could have helped to determine total processing duration of the target word, including spill-over effects. Other measures that could have been informative are duration of fixations and regressions, and mean saccadic amplitudes. It is recommended for future research to include such measures so the processes of oral and silent reading can be examined even more precisely. Furthermore, only third graders participated in the current study. Therefore no conclusions can be made about the development of oral and silent

(38)

reading. Additionally, a developmental design can also explore why the most effective reading mode for comprehension shifts from oral to silent reading around sixth grade (Prior et al., 2011). Longitudinal or cross-sectional designs are therefore recommended for future research.

Conclusions and Implications

In the present study the difference in reading speed and comprehension between oral and silent reading in third graders was examined meticulously using eye tracking methodology. It was aimed to seek explanations of differences in reading speed and comprehension by examining processes of oral and silent reading.

Silent reading was found to be the faster reading mode. Results from the present and from previous studies suggest this speed difference might be caused by skipping difficult words (Prior et al., 2011), a larger preview benefit due to more effective parafoveal processing (Ashby et al., 2012) and/ or more cognitive capacity being available (Huestegge et al., 2014) during silent reading.

The findings regarding reading comprehension were different from former studies. Overall, little difference was found between the two reading modes. It is suggested that the type of comprehension task influences the relative effectivity of the two reading modes. For delayed response comprehension tasks, oral reading is beneficial (Kragler, 1995; de Jong & Share, 2007; Prior et al., 2011), because it naturally forces the reader to read thoroughly. For immediate response comprehension tasks both reading modes seem equally effective, because the task forces the reader to read more thoroughly during silent reading.

The educational sector must become more aware of the effect of reading mode on reading comprehension. As of now, the reading mode in which a reading task is administered and the type of task are almost always chosen based on practical considerations. For example, when a child has to practice reading comprehension of long texts in class, teachers prefer to