Children with reading difficulties often demonstrate weaknesses in working memory (WM). This research study explored the relation between two WM systems (verbal and visuospatial WM) and reading ability in a sample of school-aged children with a wide range of reading skills. Children (N = 157), ages 9–12, were administered measures of short-term memory, verbal WM, visuospatial WM, and reading measures (e.g., reading fluency and comprehension). Although results indicated that verbal WM was a stronger predictor in reading fluency and comprehension, visuospatial WM also significantly predicted reading skills, but provided more unique variance in reading comprehension than reading fluency. These findings suggest that visuospatial WM may play a significant role in higher level reading processes, particularly in reading comprehension, than previously thought.

Working memory (WM) has been well researched as an essential component of executive functioning associated with reading ability (e.g., Swanson & O'Connor, 2009). It is strongly implicated in several domains of cognition, including attention, language, processing speed, writing, and mathematics (Gathercole, Alloway, Willis, & Adams, 2006). WM has been described as a finite cognitive resource system responsible for temporary storage and processing of information within immediate awareness (Baddeley, 1986, 2000; Baddeley & Logie, 1999). Previous research findings have been inconsistent as to whether correlations between WM and reading ability are primarily mediated by domain-specific processes or a domain-general system (e.g., Shah & Miyake, 1996). However, recent data from large-scale latent variable studies and meta-analytic studies examining WM interventions on academic skills support a domain-general system (Alloway, Gathercole, & Pickering, 2006; Melby-Lervag & Hulme, 2013; Titz & Karbach, 2014). Theoretically, if an individual is able to improve on domain-general WM capacity, these effects should transfer to a variety of academic skills that children are expected to learn in school, such as reading fluency and comprehension (Melby-Lervag & Hulme, 2013).

The framework used to capture WM performance is Baddeley's multicomponent model (Baddeley & Hitch, 1974; Baddeley, 2000). It is composed of a central executive system which controls the encoding and retrieval of stimuli input and monitors attention changes (Baddeley, 1986). The central executive is also responsible for controlling and manipulating information stored within two subsystems: the phonological loop, which stores verbal and linguistic input, and the visuospatial sketchpad, which stores visual and spatial input. The episodic buffer is a third subsystem which serves as a limited-capacity temporary storage and is capable of connecting information across domains to form integrated units of visual, spatial, and verbal input into long-term memory (Baddeley, 2000). Several research studies implicate the phonological loop as the main system affected in children with reading disabilities due to difficulties in storing and processing linguistic information (e.g., Swanson & Jerman, 2007). This plays a significant role in children's verbal WM, a subset of WM which is often involved in recalling letter–sound relationships, vocabulary, and text meaning during oral or silent reading.

Because of the emphasis placed on verbal WM in the literature, there has been limited research on the role of visuospatial WM as it pertains to higher level reading skills, particularly reading fluency and comprehension. Visuospatial WM is a subset of WM involved in recalling perceptual, pictorial, and spatial relationships of stimuli. Beginning readers learn to recognize the orientation and shape of letters of the alphabet, as well as associate letters with the corresponding symbol when decoding unfamiliar words (Badian, 2005). As readers become more fluent, they learn to read words automatically through orthographic processing (O'Brien, Wolf, Miller, Lovett, & Morris, 2011), allowing more efficient visual encoding of regular and irregular patterns of words into long-term storage. Advanced readers may learn to visualize details from texts as well as visually scanning the text for factual information in reading comprehension (Goff, Pratt, & Ong, 2005). Both verbal and visuospatial WM may be implicated in any of these developmental reading processes (Reiter, Tucha, & Lange, 2005); however, the strength of the predictive relationship remains unclear.

Verbal Working Memory

Although weaknesses in phonological processing are well documented in children with reading difficulties (e.g., Lyon, Shaywitz, & Shaywitz, 2003), poor verbal WM has also been implicated in problems with word recognition and reading comprehension (Jacobson et al., 2011; Swanson & Berninger, 1995; Swanson & Jerman, 2007). Verbal WM consists of complex cognitive operations in which maintenance, retrieval, manipulation, and transformation of verbal input takes place. Generally, lower verbal WM capacity makes it difficult for readers to perform phonological processes, such as blending and segmenting words, which are necessary for reading efficiently. Higher verbal WM capacity allows engagement of cognitive resources such as the generation of semantic associations, decoding, memory retrieval, and maintenance of salient information to facilitate reading comprehension (Sesma, Mahone, Levine, Eason, & Cutting, 2009).

Another function of verbal WM is to extract meaningful representations from verbal input taken in by the phonological loop (Dehn, 2010). Historically, some researchers considered the phonological loop as part of short-term memory (STM) because it involves two essential components: a phonological store and a rehearsal process (see Baddeley, 1986 for review). Currently, it has been acknowledged that STM and WM are both distinct operations (e.g., Swanson & Howell, 2003). WM tasks require higher attentional control and maintenance of memory traces in the face of interference by drawing resources from the central executive system (Baddeley, 2000). In other words, STM is often used in situations where small amounts of input are held passively and then reproduced in a sequential or untransformed fashion (e.g., recalling digits forward), while verbal WM requires additional manipulation or transformation of verbal input, and to reproduce the information in a different output (e.g., recalling digits backward).

Cain, Oakhill, and Bryant (2004) found that verbal WM (sentence span and digit span) accounted for 11.4% of the variance in reading comprehension after controlling for age, IQ, vocabulary, and word recognition. However, the study did not examine the influence of sentence span and digit span WM tasks separately in regression analyses. A similar study by Seigneuric, Ehrlich, Oakhill, and Yuill (2000) found that verbal WM (digits, sentences, and words) accounted for a significant amount of variance (5–10%) in reading comprehension after vocabulary and reading fluency were controlled, whereas spatial WM did not correlate with reading outcomes. Slow word recognition and reading fluency increase cognitive demands placed on other processes, particularly verbal WM, which in turn poses difficulties for beginning readers to attain fluency and comprehension and thus creates a processing bottleneck (Cutting, Materek, Cole, Levine, & Mahone; 2009; Wolf & Katzir-Cohen, 2001). Several studies suggest that depending on the type of academic skill, such as reading, writing, or mathematics, greater problems in performance are more likely to occur on verbal WM than visuospatial WM tasks (e.g., Gathercole, Brown, & Pickering, 2003). However, there is limited research that explores visuospatial WM in conjunction with verbal WM when assessing reading fluency and comprehension in children.

Visuospatial Working Memory

The visual–spatial sketchpad is primarily used as storage not only for visual and/or spatial input but also for linguistic input that can be recoded into nonverbal or visual forms. It allows the visual input to be processed in visuospatial WM, such as reversing the sequence of objects or manipulating images (Dehn, 2008). Visuospatial WM is primarily involved in the generation, manipulation, and maintenance of visual information (Gathercole & Baddeley, 1993). However, the literature exploring the role of visuospatial WM in reading ability overall yielded mixed results.

Gathercole, Brown, and Pickering (2003) found that visuospatial WM was associated with academic attainment on a national curriculum for children (ages 6–7). Thus, there appears to be a strong relationship between visuospatial WM and general academic performance in early elementary school grades. Some studies of visuospatial WM performance in children with reading disabilities revealed that their visuospatial WM was relatively intact compared with those of their typically developing same aged peers (Swanson & Jerman, 2007). Swanson and Howell (2001) found that reading comprehension was significantly correlated with both verbal WM measures (sentence span and digit span) and visuospatial WM (sequence of dots and sequence of directions) measures in children ages 9–14 years. The authors found that verbal WM tasks, and a WM latent variable that included overlapping variance between verbal and visuospatial WM explained significant portions of variance in reading comprehension. However, visuospatial WM alone did not consistently contribute to reading comprehension overall. They concluded that when the shared variance between the verbal and visuospatial WM tasks was not taken into consideration, the verbal WM tasks better predicted reading than did the visuospatial tasks which suggest that the influences of WM on reading comprehension may be both domain-general and domain-specific.

However, other studies have noted that visuospatial WM, but not verbal WM, is significantly related to reading comprehension. Bayliss, Jarrold, Gunn, and Baddeley (2003) conducted a study investigating factors underlying memory performance in children and adults. Results from their study revealed that a measure of visuospatial WM (forward Corsi block task), and not verbal WM (digit span), was significantly related to reading comprehension test scores. The researchers were able to replicate similar results and found relatively high correlations between visuospatial WM and reading comprehension (r = .68) (Bayliss, Jarrold, Baddeley, & Gunn, 2005). They also found relatively weaker correlations between verbal WM and reading comprehension tasks (r = .42–.48). Goff and colleagues (2005) used a computer-based Corsi block task (e.g., Milner, 1971) which requires participants to press blocks that light up on the screen in the same order. Their findings also revealed that the visuospatial WM performance made small independent contributions to reading comprehension but the contribution of verbal WM was not significant. Swanson (2000, 2010) speculated that any particular advantage that visuospatial WM gives to children with reading disabilities compared with typically achieving children fluctuates with processing demands placed on various components of WM. Based on the literature, there continues to be inconsistencies on the exact nature of the association between different modalities of WM with specific reading skills.

Purpose

Many studies have found that children with reading disabilities show deficits on WM in both verbal and visuospatial WM domains (e.g., Reiter et al., 2005), indicating a domain-general system, whereas other studies revealed that there were no significant impairments in visuospatial WM compared with control groups (e.g., Jeffries & Everatt, 2004). Depending on whether a WM model is domain-specific (e.g., Baddeley, 2000) or domain-general (e.g., Alloway et al., 2006), the role of the phonological system or the central executive system have been implicated for improving reading outcomes. Additionally, it has been shown that differences in task structure or the nature of the stimuli used can dramatically influence the predictive power of a WM measure, independently of task modality (Bayliss et al., 2003). This topic is particularly important as many intervention studies are investigating how WM training influences academic skills, specifically whether the transfer of cognitive training to academic abilities vary as a function of domain and training task (e.g., verbal vs. visuospatial training tasks) (Titz & Karbach, 2014).

One of the primary goals of this study was to identify the strongest predictors of reading fluency and reading comprehension in a sample of school-aged children with varying reading ability. Previous research studies investigating WM in children have often focused on word reading accuracy or reading comprehension as outcome measures. Few studies have used reading fluency, a skill that requires the quick recall and processing of text, and which is often used as an outcome measure used to monitor reading proficiency and progress in school-based assessment and universal screening. We hypothesized that verbal WM tasks significantly contribute more to measures of specific reading abilities (i.e., fluency and comprehension) than visuospatial WM tasks after controlling for general intellectual ability, STM, and behavioral inattention.

Methods

Participants

Participants were recruited from elementary schools and through community centers in the Midwest and Southeastern United States. Demographic data were collected regarding their gender, age, ethnicity, grade level enrolled, and whether they are currently receiving special education services at their current school. This sample included 157 children in Grades 4 and 5. Seventy-one were female (45.2%) and 86 were male (57.8%). The participants' ages ranged from 9 to 12 (mean age = 10.9). Seventy-two children (45.8%) were fourth grade students, and 85 children (54.2%) were fifth grade students. All children were native speakers of English. Racial/ethnicity information of the participants as reported by parents is as follows: 60% Caucasian, 22.7% Latino/Hispanic, 11.5% African American, 1.9% Asian, and 3.8% multiracial. Participants did not have any visual-hearing impairment or any learning, physical, behavioral, or psychiatric disorder based on parent report.

Parent socio-economic status (SES) was also determined based on reported parent education level and occupation (Hollingshead, 1975). Families generally came from middle class backgrounds (mean Hollingshead index score = 38.7, SD = 12.9). However, there was wide variability in scores ranging from 10 (low SES) to 66 (high SES).

Measures

Intellectual ability

WISC-IV Short Form

A brief standardized measure of intellectual ability was administered to each child participant. The two-subtest combination of the Wechsler Intelligence Scale for Children,Fourth Edition (WISC-IV; Wechsler, 2003,) short form (Sattler, 2008) is considered an acceptable strategy in clinical and research studies where obtaining a global FSIQ score is not the primary purpose of the assessment (Sattler & Dumont, 2004). The two subtests, Vocabulary and Matrix Reasoning, are reported to have good internal consistency (Vocabulary = .86; Matrix Reasoning = .85). The scaled scores from these two subtests were summed to yield a composite score, which was then computed into an estimated IQ score. The reliability and validity coefficients of the short form are .93 and .87, respectively. The mean estimated IQ from this sample was 102.1 (SD = 12.8).

Reading measures

WJ-III, Tests of Achievement, Third Edition—Letter Word Identification and Reading Fluency

Participants' individual word reading was screened using the Letter-Word Identification subtest of the Woodcock-Johnson, Tests of Achievement, Third Edition (WJ-III; Woodcock, McGrew, & Mather, 2001). Child participants were asked to read lists of words of increasing difficulty. They were also asked to complete the Reading Fluency subtest which required the participants to silently read statements and determine whether they are true or false under timed conditions. Internal consistency is high for both measures (.94 and .90, respectively). The standard scores obtained from the WJ-III Reading Fluency were subsequently converted into scaled scores in order to compute the Total Reading Fluency score.

GORT-IV Reading Fluency

Oral reading fluency was assessed using the Gray Oral Reading Test, Fourth Edition (GORT-4; Wiederholt & Bryant, 2001). Each child participant was asked to read aloud short passages of increasing difficulty. Multiple scaled scores were derived including reading speed (Reading Rate) and accuracy (Reading Accuracy) which was used to compute a fluency score (Reading Fluency). Internal consistency is high for this measure (.90). The Reading Fluency subtests scaled scores from both GORT-IV and WJ-III were averaged to create a new variable called Total Reading Fluency.

GORT-IV Reading Comprehension

Reading comprehension was also assessed using the GORT-4(Wiederholt & Bryant, 2001). After the child participants read the passages aloud, they were asked five multiple-choice questions about the passage. A scaled score was provided for Reading Comprehension. Internal consistency is also high for this measure (.90).

Short term memory

WISC-IV Digit Span Forward, WISC-IV-Integrated Spatial Span Forward

STM was assessed by the Digit Span Forward from the WISC-IV (Wechsler, 2003), and the Spatial Span Forward from the WISC-IV-Integrated (Kaplan, Fein, Kramer, Delis, & Morris, 2004). The child was asked to repeat a series of numbers verbatim as presented orally by the examiner. Digit Span Forward is part of the Digit Span subtest which is commonly used as a component measure of overall WM in the WISC-IV. Digit Span Forward primarily assesses auditory attention span and auditory STM, whereas Digit Span Backward assesses auditory or verbal WM since it requires additional cognitive resources to reverse numbers held in verbal STM. Spatial Span Forward requires the participant to recall or repeat the exact input, except it is a visuospatial task that involves tapping blocks one at a time sequentially as presented by the examiner. Internal consistency of Digit Span Forward is good (.83) and moderate for Spatial Span Forward (.78). Scores from both tasks were then averaged to form an STM variable.

Verbal working memory

WISC-IV Digit Span Backward, WRAML-2 Verbal Working Memory

The Digit Span Backward from the WISC-IV (Wechsler, 2003), and the Verbal WM subtest from the Wide Range Assessment of Memory and Learning, Second Edition (WRAML-2; Sheslow & Adams, 2003) were used to assess verbal WM. Digit Span Backward requires the child to repeat a series of numbers, but in the reverse order. Internal consistency is good (.80). The Verbal WM subtest required the child to recall a series of animals and non-animals, and then state the animals first in order of smallest to largest size, followed by the non-animal words. Internal consistency is relatively good (.81). Scores from both WM tasks were then averaged to form a Total Verbal WM variable.

Visuospatial working memory

WISC-IV-Integrated Spatial Span Backward, WRAML-2 Symbolic Working Memory

Spatial Span Backward from the WISC-IV-Integrated (Kaplan et al., 2004) and Symbolic WM from the WRAML-2 (Sheslow & Adams, 2003) were used to assess visuospatial WM. The Spatial Span Backward task is similar to Spatial Span Forward, except the child is asked to tap the blocks sequentially in reverse order. Internal consistency for this subtest is good (.81). The Symbolic WM subtest required the child to recall a random series of numbers and/or letters presented orally, which must then be reproduced by pointing to the location of the numbers and letters on a card, with the numbers in ascending order first, followed by the letters. Internal consistency for this subtest is also good (.80). Scores from both WM tasks were then averaged to form a Total Visuospatial WM variable.

Behavioral inattention

SNAP-IV Inattention scale

The Swanson, Nolan and Pelham, Version Four (SNAP-IV) rating scale was used to assess behaviors related to inattention (Swanson et al., 2001). Items from this scale provided similar wording to the criteria listed within the DSM-IV-TR (APA, 2000) for ADHD-Predominantly Inattentive Type. When completing the scale, parents and teachers rated the frequency they observed each behavior during the past 6 months. Frequency was determined using a Likert scale ranging from 0 to 3 points: “not at all,” “just a little,” “quite a bit,” or “very much.” Parent and teacher ratings were then summed to create an additional variable: Total Inattention. Internal consistency for parent and teacher ratings were high (.94 or above). The inter-rater reliability correlation between parent and teacher ratings was .49. Parents and teachers completed all items, and any missing data were reviewed and followed up with the raters to ensure accurate completion.

The SNAP-IV scale does not use normative cutoff points based on age or gender. However, Bussing and colleagues (2008) analyzed the SNAP-IV scores of boy and girls (ages 8–11) and reported findings that did not support the use of age- or gender-specific norms among elementary school-age students on this measure. Estimates revealed a small effect size for parent and teacher ratings (Cohen's d = .20–.33) comparing 8–10 year olds to 11 year olds. Additionally, gender differences revealed small to medium effect sizes based on parent and teacher ratings (Cohen's d = .29–.49).

Procedures

Following approval by the university Institutional Review Board, all child participants and their parents signed consent forms indicating their agreement to participate in the study. Each child was tested in a quiet area at the child's school. The individual testing took ∼60 min to complete, across two sessions. Half the participants were given the STM tasks first and half received the WM span tasks first. Within each span task type, half of the children received the verbal or the visuospatial tasks first. Trained graduate assistants administered and scored portions of the assessments. Parents and teachers were also asked to complete several forms including demographic information and rating scales regarding the child's behaviors at home and school.

Data Analyses

Hierarchical multiple regression analyses were used to examine the relative contribution of verbal and visuospatial WM measures on reading outcomes while controlling for general intellectual ability, word reading, STM, and ratings of inattention. Separate regression analyses were initially conducted for each of the four specific WM tasks. To address multicollinearity, composite scores of the WM measures (e.g., Total Verbal WM, Total Visuospatial WM) were later used in subsequent regression analyses.

Results

Mean scores and standard deviations of intellectual ability, STM and WM measures, ratings of inattention, and reading measures are presented in Table 1. Closer inspection of the means, standard deviations, and actual range of scores revealed that the scores obtained from most of the measures were normally distributed. The participants' estimated IQ scores were within the normal range. One exception was the SNAP-IV parent and teacher ratings of inattention, which showed positive skewness as a majority of the children were rated to have few attention problems (scores of zeroes). Logarithmic transformation was conducted to make the scores more closely approximate to the normal distribution. Because the distribution of the raw scores contains zeroes, a constant of one was added to each raw score in order to avoid taking the log of zero. The variable transformation eliminated the positive skewness, and therefore allowed the SNAP-IV data to be used in subsequent analyses.

Table 1.

Means and standard deviations of study variables

Variable M SD Range 
Age 10.9 1.2 9.0–12.0 
Parent SES (Hollingshead) 38.7 12.9 10.0–66.0 
Reading 
 WJ-III Letter-Word IDa 101.7 11.1 88.0–130.0 
 WJ-III Reading Fluencya 103.4 13.4 84.0–140.0 
 GORT-IV Fluencyb 10.2 2.4 5.0–18.0 
 GORT-IV Comprehensionb 11.2 2.5 6.0–18.0 
Intellectual ability 
 WISC-IV Vocabularyb 9.4 2.3 4.0–17.0 
 WISC-IV Matrix Reasoningb 10.8 2.0 5.0–17.0 
 WISC-IV Estimated IQa 102.1 12.8 78.0–140.0 
STM (average scaled) 10.5 2.4  
 WISC-IV DSFb 10.9 2.6 6.0–15.0 
 WISC-IV-I SSFb 10.1 2.3 5.0–16.0 
Total Verbal WM (average scaled) 10.1 2.2  
 WISC-IV DSBb 9.9 2.4 4.0–16.0 
 WRAML-2 Verbal WMb 10.3 2.1 6.0–17.0 
Total Visuospatial WM (average scaled) 10.1 2.3  
 WISC-IV-I SSBb 10.4 2.2 5.0–16.0 
 WRAML-2 Symbolic WMb 9.8 2.2 4.0–15.0 
Behavioral inattention 
 SNAP-IV Parent Inattentionc 0.69 0.61 0–3.0 
 SNAP-IV Teacher Inattentionc 0.62 0.80 0–3.0 
Variable M SD Range 
Age 10.9 1.2 9.0–12.0 
Parent SES (Hollingshead) 38.7 12.9 10.0–66.0 
Reading 
 WJ-III Letter-Word IDa 101.7 11.1 88.0–130.0 
 WJ-III Reading Fluencya 103.4 13.4 84.0–140.0 
 GORT-IV Fluencyb 10.2 2.4 5.0–18.0 
 GORT-IV Comprehensionb 11.2 2.5 6.0–18.0 
Intellectual ability 
 WISC-IV Vocabularyb 9.4 2.3 4.0–17.0 
 WISC-IV Matrix Reasoningb 10.8 2.0 5.0–17.0 
 WISC-IV Estimated IQa 102.1 12.8 78.0–140.0 
STM (average scaled) 10.5 2.4  
 WISC-IV DSFb 10.9 2.6 6.0–15.0 
 WISC-IV-I SSFb 10.1 2.3 5.0–16.0 
Total Verbal WM (average scaled) 10.1 2.2  
 WISC-IV DSBb 9.9 2.4 4.0–16.0 
 WRAML-2 Verbal WMb 10.3 2.1 6.0–17.0 
Total Visuospatial WM (average scaled) 10.1 2.3  
 WISC-IV-I SSBb 10.4 2.2 5.0–16.0 
 WRAML-2 Symbolic WMb 9.8 2.2 4.0–15.0 
Behavioral inattention 
 SNAP-IV Parent Inattentionc 0.69 0.61 0–3.0 
 SNAP-IV Teacher Inattentionc 0.62 0.80 0–3.0 

Notes: GORT-IV = Gray Oral Reading Test-Fourth Edition; WISC-IV = Wechsler Intelligence Scale for Children-Fourth Edition; WISC-IV-I = Wechsler Intelligence Scale for Children-Fourth Edition-Integrated; STM = short-term memory; WM = working memory; DSF = Digit Span Forward; SSF = Spatial Span Forward; DSB = Digit Span Backward; SSB = Spatial Span Backward; WRAML-2 = Wide Range Assessment of Memory and Learning-Second Edition.

aScores from these measures are based on a standard score, mean = 100, SD = +15.

bScores from these measures are based on a scaled score, mean = 10, SD = +3.

cScores from these measures are based on a raw score: 0 (not at all) to 3 (very much).

There were significant bivariate correlations between reading WM measures, ratings of inattention, and outcome variables and (Table 2). A significance level of .01 was adopted on all WM measures that were significantly correlated with reading variables due to the relatively high number of intercorrelations found. Verbal WM measures (WISC-IV DSB, WRAML-2 Verbal WM) had stronger correlations (r = .37–.47, p < .001) than STM (r = .18–.20, p < .001) and visuospatial WM (r = .20–.22, p = .012) with reading fluency measures. Similar findings also were found where verbal WM measures (r = .42–.48, p = .002) demonstrated stronger correlations than STM (r = .17–.21, p = .003), and visuospatial WM (r = .21–.38, p = .009) with reading comprehension.

Table 2.

Correlation matrix of study variables

 10 11 12 
1. GORT-IV Fluency .55** .45** .66** .51** .18* .20* .41** .37** .22* .20* −.25** 
2. GORT-IV Comprehension  .39** .59** .52** .17* .21* .48** .42** .21* .38** −.27** 
3. WJ-III Letter Word ID   .55** .47** .21* .39** .42** .33** .14 .12 −.21* 
4. WJ-III Reading Fluency    .41** .37** .53** .47** .40** .31** .21* −.24** 
5. WISC-IV Estimated IQ     .46** .47** .30** .31** .39** .38** −.24** 
6. WISC-IV DSF      .42** .23** .30** .31** .29** −.14 
7. WISC-IV-I SSF       .28** .29** .31** .39** −.16 
8. WISC-IV DSB        .41** .24** .32** −.21* 
9. WRAML-2 Verbal WM         .28** .40** −.22* 
10. WISC-IV-I SSB        .45** −.19 
11. WRAML-2 Symbolic WM           −.23** 
12. Total Inattention            
 10 11 12 
1. GORT-IV Fluency .55** .45** .66** .51** .18* .20* .41** .37** .22* .20* −.25** 
2. GORT-IV Comprehension  .39** .59** .52** .17* .21* .48** .42** .21* .38** −.27** 
3. WJ-III Letter Word ID   .55** .47** .21* .39** .42** .33** .14 .12 −.21* 
4. WJ-III Reading Fluency    .41** .37** .53** .47** .40** .31** .21* −.24** 
5. WISC-IV Estimated IQ     .46** .47** .30** .31** .39** .38** −.24** 
6. WISC-IV DSF      .42** .23** .30** .31** .29** −.14 
7. WISC-IV-I SSF       .28** .29** .31** .39** −.16 
8. WISC-IV DSB        .41** .24** .32** −.21* 
9. WRAML-2 Verbal WM         .28** .40** −.22* 
10. WISC-IV-I SSB        .45** −.19 
11. WRAML-2 Symbolic WM           −.23** 
12. Total Inattention            

Notes: GORT-IV = Gray Oral Reading Test-Fourth Edition; WJ-III = Woodcock-Johnson Tests of Achievement-Third Edition; WISC-IV = Wechsler Intelligence Scale for Children-Fourth Edition; WISC-IV-I = Wechsler Intelligence Scale for Children-Fourth Edition-Integrated; WM = working memory; DSF = Digit Span Forward; SSF = Spatial Span Forward; DSB = Digit Span Backward; SSB = Spatial Span Backward; WRAML-2 = Wide Range Assessment of Memory and Learning-Second Edition.

*p < .01 and **p < .001

Verbal WM, Visuospatial WM, and Reading

To examine the influence of verbal and visuospatial WM measures on reading outcomes in a continuous manner, hierarchical regression analyses were conducted. Because a relatively large set of predictor variables could introduce high levels of multicollinearity into regression equations (i.e., high variance inflation factors, and low tolerances), STM and WM variables were combined into their corresponding composite variables (e.g., STM, Total Visuospatial WM, and Total Verbal WM) based on principal components analysis (PCA). Visual inspection of the correlation matrix of the study variables indicated several correlations>.30. The Kaiser–Meyer–Olkin value at .72 exceeded the recommended value of .60, and the Bartlett's Test of Sphericity was statistically significant (.032), supporting the factorability of the correlation matrix. The PCA revealed the presence of three components with eigenvalues exceeding 1, explaining 13.89% (STM), 27.08% (Total Verbal WM), and 22.93% (Total Visuospatial WM) of the variance.

No significant differences were found between the two subtests scores within each composite: STM (p = .29), Total Verbal WM (p = .11), and Total Visuospatial WM (p = .32) composites. A correlation matrix of the STM and WM composite scores and Total Reading Fluency is presented in Table 3. Combining subtest scores into composite scores overall lowered collinearity, rendering acceptable variance inflation factors, and tolerance diagnostics for Total Verbal WM (VIF = 1.37, Tolerance = .73) and Total Visuospatial WM (VIF = 1.25, Tolerance = .80) when reading fluency was the dependent variable. Low collinearity was also found for Total Verbal WM (VIF = 1.63, Tolerance = .61) and Total Visuospatial WM (VIF = 1.72, Tolerance = .59) when reading comprehension was the dependent variable.

Table 3.

Correlation matrix of study composite scores and total reading fluency

 
1. Total Reading Fluency .18* .27** .20* 
2. STM  .25** .31** 
3. Total Verbal WM   .33** 
4. Total Visuospatial WM    
 
1. Total Reading Fluency .18* .27** .20* 
2. STM  .25** .31** 
3. Total Verbal WM   .33** 
4. Total Visuospatial WM    

Notes: STM = short-term memory; WM = working memory.

*p < .05 and **p < .01.

The Total Verbal WM and Total Visuospatial WM composites were entered in the regression analyses to determine the variance explained in both reading fluency and reading comprehension, while controlling for estimated IQ, word reading, inattention, and STM, which were entered as covariates in Step 1. The Total Verbal WM composite was entered in Step 2, and Total Visuospatial WM composite was entered in Step 3 in order to determine whether visuospatial WM accounted for additional variance if entered into the regression analysis after verbal WM. When Total Reading Fluency was used as the dependent variable, the total amount of variance explained by the model was 43%. Total Verbal WM provided 5% of the variance in the model, F(2, 151) = 17.42, p = .04, but Total Visuospatial WM did not provide additional variance to reading fluency, F(2, 151) = 10.01, p = .53 (Table 4). These findings were not as consistent when reading comprehension was used as the dependent variable. When Total Visuospatial WM was entered after Total Verbal WM, 50% of the variance in reading comprehension was accounted for (Table 5). Total Verbal WM significantly contributed 5% of the variance in the model, F(2, 151) = 15.30, p = .043; however, the inclusion of Total Visuospatial WM also provided significant results, contributing an additional 4% of unique variance to reading comprehension, F(2, 151) = 13.49, p = .036.

Table 4.

Hierarchical regression analyses of verbal WM and visuospatial WM predicting reading fluency

 b SEb β Model R2
 
ΔR2 p-value 
Model 1 
 Step 1: Estimated IQ .32 .44 .28    .034 
  Total inattention .10 .12 .17    .272 
  WJ-III Letter Word ID .48 .42 .41    .042 
  STM .31 .45 .50    .239 
  Reading comprehension .30 .53 .37 .37   .032 
 Step 2: Total Verbal WM .10 .38 .15 .42 .05 .040 
 Step 3: Total Visuospatial WM .13 .13 .11 .43 .01 .532 
        
 b SEb β Model R2
 
ΔR2 p-value 
Model 1 
 Step 1: Estimated IQ .32 .44 .28    .034 
  Total inattention .10 .12 .17    .272 
  WJ-III Letter Word ID .48 .42 .41    .042 
  STM .31 .45 .50    .239 
  Reading comprehension .30 .53 .37 .37   .032 
 Step 2: Total Verbal WM .10 .38 .15 .42 .05 .040 
 Step 3: Total Visuospatial WM .13 .13 .11 .43 .01 .532 
        

Notes: WJ-III = Woodcock-Johnson Tests of Achievement-Third Edition; STM = short-term memory; WM = working memory.

Table 5.

Hierarchical regression analyses of verbal WM and visuospatial WM predicting reading comprehension

 b SEb β Model R2 ΔR2 p-value 
Model 2 
 Step 1: Estimated IQ .43 .37 .18   .028 
  Total Inattention .20 .09 .10   .328 
  WJ-III Letter Word ID .26 .38 .23   .015 
  STM .28 .08 .34   .333 
  Total Reading Fluency .27 .05 .32 .40  .014 
 Step 2: Total Verbal WM .17 .25 .05 .45 .05 .043 
 Step 3: Total Visuospatial WM .09 .11 .01 .50 .04 .036 
 b SEb β Model R2 ΔR2 p-value 
Model 2 
 Step 1: Estimated IQ .43 .37 .18   .028 
  Total Inattention .20 .09 .10   .328 
  WJ-III Letter Word ID .26 .38 .23   .015 
  STM .28 .08 .34   .333 
  Total Reading Fluency .27 .05 .32 .40  .014 
 Step 2: Total Verbal WM .17 .25 .05 .45 .05 .043 
 Step 3: Total Visuospatial WM .09 .11 .01 .50 .04 .036 

Notes: WJ-III = Woodcock-Johnson Tests of Achievement-Third Edition; STM = short-term memory; WM = working memory.

Discussion

In the current study, we sought to understand the role of two different WM domains and their relation to reading outcomes among typically developing children. Children learn to read by using both linguistic and orthographic processing, which may require both verbal and visuospatial WM when processing text. However, it is unclear how strongly verbal and visuospatial WM predict reading outcomes, as children and adolescents may demonstrate reading difficulties if either or both of these WM domains are affected. This study expands on the currently literature by considering the role of WM on specific reading skills, particularly fluency and comprehension.

In this sample of children with a wide range of reading ability, all WM tasks significantly correlate to reading ability. WM contributed unique variance to reading fluency and reading comprehension beyond that contributed by STM, estimated IQ, word reading, and inattention. This suggests that STM and WM have distinct roles in the reading process (Swanson & Howell, 2003). Part of the reason for this finding is that STM taps into basic reading processes, including decoding and word recognition, whereas WM taps into complex mental processing and executive functioning, which are important for higher level reading processes, particularly reading fluency and comprehension. Studies with children and adults have established that WM is particularly important to high-level cognition and academic learning whereas STM is less so (Jacobson et al., 2011).

Much of the research exploring WM involves either word reading or reading comprehension (Goff et al., 2005). In addition to individual word reading, we explored reading fluency and comprehension since these reading outcomes require higher demands of WM than simple word reading. Our hypothesis regarding verbal WM contributing more than visuospatial WM on reading outcome measures was partially supported. Although WM measures were significantly correlated with one another and with both reading outcomes, there were some notable differences found between reading fluency and comprehension. First, the amount of overall variance WM accounted for in reading fluency differed from reading comprehension even after controlling for other reading processes. Because reading fluency requires responding to the text under timed conditions, processing speed may contribute more than WM when analyzing reading fluency outcomes, whereas reading comprehension requires WM to recall and maintain details and engage in metacognitive strategies to understand written text. Reading comprehension also requires other higher level cognitive processes including executive functioning and oral language (Cutting et al., 2009). Therefore, it is reasonable to suggest that reading comprehension tasks allow individuals to use WM at a higher capacity than reading fluency tasks.

The results showed that the significant contribution of WM to reading fluency and comprehension came from verbal WM measures. For example, the WRAML-2 Verbal WM is a verbal span WM measure, which assesses how well the child retains a series of animals and non-animals, and must transform the information by categorizing the verbal input not only into their respective groups but also by size from smallest to largest. However, Dehn (2008) described this subtest as a measure of executive WM rather than a verbal WM as the child must cope with secondary processing before producing a verbal output. Thus children who perform well on this measure should likely perform well on tasks that require heavy demands on verbal memory while also eliciting semantic processing, which is absent on the WISC-IV DSB subtest. Semantic processing has been found to be necessary in higher level cognition and reading comprehension (Cutting et al., 2009).

Visuospatial WM did not provide any significant contribution to reading fluency, which is a notable finding. Because reading fluency often relies on efficient orthographic processing, the child recognizes words based on the formation of visual representations of letters, letter patterns, and sequences of letters that serve to map spatially the temporal sequence of phonemes within words (Ehri, 2005). Most beginning readers learn to read by using phonological processing to decode novel words in text; however, considering the developmental age of the sample, upper elementary school students efficiently recognize words by sight or using orthographic processing (O'Brien, Miller, Wolf, Morris, & Lovett, 2011). When children are assessed using the WISC-IV-I SSB, a visuospatial task, the child is asked to remember a visual stream created by the examiner who is tapping blocks one at a time on a board; subsequently, the child must recreate the stream in a reverse sequence. One reason for why visuospatial WM did not provide unique variance in reading fluency is that these tasks might not directly tap into orthographic processing. Additionally, the child must also associate the visual form of the word with its associated phoneme or sound, when reading fluently. Because the task only requires the child to recall the visual stream or sequence of blocks, it does not tap into the symbolic representation (i.e., visual-auditory association) of the stimuli that is required in reading.

Unlike reading fluency, comprehension tasks require use of executive WM and visuospatial processing (Baddeley, 2000; Dehn, 2008), which works to regulate attention and manages both phonological and visuospatial input. Although used as a visuospatial WM task for the purposes of this study, the WRAML-2 Symbolic WM has been described also as an executive WM measure because it requires the child to organize and recall both visual (i.e., location) and linguistic input (i.e. number and letter) in an efficient manner. Additionally, coordination, transformation, and inhibition are all required for successful completion of the task. Reading comprehension works similarly as it coordinates diverse processes that culminate the integration of new information with an existing model. For example, readers focus attention on a main idea, create visual images, draw plausible inferences based on prior knowledge, and monitor understanding of the text as reading progresses (Dehn, 2008). Special attention should be made in exploring visuospatial WM in reading processes as they may be relevant indicators of efficient reading comprehension rather than solely focusing on verbal WM tasks.

This finding supports some previous research studies relating to the strength of verbal WM (Gathercole et al., 2003) in predicting reading skills, since verbal WM tasks tapped into linguistic and phonological memory processes more so than visuospatial WM tasks. However, visuospatial WM provided significant additional variance in reading comprehension, and not reading fluency. Finding supports the notion that influences of WM on reading may not be only domain-general but also domain-specific when taking into account the modality specificity and the storage demands of the each task.

Limitations and Future Directions

The sample used in the study included children with a wide range of reading ability and cognitive ability. Because the purpose of the study is to look at reading outcomes among typically developing children, group differences between below average, average, and above average readers were not explored. Future studies should explore group differences in verbal and visuospatial WM among typically developing readers and children with reading disabilities, or ADHD, given the high comorbidity between attention deficits and reading problems. Additionally, behaviors of hyperactivity and impulsivity were not included in the study, although should be considered a limitation as these behaviors can also influence academic achievement. Because we explored reading fluency which heavily relies on quick recognition of word reading under timed conditions, another limitation was lack of measures that assess processing speed. Longitudinal studies would be beneficial to determine whether there are significant changes in WM among older adolescents and adults on both reading fluency and comprehension measures, in addition to analyzing different forms of written text (e.g., informational vs. literary).

Because the focus of the study was examining WM components, the STM composite score was primarily used as a control variable to explore whether WM tasks provided additional variance on top of STM and other control variables (e.g., word reading, IQ). Thus, the study did not incorporate the breakdown of the STM composite by subtest or domain when incorporated into the regression analyses. Further research should also examine whether specific STM tasks mediate the role between WM and reading outcomes.

Another limitation was how reading comprehension was assessed in this study. Since the GORT-IV measure assesses reading comprehension using a multiple-choice format, children often relied more on recognition of the answers, rather than free recall, which often requires more complex cognitive demands and resources to answer successfully. The Gray Oral Reading Test, Fifth Edition (GORT-5; Wiederholt & Bryant, 2012) was entirely revised, and new comprehension questions were developed to address the criticisms addressed from the GORT-IV. Future research should consider including alternative methods of assessing reading comprehension, including cloze tasks or free-response questions.

Interventions which focus on reading fluency and comprehension should also consider addressing verbal and visuospatial WM as strategies to aid in reading development. Although many intervention strategies rely on verbal cues and repeated readings of texts, children also use visual modalities such as visual imagery or visual diagrams and outlines for planning and connecting main ideas. As children learn to extract meaning from text, they learn to incorporate these comprehension strategies that allow them to simultaneously process the text while recalling information for future use.

Conclusion

These results contributed to the literature by exploring specific verbal and visuospatial WM tasks in predicting reading outcomes. Based on the results of the study, verbal WM plays a significant role in reading outcomes, although there is overlapping variance between verbal and visuospatial WM which explained significant portions of variance particularly in reading comprehension, supporting previous research. Findings support the notion that visuospatial WM contributes to reading outcomes. Children with reading difficulties may likely rely on visuospatial WM or imagery as a strategy to process written text and comprehension, and this knowledge should be incorporated when developing or implementing reading interventions or WM training programs for children with these skills deficits.

References

Alloway
T. P.
Gathercole
S. E.
Pickering
S. J.
Verbal and visuospatial short-term and working memory in children: Are they separable?
Child Development
 
2006
77
1698
1716
American Psychiatric Association
Diagnostic and statistical manual of mental disorders
 
2000
4th ed.
Washington, DC
Author
text rev.
Baddeley
A. D.
Working memory
 
1986
London
Oxford University Press
Baddeley
A. D.
The episodic buffer: A new component of working memory?
Trends in Cognitive Sciences
 
2000
4
417
423
Baddeley
A. D.
Gathercole
S. E.
Papagno
C.
The phonological loop as a language learning device
Psychological Review
 
1998
105
158
173
Baddeley
A. D.
Hitch
G. J.
Bower
G. A.
Working memory
the psychology of learning and motivation: Advances in research and theory
 
1974
Vol. 8
New York
Academic Press
47
89
Baddeley
A. D.
Logie
R. H.
Miyake
A.
Shah
P.
Working memory: The multiple component model
Models of working memory: Mechanisms of active maintenance and executive control
 
1999
New York
Cambridge University Press
28
61
Badian
N.
Does a visual-orthographic deficit contribute to reading disability?
Annals of Dyslexia
 
2005
55
28
52
Bayliss
D.
Jarrold
C.
Baddeley
A.
Gunn
D.
The relationship between short-term memory and working memory: Complex span made simple
Memory
 
2005
13
414
421
Bayliss
D. M.
Jarrold
C.
Gunn
D. M.
Baddeley
A. D.
The complexities of complex span: Explaining individual differences in working memory in children and adults
Journal of Experimental Psychology: General
 
2003
132
71
92
Bussing
R.
Fernandez
M.
Harwood
M.
Hou
W.
Wilson Garvin
C.
Eyberg
S. M.
Parent and teacher SNAP-IV Ratings of Attention Deficit/Hyperactivity Disorder Symptoms
Assessment
 
2008
15
317
328
Cain
K.
Oakhill
J. V.
Bryant
P. E.
Children's reading comprehension ability: Concurrent prediction by working memory, verbal ability and component skill
Journal of Educational Psychology
 
2004
96
671
681
Cutting
L. E.
Materek
A.
Cole
C. A. S.
Levine
T. M.
Mahone
E. M.
Effects of fluency, oral, language, and executive function on reading comprehension performance
Annals of Dyslexia
 
2009
59
34
54
Dehn
M. J.
Working memory and academic learning: Assessment and intervention
 
2008
Hoboken, NJ
Jon Wiley & Sons
Ehri
L. C.
Learning to read words: Theory, findings, and issues
Scientific Studies of Reading
 
2005
9
167
188
Gathercole
S. E.
Alloway
T. P.
Willis
C.
Adams
A.
Working memory in children with reading disabilities
Journal of Experimental Child Psychology
 
2006
93
265
281
Gathercole
S. E.
Baddeley
A. D.
Working memory and language
 
1993
Hove
Lawrence Erlbaum Associates
Gathercole
S. E.
Brown
L.
Pickering
S. J.
Working memory assessments at school entry as longitudinal predictors of National Curriculum attainment levels
Educational and Child Psychology
 
2003
20
109
122
Goff
D. A.
Pratt
C.
Ong
B.
The relations between children's reading comprehension, working memory, language skills and components of reading decoding in a normal sample
Reading and Writing
 
2005
18
583
616
Hollingshead
A. A.
Four-factor index of social status
 
1975
New Haven, CT
Yale University
Unpublished manuscript
Jacobson
L. A.
Ryan
M.
Martin
R. B.
Ewen
J.
Mostofsky
S. H.
Denckla
M. B.
et al.  
Working memory influences processing speed and reading fluency in ADHD
Child Neuropsychology
 
2011
17
209
224
Jeffries
S.
Everatt
J.
Working memory: Its role in dyslexia and other specific learning difficulties
Dyslexia
 
2004
10
196
214
Kaplan
E.
Fein
D.
Kramer
J.
Delis
D.
Morris
R.
WISC-IV integrated
 
2004
San Antonio, TX
The Psychological Corporation
Lyon
G. R.
Shaywitz
S. E.
Shaywitz
B. A.
A definition of dyslexia
Annals of Dyslexia
 
2003
53
1
14
Melby-Lervag
M.
Hulme
C.
Is working memory training effective? A metaanalytic review
Developmental Psychology
 
2013
49
270
291
Milner
B.
Interhemispheric differences in the localisation of psychological processes in man
British Medical Bulletin
 
1971
27
272
277
O'Brien
B. A.
Miller
L.
Wolf
M.
Morris
R.
Lovett
M.
Orthographic processing efficiency in developmental dyslexia: An investigation of age and treatment factors at the sublexical level
Annals of Dyslexia
 
2011
61
111
135
Reiter
A.
Tucha
O.
Lange
K. W.
Executive functions in children with dyslexia
Dyslexia
 
2005
11
116
131
Sattler
J. M.
Resource guide to accompany assessment of children
 
2008
La Mesa, CA
J. M. Sattler Publishing Company
Sattler
J. M.
Dumont
R.
Assessment of children: WISC-IV and WPPSI-III supplement
 
2004
San Diego, CA
J. M. Sattler Publishing Company
Seigneuric
A.
Ehrlich
M. F.
Oakhill
J. V.
Yuill
N. M.
Working memory resources and children's reading comprehension
Reading and Writing: An Interdisciplinary Journal
 
2000
13
81
103
Sesma
H. W.
Mahone
E. M.
Levine
T.
Eason
S. H.
Cutting
L. E.
The contribution of executive skills to reading comprehension
Child Neuropsychology
 
2009
15
232
246
Shah
P.
Miyake
A.
The separability of working memory resources for spatial thinking and language processing: An individual differences approach
Journal of Experimental Psychology: General
 
1996
125
4
27
Sheslow
D.
Adams
W.
Wide range assessment of memory and learning
 
2003
2nd ed.
Lutz, FL
Psychological Assessment Resources
Swanson
H. L.
Are working memory deficits in readers with learning disabilities hard to change?
Journal of Learning Disabilities
 
2000
33
551
566
Swanson
H. L.
Does the dynamic testing of working memory predict growth in nonword fluency and vocabulary in children with reading disabilities?
Journal of Cognitive Education and Psychology
 
2010
9
51
77
Swanson
H. L.
Berninger
V.
The role of working memory in skilled and less skilled readers’ comprehension
Intelligence
 
1995
21
83
108
Swanson
H. L.
Howell
M.
Working memory, short-term memory, and speech rate as predictors of children's reading performance at different ages
Journal of Educational Psychology
 
2001
93
720
734
Swanson
H. L.
Jerman
O.
The influence of working memory growth on reading growth in subgroups of children with reading disabilities
Journal of Experimental Child Psychology
 
2007
96
249
283
Swanson
J. M.
Kraemer
H. C.
Hinshaw
S. P.
Arnold
L. E.
Conners
C. K.
Abikoff
H. B.
et al.  
Clinical relevance of the primary findings of the MTA: Success rates based on severity of ADHD and ODD symptoms at the end of treatment
Journal of the American Academy of Child and Adolescent Psychiatry
 
2001
40
168
179
Swanson
H. L.
O'Connor
R.
The role of working memory and fluency practice on the reading comprehension of students who are dysfluent readers
Journal of Learning Disabilities
 
2009
42
548
575
Titz
C.
Karbach
J.
Working memory and executive functions: Effects on training on academic achievement
Psychological Research
 
2014
Wechsler
D.
Wechsler intelligence scale for children
 
2003
4th ed.
New York
The Psychological Corporation
Wiederholt
J. L.
Bryant
B. R.
Gray oral reading test
 
2001
4th ed.
Austin, TX
Pro-Ed
Wiederholt
J. L.
Bryant
B. R.
Gray oral reading test
 
2012
5th ed.
Austin, TX
Pro-Ed
Wolf
M.
Katzir-Cohen
T.
Reading fluency and its intervention
Scientific Studies of Reading
 
2001
5
211
239
Woodcock
R. W.
McGrew
K. S.
Mather
N.
Woodcock-Johnson III normative update tests of achievement
 
2001
Itasca, IL
Riverside Publishing