Relations Among Executive Function, Decoding, and Reading Comprehension: An Investigation of Sex Differences

Abstract

In the current investigation, we used structural equation mediation modeling to examine the relations between executive function (indexed by measures of working memory, shifting, and inhibition), decoding ability, and reading comprehension in a sample of 298 6- to 8-year-old children (N =132 and 166 for boys and girls, respectively). Results for the full sample indicated that executive function was mediated by decoding ability. When sex was examined as a moderator of these associations, there was evidence for a trend suggesting that direct relations between executive function and reading comprehension were stronger for girls compared to boys; no significant differences were found for other direct and indirect relations. Taken together, these findings highlight the importance of executive function in supporting underlying integrative processes associated with reading comprehension and emphasize the need to further consider the role of executive function in relation to reading.

Reading comprehension, or the ability to extract meaning from text (Snow, 2002), is a highly complex process that relies on the integration of multiple cognitive skills (Kendeou & Trevors, 2012). One skill that may be associated with such integrative processes is executive function, which is defined as the ability to plan, inhibit, problem solve, and update and/or maintain information (Miyake et al., 2000). Executive function is a domain-general skill that is associated with a variety of literacy-related skills, including the ability to engage in fluent word recognition, make inferences about the text, and integrate background knowledge with the information presented in the text, all of which have been implicated in reading comprehension (Currie & Cain, 2015; Jacobson et al., 2011; Kintsch, 1988; Klauda & Guthrie, 2008; Potocki, Sanchez, Ecalle, & Magnan, 2017; van den Broek, Lorch, Linderholm, & Gustafson, 2001). Thus, the aim of the current study was to further elucidate the potential integrative role of executive function in reading comprehension by investigating relations between executive function, decoding, and reading comprehension within a mediation model framework. Furthermore, given evidence that boys and girls exhibit differential performance across both reading comprehension and word reading (e.g., Lietz, 2006; Rutter et al., 2004), we additionally examined whether sex moderated the observed associations.

The association between executive function and life outcomes has been well established over the past 50 years (e.g., Mischel et al., 2010). However, examinations of relations between executive function and academic performance (and literacy-related outcomes, in particular) are a more recent area of exploration (e.g., Altemeier, Abbott, & Berninger, 2008; Blair & Razza, 2007; Cartwright, 2012; Schmitt, Geldhof, Purpura, Duncan, & McClelland, 2017). Yet, there has been a substantial increase in the number of studies investigating the impact of executive function on reading comprehension in recent years (e.g., Authors, 2019; Christopher et al., 2012; Cutting, Materek, Cole, Levine, & Mahone, 2009; Iglesias-Sarmiento, Lopez, & Rodriguez, 2015; Karlsson, Jolles, Koornneef, van den Broek, & Van Leijenhorst, 2019; McVay & Kane, 2012; Sesma, Mahone, Levine, Eason, & Cutting, 2009). Although this has led to a greater understanding of individual differences in reading comprehension and the acknowledgement of its importance via its inclusion within more recent theoretical frameworks of reading comprehension, a substantial proportion of these studies have focused on adolescents as opposed to young readers. This is a potential limitation of previous work because associations between reading comprehension and reading-related component skills, such as decoding, can shift over time (Catts, Hogan, & Adolf, 2005). Thus, the inclusion of executive function in a younger sample may provide some insight into how these skills may be differentially associated for early readers, especially given that executive function develops early on (Garon, Bryson, & Smith, 2008).

Executive Function

Executive function is comprised of a variety of subskills. The most common conceptualization of executive function identifies three central abilities, which are working memory, shifting, and inhibition (Miyake et al., 2000). Although these skills are described as being distinct, there remains substantial overlap in their associations (Miyake & Friedman, 2012), suggesting the need to study them in conjunction. Working memory, or the ability to store and manipulate information (Baddeley, 1992); shifting, or the ability to cognitively shift between tasks (Diamond, 2013); and inhibition, or the ability to delay or stop a behavioral response (Barkley, 1997b), are each associated with both decoding and reading comprehension (e.g., Arrington, Kulesz, Francis, Fletcher, & Barnes, 2014; Meixner, Warner, Lensing, Schiefele, & Elsner, 2019; Morgan et al., 2019; Ober, Brooks, Plass, & Homer, 2019; Yeniad, Malda, Mesman, van IJzendoorn, & Pieper, 2013).

The associations between word reading and executive function are threefold. First, words’ orthographic, phonological, and semantic information is stored in memory and utilized during word recognition (Inhoff, Connine, Eiter, Radach, & Heller, 2004; Seidenberg & McClelland, 1989). This is evidenced by studies showing that working memory skills support decoding processes for individuals with decoding weaknesses (Hamilton, Freed, & Long, 2016). Second, word reading requires individuals to continuously move between multiple sources of information (e.g., Berninger & Nagy, 2008; Cole, Duncan, Blaye, 2014; Coltheart, Rastle, Perry, Langdon, & Ziegler, 2001). This view is supported by studies showing that children with dyslexia or reading disability frequently have difficulty completing cognitive flexibility or shifting tasks (e.g., Reiter, Tuchs, & Lange, 2004), suggesting that decoding difficulties may be associated with the inability to devote attention to multiple aspects of words simultaneously (e.g., Bialystok & Niccols, 1989). Third, inhibitory processes, such as the suppression of irrelevant word-level information (e.g., Coltheart et al., 2001), are important for word recognition (e.g., Conrad, Carreiras, Tamm, & Jacobs, 2009; Ziegler & Muneaux, 2007). This may explain why children with word reading difficulties display corresponding deficits in inhibition (e.g., Booth, Boyle, & Kelley, 2014; Chiappe, Hasher, & Siegel, 2000; Locascio, Mahone, Eason, & Cutting, 2010; Reiter et al., 2004) and also why attention-deficit hyperactivity disorder (ADHD) and reading disability co-occur (Dykman & Ackerman, 1991; Willcutt & Pennington, 2000).

Because reading comprehension requires the integration of several cognitive processes, executive function resources are thought to be foundational for reading comprehension as well. Working memory, shifting, and inhibition are associated with various reading comprehension-related processes, including inference generation, metacognition, and suppression of irrelevant information, respectively (Cain, 2006; Cain, Oakhill, & Bryant, 2004; Potocki et al., 2017; Torgesen, 1994). In this way, executive function skills may serve to facilitate integration by further strengthening the associations between fundamental comprehension-related skills (e.g., Cutting, Bailey, Barquero, & Aboud, 2015), which would allow readers to engage in more goal-directed behaviors during reading comprehension. This may explain why children with poor reading comprehension often have working memory deficits, perform poorly on measures of cognitive flexibility, and have greater interference of irrelevant information during text comprehension (Cain, 2006; Cartwright et al., 2017; De Beni & Palladino, 2000). We offer a more detailed discussion of the role of executive function in within theories of reading comprehension below.

Executive Function Within Theoretical Frameworks of Reading Comprehension

Multiple theories of reading comprehension exist, and all aim to capture its complexity. Perhaps the most well-known theoretical frameworks include the Construction-Integration (C-I) Model (Kintsch, 1988), the Landscape Model (van den Broek, Young, Tzeng, & Linderholm, 1999), the Reading Systems Framework (Perfetti, 1999), and the Structure Building Framework (Gernsbacher, 1991). Each of these models acknowledges that additional reader characteristics can interact with text-based features in a way that potentially further impacts reading comprehension. Although executive function (beyond working memory) is not explicitly discussed within these models, each of these reading comprehension frameworks highlight the vital integrative processes and associated cognitive abilities that rely on executive function-related skills.

The C-I Model (Kintsch, 1988) asserts that during text comprehension, readers are constructing mental representations of what is being read based on their own background knowledge. Following the initial activation of concepts encountered in the text, readers begin assimilating different sources of information to create a network of related concepts and are continuously determining the relevance of those that are activated. During this process, readers are additionally engaged in inference making, which results in the activation of additional concepts not available in the text. At this point, however, a coherent representation of the text has not yet been established. This occurs during the integration process. During integration, readers are shifting between various sources of information within their mental (situation) model to exclude irrelevant information (e.g., lexical information and inferences). Consequently, executive function resources facilitate the reading comprehension process (Kintsch, 1998). For instance, the retrieval of information from background knowledge involves working memory resources, the creation and maintenance of connections among concepts potentially necessitates cognitive shifting skills, and the suppression of irrelevant information requires inhibitory control.

Similar to the C-I Model, the Landscape Model (van den Broek et al., 1999) states that as readers are working their way through a text, they develop a mental representation of the text by creating a network of activated and associated concepts, updating them based on previously activated information (e.g., prior words and sentences), and retrieving them from memory. It is a jointly passive and active iterative process whereby readers are striving to develop a coherent representation of the text according to their own standards (van den Broek & Helder, 2017). During the knowledge-text-base integration process, executive function-related skills are being taxed. For instance, working memory resources constrain the number of concepts that can be concurrently activated. Further, readers are required to shift between multiple sources of information in the text as well as between multiple activated concepts and are required to suppress or inhibit the activation of information that is either inconsistent or erroneously activated.

Comparably, the Reading Systems Framework (Perfetti, 1999; Perfetti & Stafura, 2014) is an interactive model that asserts that readers are integrating different sources of information from both word reading and linguistic inputs during reading comprehension. Readers are constantly shifting between word identification processes, meaning-based information, and their own background knowledge to facilitate their understanding of the text. As within the C-I Model, the Reading Systems Framework asserts that reading comprehension is influenced by the interaction between inference making skills, background knowledge, and mental representations of the text. This means that during reading comprehension, working memory resources are being tapped as readers are maintaining and integrating bottom-up text-based information (e.g., words) with higher-level top-down processes, such as comprehension monitoring (Williams, 2003).

According to the Structure-Building Framework (Gernsbacher, 1991), individuals are continuously building mental models (structures) of the text as they read. Essentially, readers create text representations based on initial information presented and continue adding to their mental model as they encounter new incoming information. This continues until a cohesive and integrated mental model is achieved. Given its interactive nature, structure building processes depend on executive function resources. Working memory skills are needed for information retrieval and updating whereas inhibition is required for the suppression of irrelevant information, such as when word meanings are incorrectly activated (Gernsbacher & Faust, 1991).

The Potential Moderating Effect of Sex

There is some evidence for sex differences across word reading, reading comprehension, and executive function performance. For example, on large-scale reading comprehension assessments, girls tend to attain higher scores than boys (Lietz, 2006). Boys, on the other hand, are more likely to receive a greater proportion of reading difficulty diagnoses relative to girls (i.e., poor word recognition/decoding; e.g., Rutter et al., 2004; Quinn & Wagner, 2015). With regards to executive function, a few studies have shown that males tend to outperform females on working memory tasks (Voyer, Voyer, Saint-Aubin, 2017) whereas females perform better on measures of inhibitory control and attention shifting (Klenberg, Korkman, & Lahti-Nuutila, 2001; Yuan, He, Qinglin, & Li, 2008). These observed discrepancies may be due to underlying neurobiological differences (e.g., Shaywitz et al., 1995; Zilles et al., 2016) or to differences in how the two groups approach and process these tasks (e.g., Morgan & Fuchs, 2007; Thompson, 1987; Weiss et al., 2006).

Although the presence of sex differences is not consistent across investigations (e.g., Chiarello et al., 2009; Cross, Copping, & Campbell, 2009; McGeown, Goodwin, Henderson, &Wright, 2013; Robert & Savoie, 2007), it remains an important area of inquiry. First, several investigations that include sex as a covariate have shown significant associations between sex and reading-related outcomes (e.g., Ehm et al., 2016; Pfost, Dorfler, & Artelt, 2013; Reynolds & Turek, 2012), suggesting that its role is likely not negligible. Second, the existence of sex differences in the relation between executive function and reading, if any, would serve to inform our theoretical understanding of reading comprehension by highlighting an additional reader characteristic that may contribute to individual differences in reading comprehension that could be included within theoretical frameworks. Third, understanding how sex differences are related to associations between executive function, decoding, and reading comprehension could have practical implications as well. For instance, stronger associations in either sex would provide valuable insight into additional cognitive skills (e.g., executive function) that could be targeted in conjunction with reading comprehension for that group. Yet, despite the fact that studies have examined direct and indirect associations between executive function and reading comprehension (e.g., Arrington et al., 2014; Authors, 2019; Cantin, Gnaedinger, Gallaway, Hesson-McInnis, & Hund, 2016; Christopher et al., 2012; Georgiou & Das, 2018; Kieffer, Vukovic, & Berry, 2013; Martinussen & Mackenzie, 2015; Ober et al., 2019) as well as moderators of associations (e.g., Arrington et al., 2014; Cirino et al., 2019), these investigations did not examine a moderated mediation model that included executive function, sex, decoding, and reading comprehension. Consequently, it is unclear if the observed advantage that girls have in reading comprehension, word reading, and aspects of executive function across studies (Lietz, 2006; Klenberg et al., 2001; Yuan et al., 2008; Rutter et al., 2004) is indicative of differential associations among the three skills.

Theoretical rationale from the neurobiological literature additionally provides support for the examination of sex differences in executive function and reading-related skills. For instance, males tend to exhibit greater brain lateralization (i.e., the primary use of a single hemisphere during a particular task) compared to females (Clements et al., 2006). This is potentially important because language processing is often associated with activation within the left hemisphere (e.g., Frost et al., 1999) whereas reading-related tasks may rely more on the recruitment and coordination of brain regions across both hemispheres. This is evidenced by greater bilateral activation during more complex language comprehension as well as reading comprehension tasks (e.g., Berl et al., 2010; Jung-Beeman, 2005). Thus, differences in brain lateralization across the sexes, coupled with developmental differences in biological maturation (Dwyer, 1973; Lim, Han, Uhlhaas, & Kaiser, 2013), may lead girls to acquire language- and literacy-related skills earlier than boys (e.g., Adani & Cepanec, 2019; Verhoeven & van Leeuwe, 2011). This may subsequently impact observed associations between skills.

Earlier developmental theories of executive function posit that executive function, namely inhibition-related skills, are associated with the maturation of the pre-frontal cortex (e.g., Barkley, 1997a), a brain region that does demonstrate sex-related functional differences during inhibitory control tasks (Garavan, Hester, Murphy, Fassbender, & Kelly, 2006). Such differences may explain why boys are more likely to have externalizing behavior problems and reduced inhibitory control compared to girls (Leadbeater, Kiperminc, Blatts, & Hertzog, 1999; Yuan et al., 2008). Although associations between executive function development and sex do exist (e.g., de Luca et al., 2003), such relations may not necessarily be due to differences in biological maturation (see Wierenga, Bos, van Rossenberg, & Crone, 2019) and may instead be the result of evolutionary socio-cultural differences (Barkley, 2001). This perspective posits a female advantage with regards to some executive function-based skills (Bjorklund & Kipp, 1997).

The Current Study

In the current study, we examined associations between executive function (as indexed by measures of working memory, shifting, and inhibition), decoding, and reading comprehension. First, we investigated the magnitude of the predictive relations between executive function and reading comprehension over and above decoding and age. Second, we examined the indirect associations between executive function and reading comprehension via decoding. Third, we investigated the extent to which sex moderated the observed associations. In this way, we aimed to build upon the existing literature by further elucidating the specific role of executive function in reading comprehension while also taking into consideration the potential for sex differences.

Given prior work demonstrating that that executive function-related skills potentially facilitate reading comprehension (e.g., Cutting et al., 2015) and the female advantage on reading comprehension-related tasks (e.g., Lietz, 2006), we hypothesized that:

1. Executive function would account for unique variance in reading comprehension and that decoding would mediate these associations.

2. The contribution of executive function skills to reading comprehension would be greater for girls than boys.

Materials and Method

Participants were part of a screening database for a larger reading comprehension study in the Southeastern United States. Screening was conducted in schools by research assistants who had ≥ 90% fidelity accuracy in administering all measures. Assessment occurred across one session and included a short battery of standardized and experimenter-created cognitive and behavioral measures. These included two mathematical measures (numeracy and calculation), two reading-related measures (nonword decoding and passage comprehension), and teacher ratings of students’ executive function. Only reading-related and executive function measures were used in the current investigation.

A total of 482 children between 6.33 and 8.33 years of age were screened (M_age = 7.20 years; SD_age = 0.36 years). Approximately half of this sample was female (53.32%), and participants were identified as 74.07% White, 15.15% Black/African American, 7.47% Multiracial, and 2.70% Asian; 0.62% did not answer. However, only a subsample of the full screening sample was included in the analysis due to missing data. The final analysis sample included 298 children between 6.42 and 8.33 years (M_age = 7.24 years; SD_age = 0.34 years). Participants were identified as White (75.17%), Black/African American (14.77%), Multiracial (6.71%), and Asian (3.36%); there were slightly more females (55.70%) than males. Written parental consent and child assent were obtained, and all study-related procedures were conducted in accordance with the Institutional Review Board.

Measures

In the current study, we included several measures of executive function, decoding, and reading comprehension.

BRIEF-2 (Gioia, Isquith, Guy, & Kenworthy, 2015). Teachers are asked to rate the frequency (never, sometimes, often) of students’ various problematic behaviors for the past six months. The assessment consists of 63 items total. The Shift (8 items), Inhibit (8 items), and Working Memory (8 items) clinical subscales were used. The Shift, Inhibit, and Working Memory subscales assess the ability to shift between tasks, inhibit impulsive behaviors, and retain information needed to complete tasks, respectively. The assessment uses reverse scoring. Reported reliability for the BRIEF-2 teacher forms exceed .90 (Gioia et al., 2015). T scores were used (i.e., standardized scores with a mean of 50 and a standard deviation of 10).

Woodcock-Johnson, Fourth Edition (WJ-IV; Schrank, Mather, & McGrew, 2014). Two subtests from the WJ-IV were used. Word Attack requires participants to read a list of nonwords across 32 items. Participants are awarded one point for each correct answer, and administration ends after six consecutive errors. Reported reliability for this assessment is high (.94 – .96 for children 6–8 years; McGrew, LaForte, & Schrank, 2014). W scores were used (i.e., Rasch ability transformed scaled scores representative of a child’s developmental level).

Passage Comprehension requires participants to read passages and provide a missing word across 52 items. This subtest is untimed. Administration is discontinued after six consecutive errors. Reported reliability for this subtest is high (.93 – .98 for children 6–8 years; McGrew et al., 2014). W scores were used.

Analytic Approach

The original sample included 482 participants; however, over 38% had missing data for the BRIEF-2 (N = 184). Given that this missingness was likely systematic (i.e., teachers who failed to complete the BRIEF-2 likely did so for all students who were screened within their classroom rather than a random subset of students within their class), we decided to exclude all students who did not have complete data for the BRIEF-2. Children with missing data on measures of reading comprehension and/or decoding were allowed to remain in the analyses, and missingness for these measures was examined separately. Thus, the final analysis sample consisted of 298 children. We analyzed data using SPSS software (Version 25.0; IBM Corp., 2017) and Mplus software (Version 7.11; Muthen & Muthen, 2013). For the mediation models, we conducted bootstrapping with 5,000 iterations to obtain bias-corrected (BC) 95% confidence intervals (CI), and unstandardized parameter estimates are reported. We specified multi-group models to examine whether sex moderated relations between variables; parameter constraints were used to test the equality of the estimates across boys and girls. Model fit was based on the chi-square statistic (χ²), comparative fit index (CFI), Tucker-Lewis index (TLI), root mean square error of approximation (RMSEA), and standardized root mean square residual (SRMR). A good-fitting model would have non-significant chi-square values, CFI and TLI values greater than .95, RMSEA values at or below .05, and SRMR values at or below .08 (Hu & Bentler, 1999; MacCallum, Browne, & Sugawara, 1996).

Results

Preliminary Analyses

Prior to structural equation modeling (SEM), we investigated missingness, outliers, skewness, and kurtosis for the full analytical sample. We identified univariate outliers using the median +/– 2 interquartile ranges criterion. Using this criterion, 2.84% of the data included outliers (0.27% and 2.57% at the lower and upper ends, respectively). Outlier values were then replaced with values at the lower or upper end of the range depending on whether the values were low or high, respectively. Mahalanobis distance was used to identify multivariate outliers across the five measures; none were detected. Following this, we examined skewness and kurtosis, which were within the acceptable range of +/– 2 (see Table 1 for descriptive statistics and correlations for boys and girls and Appendix A for the full sample). Next, we examined missingness for the reading comprehension and decoding measures (0.81% of the total data for the analysis sample), we found that these data were missing completely at random regardless of whether the BRIEF was included in the analysis [Little’s MCAR test χ² = 12.916 (11), p = .299 vs. χ² = 2.988 (2), p = .224 with and without the BRIEF-2, respectively]. Given the observed pattern of missingness, we used maximum likelihood estimation during SEM.

Table 1.

Descriptive Statistics and Correlations Across Measures of Decoding, Reading Comprehension, and Executive Function for Boys and Girls

Sample	Variables	1	2	3	4	5	6
Boys
	1. Age	–
	2. WJ-IV Word Attack a	.011	–
	3. WJ-IV Passage Comp. a	.148	.622**	–
	4. BRIEF-2 Shift b	−.104	−.253**	−.114	–
	5. BRIEF-2 Inhibit b	−.202*	−.136	−.120	.532**	–
	6. BRIEF-2 WM b	−.097	−.253**	−.297**	.614**	.602**	–
	N	132	131	128	132	132	132
	Observed Min. to Max.	6.50–8.17	455–522	441–503	40–69	39–76	36–76
	Mean	7.293	490.595	477.180	48.856	48.955	51.379
	SD	0.341	13.666	14.171	9.738	11.341	11.848
	Skew	−0.075	−0.727	−0.363	0.973	1.036	0.615
	Kurtosis	−0.232	0.116	−0.220	−0.344	−0.096	−0.871
Girls
	1. Age	–
	2. WJ-IV Word Attack a	.047	–
	3. WJ-IV Passage Comp. a	.023	.590**	–
	4. BRIEF-2 Shift b	.044	−.135	−.164*	–
	5. BRIEF-2 Inhibit b	.050	−.098	−.211**	.424**	–
	6. BRIEF-2 WM b	.062	−.187*	−.309**	.563**	.590**	–
	N	166	162	163	166	166	166
	Observed Min. to Max.	6.42–8.33	456–517	441–512	40–69	43–76	41–76
	Mean	7.202	487.130	475.135	47.319	51.892	50.940
	SD	0.327	11.896	13.039	8.137	9.708	9.674
	Skew	0.100	−0.304	−0.358	1.347	1.023	1.262
	Kurtosis	−0.005	0.096	0.188	0.915	−0.006	0.651

Note. Across correlations, N range = 127–132 and 161–166 for boys and girls, respectively. The BRIEF-2 is reverse scored. WJ-IV = Woodcock-Johnson, Fourth Edition; BRIEF-2 = Behavioral Rating Inventory of Executive Function, Second Edition, Teacher Form; Min. = Minimum; Max. = Maximum; SD = Standard deviation;

^aW scores;

^bT scores.

^*p < .05

^**p < .01

SEM

Executive function was measured using three subscales from the BRIEF-2. Thus, we created an averaged composite score of executive function to include within the model. Using multi-group modeling, we first modeled boys and girls separately and then examined whether sex moderated observed associations. Age was included as a control variable (see Appendix B).

We began by specifying a fully unconstrained model (see Appendix C). We then tested whether the magnitudes of the estimates varied across boys and girls by systematically constraining (direct and indirect) parameters to equality and using chi-square difference testing to examine whether specific parameter constraints impacted model fit (see Table 2). First, we constrained the covariance between age and executive function to equality. This resulted in a significant degradation of model fit (Δχ² = 3.900 (1), p = .048), suggesting that differences in the magnitudes of the estimates exist; thus, we removed this constraint for all subsequently tested models. No significant differences were found between boys and girls for associations between age and reading comprehension and age and decoding. Similarly, no differences were observed in the magnitudes of the estimates for the path from decoding to reading comprehension. When we examined executive function, however, we found a trend for the direct association between executive function and reading comprehension (Δχ² = 3.626 (1), p = .057)¹; no other associations were significantly different across sex. Thus, in our final model, we allowed the covariance between age and executive function to vary across groups as well as the direct effect between executive function and reading comprehension; all other predictive paths were constrained to equality (see Table 3). This final model provided a good fit to the data (χ²= 2.554 [4], p = .635; CFI= 1.000; TLI = 1.024; RMSEA = 0.000, 90% CI [0.000, 0.101], p-close = .772; SRMR = 0.122) and predicted nearly 40% of the variance in reading comprehension (boys R² = .382; girls R² = .396).

Table 2.

Model Comparisons for Tested Parameter Constraints

			Difference Testing
Model	Path Constraint	Comparison	Δχ2	Δdf	p-value
Model A	–	–
Model B	Covariance between Age and EF a	Model B vs. A	3.900	1	.048
Model C	Executive Function → Reading Comprehension	Model C vs. A	3.626	1	.057
Model D	Executive Function → Decoding → Reading Comprehension	Model D vs. C	0.258	1	.611
Model E	Age → Reading Comprehension	Model E vs. D	1.441	1	.230
Model F	Age → Decoding → Reading Comprehension	Model F vs. E	0.500	1	.480
Model G	Decoding → Reading Comprehension b	Model G vs. F	0.137	1	.711

Note. χ² = Chi-square; df = Degrees of freedom.

^aThis constraint was freed across Models C–G.

^bThe inclusion of this constraint resulted in additional simultaneous parameter constraints for the paths between age and decoding and executive function and decoding.

Table 3.

Parameter Estimates for the Final Multi-Group Mediation Model

			Bias-Corrected 95% CI
Group	Path	Estimate	LL	UL
Boys
	Age → Reading comprehension	2.753	−0.706	6.449
	Decoding → Reading comprehension	0.619	0.519	0.724
	Executive function → Reading comprehension	−0.055	−0.297	0.177
	Age → Decoding	0.852	−3.447	5.105
	Executive function → Decoding	−0.308	−0.485	−0.129
	Age → Decoding → Reading comprehension	0.528	−2.149	3.191
	Executive function → Decoding → Reading comprehension	−0.191	−0.316	−0.080
	Covariance between age and executive function	−0.502	−1.108	0.062
Girls
	Age → Reading comprehension	2.753	−0.706	6.449
	Decoding → Reading comprehension	0.619	0.519	0.724
	Executive Function → Reading comprehension	−0.333	−0.597	−0.084
	Age → Decoding	0.852	−3.447	5.105
	Executive function → Decoding	−0.308	−0.485	−0.129
	Age → Decoding → Reading comprehension	0.528	−2.149	3.191
	Executive function → Decoding → Reading comprehension	−0.191	−0.316	−0.080
	Covariance between age and executive function	0.157	−0.182	0.508

Note. Estimates are unstandardized. Confidence intervals (CI) that do not include zero are significant (bolded). LL = Lower level; UL = Upper level.

For both boys and girls, decoding was directly related to reading comprehension (b = 0.619, bias-corrected 95% CI [0.519, 0.724]; see Table 3)². Further executive function was directly related to decoding (b = –0.308, bias-corrected 95% CI [–0.485, –0.129]) and indirectly associated with reading comprehension via decoding (b = –0.191, bias-corrected 95% CI [–0.316, –0.080]). For boys, executive function was not directly associated with reading comprehension (b = –0.055, bias-corrected 95% CI [–0.297, 0.177]) whereas this relation was significant for girls (b = –0.333, bias-corrected 95% CI [–0.597, –0.084]). Taken together, these findings suggest some evidence for stronger a direct association between executive function and reading comprehension for girls compared to boys (Δb = 0.278).

Discussion

A growing body of evidence indicates that executive function skills are related to children’s academic performance (Biederman et al. 2004; Blair & Razza, 2007; Bull, Espey, & Wiebe, 2008). Even though prior studies have investigated associations between executive function and reading comprehension, a relatively limited number have focused on young children and even fewer have examined sex differences. Consequently, the current investigation aimed to fill this gap in the literature. In the present study, we investigated associations between executive function, decoding, and reading comprehension in a large sample of elementary-aged readers and also whether these relations were moderated by sex.

Executive function exhibited direct and indirect effects on reading comprehension via decoding. The observed association between executive function and decoding may be explained by the fact that decoding was based on a nonword reading task (i.e., phonological decoding; Simos, Breier, & Fletcher, 2001) and thus relies on the ability to access and manipulate phonological representations from memory (Baddeley, Lewis, & Vallar, 1984), shift between phonological units (Goswami, Ziegler, Dalton, & Schneider, 2003), and inhibit irrelevant phonological information (Colangelo & Buchanan, 2006). Therefore, it is perhaps not surprising that our measure of executive function, which was comprised of working memory, shifting, and inhibition, would be associated with decoding. There was no sex difference detected for the indirect effect of executive function on reading comprehension; however, there was some evidence that girls had a stronger direct association between executive function and reading comprehension (p = .057). Although we did not examine this directly in the current study, this pattern of findings may be due, in part, to the observed female advantage in both executive function- and reading comprehension-related skills (e.g., Klenberg et al., 2001; Lietz, 2006; Yuan et al., 2008).

The observed association between executive function and reading comprehension is supported by numerous studies showing that executive function is related to reading comprehension (e.g., Arrington et al., 2014; Authors, 2019; Cain et al., 2004; Cartwright, 2015; Cartwright et al., 2010; Cirino et al., 2019; Georgiou & Das, 2018; Karlsson et al., 2019; Ober et al., 2019). Our results align with comprehensive frameworks of reading comprehension, including the C-I Model (Kintsch, 1988), Landscape Model (van den Broek et al., 1999), Reading Systems Framework (Perfetti, 1999), and Structure-Building Framework (Gernsbacher, 1991), all of which assert that reader-based characteristics (e.g., retrieval of relevant concepts from memory, ability to shift between multiple sources of information, and ability to suppress irrelevant/incorrect information) likely interact with the text in ways that can support integration processes and also lead to individual differences in reading comprehension. This outcome provides additional support for theoretical frameworks of reading comprehension that include executive function-related skills as mediators and supports of bottom-up and top-down reading comprehension processes (Authors, 2019; Cutting et al., 2015). Further, it is also interesting to compare the current findings to those from studies examining direct and inferential mediation models, in which nearly all of the variance in reading comprehension was explained without including multiple aspects of executive function (e.g., Ahmed et al., 2016; Kim, 2017). This discrepancy may be explained by the fact that we did not include oral language, background knowledge, and/or reading-specific strategies in our models (Elbro & Buch-Iversen, 2013; Hoover & Gough, 1998; Kintsch, 1988).

The observed association between executive function and decoding, which was similar across boys and girls, was anticipated based on models of word recognition (Coltheart et al., 2001; Seidenberg & McClelland, 1989; Zorzi et al., 1998). Many of these frameworks argue that successful word recognition results from the ability to process, integrate, and inhibit multiple features of words during word reading, a process that likely relies heavily on executive function-based skills. There is also an abundance of empirical evidence in support of these relations (Arrington et al., 2014; Peng et al., 2018; Yeniad et al., 2013).

Theoretical and Practical Implications of the Findings

Our results have both theoretical and practical implications. First, the findings provide greater insight into additional reader characteristics that may contribute to individual differences in reading comprehension. Our results emphasize the need for executive function to be integrated within theoretical frameworks of reading comprehension, as even the more comprehensive integrative frameworks (e.g., C-I Model) do not explicitly include a comprehensive construct of executive function. Although it is generally acknowledged that certain components of executive function, such as working memory, are important for the reading comprehension process (e.g., Perfetti & Stafura, 2014), a more explicit inclusion of executive function may be warranted (e.g., Cutting et al., 2015). Second, although the present findings are correlational, the models provide a foundation for better understanding how executive function interacts with other literacy-based skills in a way that may subsequently influence reading comprehension, such as the fact that executive function in the current study was comprised of skills beyond working memory. This is in line with interactive theories of reading comprehension, which assert that the integration of bottom-up and top-down processes potentially relies on additional dimensions of executive function, such as cognitive shifting or inhibition, in addition to the activation and manipulation of information from memory (Kintsch, 1988; Perfetti, 1999). Third, the observed trend for sex differences regarding the direct associations between executive function and reading comprehension provides some support for the potential inclusion of additional sources of individual differences within models of reading comprehension, including those that are nonmalleable. Finally, the association between executive function and decoding aligns well with multiple theories of word recognition, which argue for the need to activate, integrate, and shift between multiple aspects of word features (e.g., Coltheart et al., 2001).

Turning to practical implications, the findings suggest, first, that measures of executive function may be a potentially valuable addition to test batteries designed to identify children who struggle with decoding or reading comprehension. This is significant given that executive function skills tend to develop prior to formal schooling (Garon et al., 2008) and thus may facilitate the early identification of poor reading outcomes. Second, the inclusion of executive function within screening batteries may be sensitive in detecting reading comprehension-related difficulties in girls given the potential for executive function to be more strongly directly associated with reading comprehension for girls (–0.333 vs. –0.055 for girls vs. boys, respectively). However, this must be approached cautiously, as our data demonstrated only a trend for differences in relations. Overall, the findings further elucidate what is known about the interrelations between executive function, decoding, and reading comprehension and provide further information as to how executive function facilitates in the integration of reading comprehension processes regardless of sex.

Study Limitations and Conclusions

Before closing, we note a few limitations. First, no language variables were included within our models. Given the impact of language comprehension on reading comprehension (e.g., Hoover & Gough 1990), the implications of the current study are limited with regards to our understanding of the contribution of executive function over and above language-related skills. Second, although we included measures that captured the three main facets of executive function (Miyake et al., 2000), other aspects of executive function are also related to reading comprehension (e.g., Kieffer et al., 2013), and the inclusion of these skills may change the observed pattern of findings. Relatedly, measurement factors may also account for these observed relations. For example, the BRIEF is an indirect measure of executive function; direct measures of executive function, even across the same constructs, may have different relations. Fourth, because our data are concurrent, it is unknown whether or not these relations remain the same over the course of development. Fifth, we acknowledge that our findings can only be extrapolated to English-speaking readers, and the large portion of missing data for the original sample may impact the overall generalizability of the results. Finally, we had limited power to detect sex differences in effects given that multi-group modeling results in reduced sample sizes. Taken together, these limitations highlight the need for future studies to include additional predictors and measures of executive function and more diverse samples.

Despite these limitations, the present investigation contributes to our understanding of the association between executive function skills and lower- and higher-order literacy-related skills. Specifically, the findings of our study facilitate a greater understanding of the process by which executive function may support reading comprehension and how this may be different for boys versus girls. Gaining a greater understanding of the ways in which domain-general cognitive skills are associated with reading comprehension, such as the potential for differing relations across sex, can lead to the generation of more comprehensive theoretical frameworks of reading comprehension as well as the development of more effective identification and remediation programs for children with reading comprehension difficulties.

Appendix A.

Descriptive Statistics and Correlations Across Measures for the Full Sample

Variables	1	2	3	4	5	6	7
1. Age	–
2. Sex	–.135*	–
3. WJ-IV Word Attack a	.046	−.135*	–
4. WJ-IV Passage Comp. a	.091	−.075	.608**	–
5. BRIEF-2 Shift b	−.018	−.086	−.183**	−.133*	–
6. BRIEF-2 Inhibit b	−.091	.139*	−.133*	−.174**	.462**	–
7. BRIEF-2 WM b	−.015	−.020	−.217**	−.301**	.590**	.587**	–
N	298	298	293	291	298	298	298
Observed Min. to Max.	6.42–8.33	0–1	455–522	441–512	40–69	39–76	36–76
Mean	7.242	0.557	488.679	476.034	48.000	50.591	51.134
SD	0.336	0.498	12.812	13.563	8.899	10.546	10.675
Skew	0.034	−0.231	−0.467	−0.338	1.173	0.941	0.899
Kurtosis	−0.168	−1.960	−0.029	−0.037	0.254	−0.138	−0.231

Note. Across correlations, N range = 288–298. The BRIEF-2 is reverse scored. WJ-IV = Woodcock-Johnson, Fourth Edition; Comp. = Comprehension; BRIEF-2 = Behavioral Rating Inventory of Executive Function, Second Edition, Teacher Form; Min. = Minimum; Max. = Maximum; SD = Standard deviation;

^aW scores;

^bT scores.

^*p < .05

^**p < .01

Appendix B.

Tested Partial Mediation Model

Note. Executive function (EF) is an averaged composite represented by the Working Memory, Shift, and Inhibit subscales from the Behavioral Rating Inventory of Executive Function, Second Edition, Teacher Form; Comp. = Comprehension.

Appendix C.

Parameter Estimates for the Initial Unconstrained Multi-Group Mediation Model

			Bias-Corrected 95% CI
Group	Path	Estimate	LL	UL
Boys
	Age → Reading comprehension	5.578	−0.084	11.672
	Decoding → Reading comprehension	0.630	0.487	0.773
	Executive function → Reading comprehension	−0.034	−0.284	0.210
	Age → Decoding	−1.199	−7.803	5.584
	Executive function → Decoding	−0.371	−0.634	−0.099
	Age → Decoding → Reading comprehension	−0.756	−5.072	3.464
	Executive function → Decoding → Reading comprehension	−0.234	−0.447	−0.065
	Covariance between age and executive function	−0.502	−1.108	0.062
Girls
	Age → Reading comprehension	0.663	−3.674	5.001
	Decoding → Reading comprehension	0.614	0.470	0.764
	Executive Function → Reading comprehension	−0.328	−0.595	−0.074
	Age → Decoding	2.043	−3.345	7.288
	Executive function → Decoding	−0.265	−0.494	−0.025
	Age → Decoding → Reading comprehension	1.254	−1.991	4.493
	Executive function → Decoding → Reading comprehension	−0.162	−0.317	−0.018
	Covariance between age and executive function	0.157	−0.182	0.508

Note. Estimates are unstandardized. Confidence intervals (CI) that do not include zero are significant (bolded). LL = Lower level; UL = Upper level.