Executive functions and components of oral reading fluency through the lens of text complexity

Abstract

As readers struggle to coordinate various reading- and language-related skills during oral reading fluency (ORF), miscues can emerge, especially when processing complex texts. Following a miscue, students often self-correct as a strategy to potentially restore ORF and online linguistic comprehension. Executive functions (EF) are hypothesized to play an interactive role during ORF. Yet, the role of EF in self-corrections while reading complex texts remains elusive. To this end, we evaluated the relation between students’ probability of self-correcting miscues—or P(SC)—and their EF profile in a cohort of 143 participants (aged 9–15) who represented a diverse spectrum of reading abilities. Moreover, we used experimentally-manipulated passages (decoding, vocabulary, syntax, and cohesion) and employed a fully cross-classified mixed-effects multilevel regression strategy to evaluate the interplay between components of ORF, EF, and text complexity. Our results revealed that, after controlling for reading and language abilities, increased production of miscues across different passage conditions was explained by worse EF. We also found that students with better EF exhibited greater P(SC) when reading complex texts. While text complexity taxes students’ EF and influences their production of miscues, findings suggest that EF may be interactively recruited to restore ORF via self-correcting oral reading errors. Overall, our results suggest that domain-general processes (e.g., EF) are associated with production of miscues and may underlie students’ behavior of self-corrections, especially when reading complex texts. Further understanding of the relation between different components of ORF and cognitive processes may inform intervention strategies to improve reading proficiency and overall academic performance.

Introduction

One determinant of skilled reading is oral reading fluency, which, as decades of research has shown, involves an interactive orchestration of multiple reading- and language-related processes (Allington, 1983a, b; Fuchs, Fuchs, Hosp, & Jenkins, 2001; Perfetti & Roth, 1981; Pikulski & Chard, 2005; Stanovich, 1980; Wolf & Katzir-Cohen, 2001). Oral reading fluency (ORF) is typically measured as words read per minute (WPM) or words read correctly per minute (WCPM), taking into account both rate and accuracy (Fuchs, Fuchs, Hamlett, Walz, & Germann, 1993; Good, Simmons, & Kame’euni, 2001; Hasbrouck & Tindal, 1992; Marston, Fuchs, & Deno, 1985). Research highlights ORF as a multi-componential construct (Breznitz, 2006; Frederiksen, 1981a, b; Hudson, Pullen, Lane, & Torgesen, 2008; Klauda & Guthrie, 2008; Stanovich, 1980; Rumelhart, 1985). Stanovich (1980), for instance, posited that ORF taps into syntactic and semantic processes as readers incorporate supporting context (see also Hudson et al., 2008). These processes can, in turn, interact with lower-level ones (e.g., decoding) to allow readers to interpret text as individual words within meaningful sentences and in larger contexts (Klauda & Guthrie, 2008). This complex orchestration across reading- and language-related domain-specific skills has also been hypothesized to recruit support from cognitive domain-general processes beyond those within the linguistic realm: executive functions (EF). The current study seeks to build upon these findings by examining the interplay between text complexity and EF in relation to more nuanced components of ORF—namely, production of miscues and the probability of self-correcting them.

Executive functions (EF) encompass an array of domain-general abilities that include, but are not limited to: working memory, cognitive flexibility, planning/organization, and reasoning (Diamond, 2013; Miyake & Friedman, 2012; Miyake, Friedman, Emerson, Witzki, & Howerter, 2000; St. Clair-Thompson & Gathercole, 2006; Stuss & Alexander, 2000). Studies have highlighted the importance of EF in reading comprehension (Cutting, Materek, Cole, Levine, & Mahone, 2009; Sesma, Mahone, Levine, Eason, & Cutting, 2009) and word recognition and decoding (Altemeier, Abbott, & Berninger, 2008; Christopher et al., 2012; see also Locascio, Mahone, Eason, & Cutting, 2010), but less so in ORF (Miller et al., 2014; Spencer et al., 2018; see also Cirino et al., 2019). Since ORF acts as a proxy for online linguistic comprehension and involves decoding ability (Fuchs et al., 2001; Jenkins, Fuchs, van den Broek, Espin, & Deno, 2003a, b), it can be inferred that EF might contribute to skilled ORF. Put another way, EF likely supports readers in coordinating a range of reading- and language-related skills as they read aloud and comprehend texts (Frederiksen, 1981a, 1981b; Stanovich, 1980; Roberts & Pennington, 1996).

EF and components of ORF

EF has been proposed, and shown to some extent, to harmonize fluent reading by supporting the coordination of a range of reading- and language-related processes (Breznitz, 2006; Cutting et al., 2009; Perfetti, 1985; Wood, Flowers, & Grigorenko, 2001). One speculation as to why EF contributes to poorer reading performance is that poor readers face challenges in executive coordination across word-, sentence-, and passage-level processes (Fuchs et al., 2001; Klauda & Guthrie, 2008; see also Clay, 1968; Deutsch & Bentin, 1996). When word-level skills such as sight word recognition and decoding are automatized, readers are able to attend to higher-level factors of the written text (sentence and passage), as well as dedicate their cognitive resources (e.g., working memory and the ability to plan and organize) toward online linguistic comprehension (Altemeier et al., 2008; Cutting et al., 2009; Sesma et al., 2009). Via reasoning, individuals comprehend and draw inferences, as well as predict incoming information (e.g., words), based on the cues that are drawn from prior text and stored within working memory (Cain, Oakhill, & Bryant, 2004; Duffy et al., 1987; Gerrig & O’Brien, 2005; Thorndike, 1971). Yet, in order to achieve proficient comprehension during oral reading, readers must also identify and/or suppress irrelevant idea units (via inhibitory control and monitoring), which are then used to construct a coherent mental representation of the current text (Kendeou, van den Broek, Helder, & Karlsson, 2014; Miller et al., 2014). Such mental representation then informs word-level abilities through lexical retrieval and the spreading of semantic activation, a cognitive process that is often known as contextual facilitation (Stanovich, 1980). Notably, contextual facilitation has been shown to be enhanced for poor readers who tend to rely more the surrounding context to compensate for their insufficient word-level skills (Stanovich & West, 1979a; West & Stanovich, 1978). As such, unskilled readers’ reliance on contextual facilitation may further deprive them of the necessary cognitive resources, including various components of EF, for successful ORF and comprehension.

Dysfluent coordination and multi-componential failure while reading aloud can result in oral reading errors—or miscues. Here oral reading errors are referred to as miscues, which is in line with previous studies and reports on ORF (Biemiller, 1970; Breznitz, 2006; Chinn, Waggoner, Anderson, Schommer, & Wilkinson, 1993; Fuchs et al., 2001; Leslie & Caldwell, 2011; see also Amendum et al., 2018). Miscues are typically defined as any repetitions, omissions, transpositions, additions, or mispronunciations that a reader produces while reading aloud (Leslie & Caldwell, 2011). However, no studies to date have provided empirical evidence for the link between EF and readers’ behavior of self-correcting miscues. Self-correction potentially indicates readers’ comprehension-monitoring efforts, whereby individuals adopt deliberate “goal-directed” reading strategies (e.g., self-correction) to resolve their reading difficulties, or miscues, while reading aloud (Anastasiou & Griva, 2009; Borkowski, Chan, & Muthukrishna, 2000; Bowey, 1986; Vorstius, Radach, Mayer, & Lonigan, 2013; see also Nozari, Dell, & Schwartz, 2011). A handful of studies and reports have argued that reading strategies, such as self-corrections, can reveal the direct interplay between textual and contextual features (such as that of text complexity) and the individuals’ domain-general, goal-directed abilities (e.g., EF) in managing any discrepancies or inconsistencies that may emerge during reading (Baker, 1984; Jenkins et al., 2003a, b; Paris, Wasik, & Turner, 1991; Van Dijk & Kintsch, 1983; see also Afflerbach, Pearson, & Paris, 2008; Nozari & Schwartz, 2012; Postma, 2000). This interplay is especially important given the growing attention in text complexity, which could present more opportunities for readers to not only produce miscues but also self-correct while reading difficult texts.

EF and components of ORF through the lens of text complexity

The relation between ORF and EF is particularly important when reading complex texts (Frederiksen, 1981a, b; Chinn et al., 1993; Mesmer, Cunningham, & Hiebert, 2012; Miller et al., 2014; Spencer et al., 2018; Perfetti, 1985). While the terms text complexity and text difficulty are often used interchangeably, reports distinguish the two based on temporal precedence (Mesmer et al., 2012). Text complexity refers to the specific levels (i.e., word, sentence, and passage) and features (e.g., decodability, semantics and vocabulary, syntax, and cohesion) that can be manipulated (for review, see Amendum et al., 2018). Notably, there are several approaches to manipulate levels of text complexity, ranging from fine-grained (phoneme-graph-eme correspondence) to global (as the cohesiveness of idea units) dimensions. More specifically, word-level manipulations can operate by, for example, increasing the difficulty of decodability and exchanging high-frequency words with low-frequency vocabulary (Compton et al., 2004; Faulkner & Levy, 1994; Hoffman, Roser, Salas, Patterson, & Pennington, 2001). A recent study (Hiebert et al., 2019) found that word location, frequency, as well as concreteness were the strongest predictors of students’ reading performance. Sentence- and passage-level factors—such as syntactic difficulty and cohesiveness of text—can also be manipulated to construct complex texts (Benjamin, 2011; Graesser, McMamara, Cai, Conley, & Pennebaker, et al. 2014; Freebody & Anderson, 1983; Norris & Bruning, 1988). Thus, text complexity is typically treated as an independent or predictor variable (see Mesmer et al., 2012). In contrast, text difficulty refers to the interplay between students’ cognitive and behavioral characteristics and the complex texts.

Text complexity has garnered a lot of attention since the widespread adoption of the Common Core State Standards (CCSS; www.corestandards.org; CCSS Initiative, 2010). The CCSS was created to address the lack of college-readiness skills and requires that students have a mastery of complex texts by the end of 12th grade (CCSS Initiative, 2010; National Center for Education Statistics, 2011). The specificity with which the CCSS treats text complexity represents a change from previous standards, which some see as problematic given the lack of research on how increased text complexity could influence students’ reading outcomes, such as comprehension and ORF (e.g., Best, Rowe, Ozuru, & McNamara, 2005; Compton, Appleton, & Hosp, 2004; Eason, Goldberg, Young, Geist, & Cutting, 2012; Hiebert & Mesmer, 2013).

Prior studies have examined the interplay between text complexity and EF to predict ORF (specifically, as indexed by rate [WPM]) (Miller et al., 2014; Spencer et al., 2018). Miller and colleagues (2014) found that EF, particularly the ability to plan and organize, was involved in ORF across different experimentally-manipulated conditions for text complexity (decodability, vocabulary, syntax, and cohesion) (see also Spencer et al., 2018). Of note, these studies operationalize ORF by rate (WPM) to highlight the potential role of EF in orchestrating across word-, sentence-, and passage-level factors while students read aloud complex texts. Yet, a number of reports (Clay, 1969; Leu, 1982; McGee, Kim, Nelson, & Fried, 2015; Pratt & Urbanowski, 2015; Recht, 1976; see also Vellutino & Scanlon, 2002) suggest that evaluating oral reading errors and difficulty—i.e., as indexed by reader’s production of miscues—could provide a deeper understanding into the multi-componential nature of ORF, especially within the framework of text complexity.

Increasing text complexity may place higher demands on students’ reading and language skills as well as EF (Miller et al., 2014; Spencer et al., 2018; see also Mesmer et al., 2012). As a consequence, reading difficulty—i.e., miscues—likely emerges (for a review, see Leu, 1982). Yet, readers are often able to leverage textual and contextual information from the current text in order to attend to and resolve such miscues—i.e., via self-correction (Chinn et al., 1993; D’Angelo 1981; Frederiksen, 1981a, 1981b; Fuchs et al., 2001; Stanovich, 1980; see also Forbes, Poparad, & McBride, 2004). Self-correction likely involves simultaneous recruitment of domain-specific (reading and language) as well as domain-general (EF) processes when confronted with online reading difficulty; yet, whether this linkage changes as text complexity increases remains elusive.

Study rationale

While studies are beginning to reveal the importance of EF in skilled reading, exactly how EF relates to text complexity when measuring miscues and self-corrections is under-explored. Given the multi-componential nature of ORF, and its complex orchestration of multiple cognitive processes, it is challenging to tease apart the underlying factors. One way to tackle this issue is through the use of carefully manipulated texts (i.e., controlling for text complexity) while examining how readers’ behavioral and cognitive characteristics, such as reading- and language-related and EF skills predict miscues and self-corrections. While there are a handful of studies on the relation between readers’ behavior of self-correcting a miscue and reading- and language-related skills (e.g., Bowey, 1986; Deutsch & Bentin, 1996; Stanovich & West, 1979b; see also Rasinski, 1985, 2003), less is known about the involvement of EF in miscues and self-corrections overall. Thus, the current study aims to address this gap in the understanding of the production of miscues and readers’ behavior of self-correcting them.

Research questions

The current study evaluates the relation between ORF, EF, and text complexity. Specifically, we investigated the following research questions: Does EF predict miscues and self-corrections when reading complex texts, controlling for reading- and language-related skills? If so, which components of EF are related to miscues and self-corrections?

We hypothesized that, regardless of reading- and language-related abilities, students with lower EF skills would produce more miscues when reading texts of increased complexity. Furthermore, we hypothesized that higher EF skills would be associated with a greater probability of self-correcting their miscues across different passages manipulated for text complexity.

To answer these questions, we used passages that were experimentally manipulated on four different text complexity features: (1) decodability (word-level); (2) vocabulary (word-level); (3) syntax (sentence-level); and (4) cohesion (passage-level). This approach allowed us to examine how students’ behavioral and cognitive characteristics interact with these text factors and result in the production of miscues and behavior of self-corrections. To control for variability across passages manipulated for text complexity, we applied a fully cross-classified multilevel regression modeling strategy. We used an array of EF measures to capture working memory, planning/organization, cognitive flexibility, and reasoning (nonverbal) skills.

Methods

Participants

The current study was approved by—and the procedures were carried out in accordance with Vanderbilt University’s Institutional Review Board. Participants were recruited from local schools, clinics, and pediatrician’s offices as well as the greater Nashville, TN. Children were excluded if they met the following criteria: (1) known uncorrectable visual impairment; (2) treatment of any psychiatric disorder (other than ADHD) with psychotropic medications, and children with ADHD who were treated with medications other than stimulants were excluded; (3) history of known neurologic disorder (e.g., epilepsy, spina bifida, cerebral palsy, traumatic brain injury); (4) documented hearing impairment greater than or equal to a 25 dB loss in either ear; (5) individuals known to have full-scale IQs below 80, or a score below 70 on either Performance or Verbal scales of the Wechsler Abbreviated Scales of Intelligence (4th Edition; Wechsler, 1999); (6) history or presence of a pervasive developmental disorder; and (7), if during testing, parental responses from the Diagnostic Interview for Children and Adolescents (Version IV; DICA-IV; Reich et al., 2000) indicated the presence of any severe psychiatric diagnoses, including major depression, bipolar disorders, and conduct disorder. The final sample included 143 native English-speaking participants between nine and 15 years old. The sample represented a normal distribution of reading ability, ranging from those with reading difficulties to typically-developing individuals. Participants were tested with a large battery of behavioral and cognitive assessments over two days by trained graduate-level research assistants and/or staff members. Participants’ childhood socioeconomic status (SES) was estimated using the Hollingshead Four-Factor Index (Hollingshead, 1975), which created a social status composite score from parental (highest) educational attainment, occupational position, and marital status. Participants’ age, sex, and SES were treated as control variables for demographic information in the regressions.

Experimental passages

Nine baseline texts about science and animals were initially created by doctoral-level lab members. Then, eight additional texts were constructed by manipulating the passages in terms of their decodability, vocabulary, syntax, and cohesion. In total, 17 unique expository texts were used in the experiment, where each passage had approximately 300 words (see “Appendices 1, 2, and 3” for examples for the baseline versus manipulated texts). All of the passages were piloted on adults.

Each participant read only nine of the 17 passages: one common text, four baseline texts, and four manipulated texts (one from each dimension). The order of administration of these passages was counter-balanced across 10 lists (see “Appendix 2”). For each text, the number of words correctly read per minute (WCPM) was computed by subtracting the number of miscues (See Miscues and Self-Corrections below) from the number of words in that text, then dividing by the seconds that the participants took to read each text.

Coh-Metrix indices (Graesser, McNamara, Louwerse, & Cai, 2004) were used with a 90% equivalency confidence interval (CI) to ensure that each passage was manipulated only for the variable of interest while the other parameters remained constant. For example, two passages for vocabulary were manipulated by replacing high-frequency words with low-frequency words. Then these passages were checked to make sure the pre- versus post-manipulation values for all other relevant variables (i.e., decodability, syntax, and cohesion) remained equivalent. Although Coh-Metrix has indicators that measure readability of passages (CELEX word frequency), it does not analyze texts for word-level decodability. Therefore, an in-house developed measure an in-house developed measure, Decoding System Measure (DSyM; Cutting, Saha, & Hasselbring, 2017), was used to measure text decodability after text manipulations as a way to check for the decoding difficulty of the passages. The DSyM evaluates text decodability based on sub-lexical features of words using an additive formula that takes into account letter-sound discrepancies and grapheme-phoneme correspondence frequencies (Cutting, Saha, & Hasselbring, 2017). This web tool results in a decodability score for individual words and a summary score for the passage.

Decoding Manipulation To manipulate passages in terms of decoding, phonetically-regular words were replaced with irregular ones. For example, the word rubbish (which has a more common, short /i/ sound for the letter “i”) was replaced with debris (which has a less common, long /e/ sound for the letter “i”). All decoding-manipulated passages fell at or above the 90th %ile CI, which remained the case when removing the one baseline passage with a decoding score that fell below the lower end of the 90^th %ile CI (i.e., easier to decode).

Vocabulary Manipulation To manipulate passages in terms of vocabulary, high-frequency words were replaced with low-frequency/less-familiar words. For example, store was replaced with the word amass.

Syntax Manipulation More syntactically complex sentences were constructed while ensuring that inferential relations among propositions were maintained. To do this, the order of sentences as well as verb tense were manipulated. For example, the sentence: “Eskimos use snow from a single snowfall when building an igloo.” Was replaced with a more syntactically complex sentence: “The blocks of snow in an igloo consist of snow from a single snowfall.”

Cohesion Manipulation Passages were manipulated to be less cohesive and more inferentially difficult. To do this, referents across sentences were made to be intentionally vague, and causal connections were spaced further (Vidal-Abarca, Martinez, & Gilabert, 2000). Also, active voice was substituted with passive voice. For example, “Then they used a small bit of string for tying the bag shut” was substituted with “Using a small bit of string, it was tied shut.” (see “Appendix 3”).

Miscues and self-corrections

Reading behaviors were recorded by trained research assistants as the participants read the passages. All errors that participants made while reading the passages aloud were considered to be miscues. Leslie and Caldwell (2011)’s miscue definition and error coding protocols were used, which included repetitions, omissions, additions, transpositions, mispronunciations, and word(s) provided by the tester. The total number of miscues was used to calculate WCPM for each passage read. During initial instructions, participants were told that they would be allowed to correct any oral reading errors. However, while reading the passages aloud, participants were not prompted to make any corrections following an error (miscue). Each miscue followed by an unprompted correction was considered a self-correction. We scaled each participant’s total number of self-corrections by their total number of miscues from each passage to create a metric of individual probability of self-corrected miscues—or P(SC).

Executive functions

Reasoning (nonverbal) was estimated based on participants’ deductive, linear logic, using the Analysis-Synthesis subtest subtest of the Woodcock-Johnson Psychoeducational Battery (3rd Version) (WJ-III; Woodcock, McGrew, & Mather, 2001). Participants were asked to solve a series of nonverbal problems using various “keys” or combinations of colored square boxes that made other colors. The “keys” provided the basis for solving the problem across 35 items. Median reliability for this subtest is reported as 0.89. The standard scores for these subtests have M = 100, SD = 15.

Cognitive Flexibility was measured using the Card Sorting subtest of the Delis-Kaplan Executive Function System (DKEFS; Delis et al., 2001). The Card Sorting subtest had two conditions to measure participants’ cognitive flexibility (i.e., set-shifting ability). This subtest has also been purported to tap into participants’ problem-solving skills and inhibitory control (Delis, Kaplan, & Kramer, 2001; Swanson, 2005). Cognitive flexibility (set-shifting ability) overlaps substantially with inhibitory control (Miyake et al., 2000). During the first condition, Free Sorting, the participants were asked to sort cards into two groups in as many different ways as they could. While sorting, the participants were also asked to verbally describe why and how the cards were being sorted. In the second condition, the examiner, rather than the participant, sorted the cards, while the participant was asked to describe how the cards were being sorted. Across both conditions and sets, there were up to 36 sorts. The final score was calculated by the sum of correct sorting from both sets divided by the total card sets. Test–retest reliability for the Card Sorting subtest is reported as 0.49, and internal consistency falls between 0.55 and 0.82, depending on age (see also Spencer, Cho, & Cutting, 2019). The scaled score used in analyses has M = 10, SD = 3.

Planning/Organization was measured with the Elithorn Mazes subtest of the Wechsler Intelligence Scale for Children, 3^rd Edition: Process Instrument (WISCIII; Kaplan, Fein, Kramer, Delis, & Morris, 1999). Participants were asked to examine a visually presented maze and choose a single path that passes through circles within a “lattice” of lines in an inverted triangular structure, without backtracking. The measure provides information about strategic planning and response organization skills. The internal consistency reliability for this test is reported as 0.71. The scaled score has M = 10, SD = 3.

Working Memory was measured using Digit Span task from the Wechsler Intelligence Scale for Children, Fourth Edition (WISC-IV; Wechsler, 2003). In this test, participants must attend to verbal information and then repeat progressively longer series of numbers in backward order for the Digit Span Backward portion. Scaled score for the Backward trial was used in analyses. Reliability coefficient for the Digit Span task is between 0.89 and 0.93, depending on age. The scaled score has M = 10, SD = 3.

Reading and language measures

Decoding was measured with the Woodcock Reading Mastery Test (WRMT-R/NU; Woodcock, 1998). Participants were asked to read aloud single words (Word Identification subtest) and non-words (Word Attack subtest). Split-half reliability coefficients for Word Identification and Word Attack are reported as 0.91–0.99 and 0.84–0.97, respectively. The standardized scores from both of these subtests were used to determine the participants’ basic reading (i.e., decoding ability) composite score. Split-half reliability coefficient for the composite score falls within 0.91–0.98. The standard scores have M = 100, SD = 15.

Vocabulary was measured with Receptive and Expressive Vocabulary and Synonyms subtests from the Test of Word Knowledge (TOWK; Wiig & Secord, 1989). For the Receptive Vocabulary subtest, participants were asked to select one out of a set of pictures that would best match an orally-presented word. The opposite is true for the Expressive Vocabulary subtest, where the participants used one word to describe a picture. Finally, the Synonyms subtest asked the participants to select one out of four printed words that would best match the cued word. Test–retest relaibility for the Receptive, Expressive, and Synonyms subtests are 0.89, 0.92, and 0.93, respectively. A composite score across the three subtests were derived for any further analyses. The scaled scores have M = 10, SD = 3.

Morphology was measured using the Test of Morphological Relatedness (TMR; adapted from Mahony, Singson, & Mann, 2000). The TMR consisted of 12 items in writing. For each item, participants were asked to indicate whether the two words are related (e.g., “happy—happiness” [YES]; “cat—catalogue” [NO]). For each correct answer, the participant received one point. Test–retest reliability for the Test of Morphological Related is reported as 0.62. The raw score has a possible range 1–12.

Statistical analyses

Our final sample was comprised of 143 participants. First, pairwise correlations were run for covariates and predictors of interest to confirm the relation among participants’ reading- and language-related abilities and EF. Demographic factors (e.g., sex, age, and SES) were also included in these correlations. Prior to answering central questions of interest with manipulated texts as covariates, we first confirmed whether EF predicted components of ORF—namely, production of miscues and P(SC)—in passages without any manipulations (i.e., baseline passages only). To confirm as well as address the relation between the production of miscues, text complexity, and EF, we built a series of fully cross-classified mixed-effects multilevel regression models. To accommodate this approach, we used the lmer function in the R suite (version 3.4.0) (Baayen, Davidson, & Bates, 2008). We applied this regression strategy since the covariates of interest (EF, and reading- and language-related abilities) were nested within individual participants and the corresponding passages. As such, individual participants were treated as a random-effect factor, along with passage topics. Randomized group assignment for the order of passage administration was also included as a random-effect factor to avoid any potential confounding and/or biased order effects. All variables were normalized, except for age (mean-centered) and sex (dummy-coded; 0 = female, 1 = male). Before evaluating the contributions of EF and text complexity to miscues and self-corrections, we built these regression models to evaluate the links between EF and WCPM for baseline passages only and in relation to text complexity with experimentally-manipulated passages.

To build the regression model for predicting the production of miscues, we used a step-wise approach. In Step 1, Passage Covariates (i.e., baseline and manipulated passages) were entered, while controlling for Demographics. Next, Executive Functioning was added in Step 2. Finally, Reading and Language covariates were added in Step 3 to confirm whether the relative contributions of text complexity and EF would remain significant for production of miscues.

Finally, to answer our central question of interest, regarding the relation between P(SC), text complexity, and EF, another regression model was built similarly using the step-wise approach. Passage Covariates were entered first (Step 1; while controlling for Demographics), followed by Executive Functioning (Step 2). Lastly, Reading and Language covariates were entered last (Step 3) to evaluate whether relative contributions of text complexity and EF would still hold for P(SC) variance.

Results

The descriptive statistics and the results of pairwise correlations across participants’ reading- and language-related abilities, EF, and demographic factors are presented in Tables 1 and 2, respectively. The descriptive statistics for two components of ORF—namely, miscues and P(SC), as well as WCPM—across passages are presented in Table 3.

Table 1.

Descriptive statistics across covariates

Covariates	Mean	SD	Min	Max
Reading and language
Decoding	102.2	14.42	60	136
Vocabulary	10.65	3.024	1.1	17
Morphology	9.655	3.951	3	12
Executive functioning
Reasoning	108.1	14.41	52	151
Cognitive flexibility	9.628	2.756	1	16
Planning/organization	10.67	3.138	3	18
Working memory	7.247	2.207	3	14
Demographics
Sex	55% female;	45% male
Age	11.65	1.38	9	14.83
Socioeconomic status	49.59	7.65	26	62

Table 2.

Correlations between EF, reading and language measures, and demographic factors

		1	2	3	4	5	6	7	8	9	10
1	Reasoning
2	Cognitive flexibility	0.393	-
3	Planning/organization	0.188	0.370	-
4	Working memory	0.343	0.385	0.300	-
5	Decoding	0.459	0.441	0.349	0.530	-
6	Vocabulary	0.519	0.430	0.226	0.345	0.657	-
7	Morphology	0.335	0.249	0.163	0.242	0.384	0.312	-
8	Sex	0.202	0.157	0.072	0.160	0.165	0.174	0.099	-
9	Age	−0.045	-0.071	−0.022	0.232	−0.002	-0.218	0.005	0.082
10	Socioeconomic status	0.313	0.273	0.153	0.189	0.264	0.322	0.203	0.004	−0.084	-

Correlations significant at the p < 0.05 level are bolded

Table 3.

Descriptive statistics for passage covariates across components of ORF

Passage covariates	Miscues		SC		P(SC)		WCPM
	Mean	SD	Mean	SD	Mean	SD	Mean	SD
Baseline	16.91	13.49	3.32	2.51	0.23	0.15	141.2	60.18
Decoding
Baseline	20.73	17.51	3.75	3.16	0.24	0.19	134.7	50.84
Manipulated	23.20	18.13	3.49	2.98	0.19	0.17	123.3	49.57
Vocabulary
Baseline	15.14	12.49	1.95	1.69	0.16	0.14	152.1	64.50
Manipulated	17.25	14.28	1.99	1.95	0.12	0.12	135.6	57.97
Syntax
Baseline	13.39	11.98	2.91	2.36	0.28	0.22	150.7	56.38
Manipulated	17.43	14.57	3.66	2.71	0.25	0.17	133.8	50.79
Cohesion
Baseline	15.24	14.29	3.43	3.00	0.27	0.22	133.5	52.76
Manipulated	18.04	16.29	3.51	3.46	0.25	0.18	132.3	53.85

Our results indicated that, for baseline passages only, WCPM was significantly predicted by students’ ability to plan and organize (B = 0.148, se = 0.072) and working memory (B = 0.171, se = 0.065) (p < 0.05) (“Appendix 4”). When taking into account different passage manipulations, lower WCPM was associated with reading passages with more difficult-to-decode words (B = − 0.291, se = 0.114), higher corpus low-frequency words (B = − 0.423, se = 0.099), and that were more syntactically complex (B = − 0.284, se = 0.110) (p < 0.05), but not for the cohesion condition (B = − 0.029, se = 0.110) (p > 0.05).

Executive functions predicting production of miscues and P(SC)

Prior to answering our central questions of interest, we examined whether EF played a role in students’ production of miscues and behavior of self-correcting them—as indexed by P(SC). Our regression results suggested that, after taking into account reading- and language-related abilities and demographic factors, having poorer EF explained students’ production of more miscues (Table 4a). Specifically, students with worse reasoning (B = − 0.165, se = 0.062), cognitive flexibility (B = − 0.222, se = 0.080), ability to plan and organize (B = − 0.156, se = 0.072), and working memory (B = − 0.265, se = 0.075) are more likely to produce miscues while reading aloud (p < 0.05). Yet, with better EF, particularly reasoning (B = 0.117, se = 0.057), cognitive flexibility (B = 0.163, se = 0.060), and working memory (B = 0.165, se = 0.057), students self-corrected their miscues more often (p < 0.05)—that is, they had greater P(SC) (Table 4b).

Table 4.

Summary of regression results predicting (a) production of miscues and (b) P(SC) from EF and reading and language skills for baseline passages only

	(a) Miscues		(b) P (SC)
	B	se	B	se
Fixed effects
Intercept	−0.012	0.099	0.030	0.083
+ Executive functioning
Reasoning	−0.165*	0.062	0.117*	0.057
Cognitive flexibility	−0.222**	0.080	0.163**	0.060
Planning/organization	−0.156*	0.072	0.072	0.061
Working memory	−0.265**	0.075	0.165**	0.057
+ Reading + language
Decoding	−0.553**	0.086	0.308**	0.076
Vocabulary	−0.145†	0.082	0.018	0.073
Morphology	−0.119†	0.062	0.094†	0.054
+ Demographics
Sex	0.090	0.107	−0.099	0.094
Age	−0.212**	0.046	0.130**	0.042
Socioeconomic status	0.070	0.056	0.008	0.050
Random effects	σ	SD	σ	SD
Participant	0.337	0.581	0.103	0.322
Topic	0.032	0.180	0.021	0.145
Group	0.000	0.000	0.000	0.000
Residual	0.185	0.430	0.737	0.858

^†p < 0.10,

^*p < 0.05,

^**p < 0.01

Text complexity and executive functions predicting miscue production

As hypothesized, our regression results suggested an association between reading more complex texts and increased production of miscues. That is, students produced significantly more miscues when reading passages manipulated for decoding (B = 0.209, se = 0.047), vocabulary (B = 0.112, se = 0.044), syntax (B = 0.137, se = 0.049), and cohesion (B = 0.159, se = 0.049), as compared to their baseline conditions (p < 0.05) (Steps 1–3, Table 5). When adding EF as covariates into our model, worse cognitive flexibility (B = − 0.171, se = 0.079), ability to plan and organize (B = − 0.155, se = 0.078), and working memory (B = − 0.254, se = 0.073) explained higher production of miscues (p < 0.05) (Step 2, Table 5). After accounting for the significant roles of reading- and language-related abilities, only cognitive flexibility (B = − 0.073, se = 0.069; p > 0.05) was no longer predictive of students’ production of miscues (Step 3, Table 5).

Table 5.

Summary of regression results predicting production of miscues

	Step 1		Step 2		Step 3
	B	se	B	se	B	se
Fixed effects
Intercept	−0.192	0.128	−0.141	0.114	−0.162	0.113
+ Passage covariates
Decoding	0.209**	0.047	0.209**	0.047	0.208**	0.047
Vocabulary	0.112*	0.044	0.111*	0.044	0.112*	0.044
Syntax	0.137**	0.049	0.137**	0.049	0.137**	0.049
Cohesion	0.159**	0.049	0.159**	0.049	0.159**	0.049
+Executive functioning
Reasoning			−0.095	0.073	−0.050	0.074
Cognitive flexibility			−0.171*	0.079	−0.073	0.069
Planning/organization			−0.155*	0.078	−0.173*	0.075
Working memory			−0.254**	0.073	−0.235**	0.072
+ Reading + language
Decoding					−0.565**	0.087
Vocabulary					−0.225*	0.093
Morphology					−0.136*	0.063
+ Demographics
Sex	0.042	0.145	0.035	0.129	0.107	0.106
Age	−0.199**	0.051	−0.187**	0.054	−0.190**	0.047
Socioeconomic status	−0.127	0.072	0.027	0.069	0.048	0.056
Random effects	σ	sd	σ	sd	σ	sd
Participant	0.676	0.822	0.589	0.747	0.529	0.727
Topic	0.025	0.159	0.025	0.159	0.025	0.148
Group	0.016	0.127	0.012	0.108	0.004	0.063
Residual	0.174	0.417	0.174	0.417	0.174	0.417

^*p < 0.05,

^**p < 0.01

Text complexity and executive functions predicting P(SC)

To address the relation between text complexity, EF, and students’ behavior of self-correcting miscues, we again employed a series of multilevel cross-classified mixed-effects regression models. Similar to our miscue findings, our results revealed that readers were less likely to self-correct their miscues when reading experimentally-manipulated passages, as compared to baseline conditions (Steps 1–3, Table 6). This was true for passages manipulated for decoding (B = − 0.264, se = 0.064), vocabulary (B = − 0.218, se = 0.063), and syntax (B = − 0.139, se = 0.066) (p < 0.05), but less so for cohesion (B = − 0.068, se = 0.065; p > 0.05). When adding EF as covariates in our regression model, better cognitive flexibility (B = 0.184, se = 0.057) and working memory (B = 0.159, se = 0.052) explained greater P(SC) (p < 0.05) (Step 2, Table 6). Of note, the predictive effect of reasoning (B = 0.090, se = 0.053) was approaching significance (p < 0.10). In other words, students with better EF were more likely to self-correct their miscues when reading more difficult texts. These relations between components of EF and P(SC) during complex texts remained significant even after controlling for reading- and language-related skills (Step 3, Table 6). Notably, the predictive effect of reasoning became significant (B = 0.104, se = 0.057; p < 0.05).

Table 6.

Summary of regression results predicting probability of self-correcting miscues [P(SC)]

	Step 1		Step 2		Step 3
	B	se	B	se	B	se
Fixed effects
Intercept	0.304**	0.113	0.277**	0.107	0.310**	0.106
+ Passage covariates
Decoding	−0.264**	0.064	−0.264**	0.064	−0.261**	0.064
Vocabulary	−0.218**	0.063	−0.216**	0.063	−0.224**	0.063
Syntax	−0.139*	0.066	−0.142*	0.066	−0.136*	0.066
Cohesion	−0.068	0.065	−0.071	0.065	−0.067	0.065
+ Executive functioning
Reasoning			0.090†	0.053	0.104*	0.057
Cognitive flexibility			0.184**	0.057	0.161**	0.055
Planning/organization			0.060	0.057	0.088†	0.051
Working memory			0.159**	0.052	0.121*	0.050
+ Reading + language
Decoding					0.402**	0.067
Vocabulary					0.105	0.072
Morphology					0.057	0.048
+ Demographics
Sex	−0.124	0.102	−0.107	0.096	−0.125	0.080
Age	0.125**	0.039	0.084*	0.040	0.088*	0.038
Socioeconomic status	0.130**	0.049	0.048	0.050	0.035	0.043
Random effects	σ	sd	σ	sd	σ	sd
Participant	0.218	0.467	0.172	0.415	0.082	0.287
Topic	0.000	0.000	0.000	0.000	0.000	0.000
Group	0.005	0.073	0.000	0.000	0.008	0.092
Residual	0.722	0.850	0.721	0.849	0.724	0.851

^†p < 0.10,

^*p < 0.05,

^**p < 0.01

Interaction between text complexity and executive functions

Finally, while there is not sufficient power to detect interaction effects in our models, we conducted exploratory “proof-of-concept” analyses to examine whether the interaction between text complexity and EF is predictive of students’ production of miscues and P(SC). In other words, the following results should be interpreted with caution. Different manipulations of the passages (decoding, vocabulary, syntax, and cohesion) relative to their baselines were allowed to interact with individual components of EF, including reasoning, cognitive flexibility, planning/organization, and working memory. In terms of production of miscues, only the interaction between decoding-manipulated passages and cognitive flexibility was significant (B = − 0.093, se = 0.040; p < 0.05). That is, for passages with more difficult-to-decode words, students at one standard deviation below the mean cognitive flexibility score would likely produce more miscues. In terms of P(SC), the interactions that were significant at p < 0.05 were between decoding-manipulated passages and reasoning (B = 0.155, se = 0.077), between syntax-manipulated passages and working memory (B =0.146, se = 0.074), and between cohesion-manipulated passages and reasoning (B = 0.175, se = 0.079) and working memory (B = 0.149, se = 0.074). Specifically, students at one standard deviation higher than the mean reasoning and/or working memory would likely self-correct their miscues when reading more difficult texts, including cohesion-manipulated passages (i.e., with less cohesiveness across idea units and more difficult to draw inferences).

Discussion

The overarching aim of the current study was to examine the role of EF in students’ production of miscues and behavior of self-correcting them while reading baseline versus experimentally-manipulated passages for text complexity. Our results revealed that production of miscues was associated with reading more complex texts (experimentally-manipulated passages for decoding, vocabulary, syntax, and cohesion). In addition, components of EF, particularly planning/organization and working memory, contributed significant variance in production of miscues beyond reading- and language-related skills. Central to our study were the findings for students’ behavior of self-correcting their miscues, as indexed by probability of self-corrected miscues [P(SC)]. Decreased P(SC) was related to reading more complex texts experimentally manipulated for decoding, vocabulary, and syntax but not cohesion. Higher P(SC) was predicted by better EF, including reasoning, cognitive flexibility, and working memory, even after controlling for reading- and language-related abilities. To our knowledge, this is the first study to explicitly evaluate whether EF explained students’ ability to self-correct their miscues using an experimental paradigm of increasing text complexity. Overall, our results elucidate the importance and involvement of EF in the multi-componential nature of ORF, while highlighting students’ handle on contextual and textual discrepancies (i.e., as indexed by miscues) through self-correction. First, we address the results regarding students’ production of miscues in relation to EF and text complexity. Then, and central to our study, we discuss the unique association between students’ behavior of self-correcting miscues and EF while controlling for text complexity. Limitations as well as future directions follow.

Do executive functions predict the production of miscues when students read texts with varying complexity?

Production of miscues and executive functions

While the link between EF and the production of miscues has been hypothesized throughout previous literature (e.g., Frederiksen, 1981a, 1981b; Stanovich, 1980, 1986), no studies to date have lent empirical support for this assertion. Our results suggest that students with poorer EF are likely to produce more miscues while reading aloud, even after taking into account reading- and language-related skills. Poor readers rely on cues drawn from sentence- and passage-level contents (i.e., contextual facilitation) to in part buffer their difficulties with word-level skills (word recognition) and processing (e.g., more difficult-to-decode words) (Jenkins et al., 2003b; LaBerge & Samuels, 1974; Nation & Snowling, 1998; Stanovich, 1980). As a result, miscues emerge, and students’ ORF performance worsens. Alternatively, with efficient executive coordination, students are able to orchestrate across word-, sentence-, and passage-level reading- and language-related processes to maintain the textual and contextual flow during ORF (Klauda & Guthrie, 2008; see also Cutting et al., 2009; Kim, 2015; Kuhn, Schwanenflugel, & Meisinger, 2010). Our results reveal the importance of EF in students’ attempt to minimize the production of miscues during ORF.

Our results align with previous studies, which demonstrate that deficits with specific EF components may be related to poor ORF and overall reading competence. Since our participants are between the ages of 9 and 15, they have entered the “reading to learn” phase and are expected to have proficient ORF and its reading- and language-supporting skills (e.g., decoding; see Wang, Sabatini, O’Reilly, & Weeks, 2018). Studies suggest that working memory plays an important role in developing proficient reading fluency—specifically, during the “learning to read” phase (Nevo & Breznitz, 2011, 2013; Swanson & Jerman, 2007; Wilcutt, Pennington, Olson, Chabildas, & Hulslander, 2005). Working memory has been clearly demonstrated to be important well beyond the “learning to read” and into the “reading to learn” phases. In other words, having poor EF may also hinder students’ proficiency with reading- and language-related abilities needed for skilled ORF (Locasio et al. 2010; Sesma et al., 2009; see also Daneman, 1991). Overall, our results reveal that oral reading dysfluency, as indexed by more miscues, is related to less effective EF, beyond reading- and language-related difficulties.

Production of miscues and text complexity

Reading errors (miscues) emerge in part as a product of reading difficulty (for a summary of multiple sources of dysfluency, see Wolf & Katzir-Cohen, 2001; Meyer & Felton, 1999). While the linkage between multiple sources of dysfluency is likely true for when students read complex texts, only a few studies have substantiated this claim with empirical evidence. Chinn and colleagues (1993) examined the production of different types of miscues, including substitutions, non-words, hesitations, and insertions or omissions, using text-leveling paradigm (see also Compton et al., 2004; Leu, 1982). Overall, increased count in oral reading errors was observed when students were reading passages and stories that were above their reading level (more difficult). Interestingly, Chinn and colleagues (1993) found that poor readers made more grapho-phonemic miscues, which could be explained by their lower decoding ability. In contrast, skilled readers were more prone to make semantically unacceptable miscues than grapho-phonemic ones (see also D’Angelo, 1981).

In the current study, all of the four manipulations (decoding, vocabulary, syntax, and cohesion) significantly increased students’ production of miscues. While we did not specifically isolate and categorize different types of miscues (Chinn et al., 1993; D’Angelo, 1981; Frederiksen, 1981a, 1981b), manipulations of different text dimensions and factors in the current study could potentially be perceived as a proxy for characterizing and tapping into the nature of the miscues (e.g., grapho-phonemic, syntactic, and semantic acceptability) (see also Cattell, 1886; Doehring, 1976). For example, findings related to the decoding-manipulated condition provided insights about the linkage between miscues and word-level factors (i.e., grapho-phonemic) (Rasinski, 2003; Schwanenflugel, Hamilton, Wisenbaker, Kuhn, & Stahl, 2009; see also Chinn et al., 1993; D’Angelo, 1981). Furthermore, the production of syntax-related miscues may be implicated in our findings using the syntax-manipulated passages (Cattell, 1886; Doehring, 1976; Schreiber, 1980). Lastly, semantically constrained miscues are likely associated with the significant effects of vocabulary- and cohesion-manipulated conditions (Smith & Elkins, 1985). An additional explanation is that students’ production of miscues may occur across word-, syntax-/sentence-, and passage-level textual and contextual factors (Biemiller, 1970; Klauda & Guthrie, 2008; Stanovich, 1980; Weber, 1970), as evidenced by the significant effects across all four manipulations. Regardless of manipulations, text complexity increased students’ production of miscues and in part hindered their ORF and online comprehension.

Production of miscues, text complexity, and executive functions

We hypothesized that reading texts that varied in their complexity may tax students’ EF and thereby increase the production of miscues. While past literature suggested that poor EF is linked to poor reading performance by examining the role of working memory (Peng et al., 2017), the current study employed an array of assessments to capture the multiple components of students’ EF (Diamond, 2013; Miyake et al., 2000; Miyake & Friedman, 2012). As expected, different components of EF were found to predict students’ production of miscues while reading experimentally-manipulated texts. Specifically, students with worse cognitive flexibility, ability to plan and organize, and working memory were likely to produce more miscues. After controlling for reading- and language-related abilities, the relative contributions of planning/organization and working memory remained significant. While the large variance explained by reading- and language-related skills is not surprising (Stanovich & West, 1979a, b), what is less known is the link between EF and contextual facilitation. Contextual facilitation arguably taxes individuals’ attentional resources (working memory and the ability to plan/organize) in order to leverage sentence- and passage-level contents to guide word-level processing (Brothers, Swaab, & Traxler, 2015; Ehrlich & Rayner, 1981; Stanovich & West, 1979a, b; West & Stanovich, 1982). This also means that contextual facilitation would deprive the cognitive resources needed to attend to more demanding, higher-level processes (language comprehension). In other words, contextual facilitation (and reading- and language-related processes) as well as EF may play competing roles in and production of miscues (Stanovich, Nathan, & Zolman, 1988; Stanovich & West, 1979a).

The relation between production of miscues and specific components of EF reported here, after controlling for reading- and language-related skills, is not altogether surprising. The capacity of students’ working memory and planning/organizing is likely taxed by more difficult-to-decode words and/or reading texts with more low-frequency vocabulary words (Daneman, 1991; Ellis & Sinclair, 1996; Swanson & Jermon, 2007). EF is also likely recruited to support students in constructing syntactically and semantically coherent sentences (Hudson et al., 2008; Miller et al., 2014; Spencer et al., 2018; Rasinski, 1985, 2003). Working memory may allow students to hold the content of sentences in their memory, which includes mentally registering words and content ideas (Daneman, 1991; Daneman & Carpenter, 1980; King & Just, 1991; MacDonald, Just, & Carpenter, 1992; Vos, Gunter, Kolk, & Mulder, 2001). The content of sentences— and potentially of the overall passage—stored within students’ working memory is thought to aid in predicting what information (e.g., words) comes next. With poorer working memory, students may be less able to bridge and map the contents drawn from prior text with incoming information. This role of working memory likely holds while reading more textually and contextually complex passages (e.g., syntax- and cohesion-manipulated ones) (Long, Oppy, & Seely, 1997; Oakhill, Hart, & Samols, 2005). Moreover, planning/organizing is likely associated with students’ ability to monitor overall textual and contextual cohesion and, at the same time, process sentences (Locascio et al., 2010). Impairments with planning and organizing contents extracted from the written text may also tax working memory, which could otherwise underlie contextual facilitation in buffering students’ word-level deficits (Jenkins et al., 2003b; Stanovich, 1980). Overall, findings from the current study suggest that the interactive, and potentially compensatory, recruitment of EF when students read difficult texts is an area in need of further exploration.

When reading complex texts, can students’ behavior of self-correcting miscues be explained by executive functions?

The orchestration of multiple reading- and language-related skills during ORF is undoubtedly complex and laborious. This is especially true for poor readers, whereby the production of miscues reflects their difficulties with cognitive orchestration while reading aloud. In contrast, self-corrections are suggested to be a cognitive and behavioral strategy that permits students to resolve miscues, alleviate reading difficulty, and potentially restore the semantic coherence of the written text (Clay, 1969; Fuchs et al., 2001; Weber, 1970; see also Allington, 1983a; Tunmer & Greaney, 2010). The ability to self-correct likely reflects the multi-componential nature of ORF, whereby reassessing and resolving miscues may require an executive (re-)coordination of multiple cognitive abilities (see McGee et al., 2015; Stanovich, 1980). To this end, the current study posited that students additionally recruit EF to self-correct their miscues, beyond their reading- and language-related skills, during ORF.

Behavior of self-correcting miscues and executive functions

While having poorer EF in part explained the production of miscues, self-correcting such miscues was associated with better EF skills. Specifically, students with better working memory and cognitive flexibility exhibited higher probability of self-correcting miscues, as indexed by P(SC). This relation held true even after controlling for students’ decoding, vocabulary, and morphology abilities. Our findings are consistent with prior literature showing a relation between stronger word-level skills (e.g., decoding) and students’ behavior of self-correct oral reading errors more frequently (Bowey, 1986; D’Angelo, 1981; Chinn et al.; Stanovich, 1986; Tunmer & Nicholson, 2011; Weber, 1970; see also Johnston & Goatley, 2014). Unique to our findings is the contribution of EF to self-corrections even after accounting for basic reading skills. In addition to contributing to the growing literature on the role of EF in reading performance, our study confirms the relation between the ability to self-correct miscues and domain-specific and domain-general cognitive abilities that was previously hypothesized, but not empirically tested until now (Allington, 1983b; McNaughton & Glynn, 1981). EF may play an interactive-compensatory role in ORF, specifically through the ability to self-correct miscues.

Behavior of self-correcting miscues and text complexity

Reading complex texts has been previously shown to negatively impact students’ oral reading performance, but less is known about readers’ self-corrections. We found that students’ probability of self-correcting their miscues was significantly reduced when reading more complex passages. Previous studies found that the miscues that the readers self-corrected had word-level differences (e.g., grapho-phonemic and semantic) when compared to those that were not self-corrected (Chinn et al., 1993; Frederiksen, 1981a, 1981b). Our findings highlight the importance of word-level factors in self-corrections, as the decoding- and vocabulary-manipulated passages to a higher probability of self-corrections. Previous reports also suggest that readers with poor word-level abilities are more likely rely on contextual facilitation while reading aloud (Chinn et al., 1993; D’Angelo, 1981; Frederiksen, 1981a, 1981b). As demonstrated by our findings for the syntax- and, to some extent, cohesion-manipulated passages, such contextual facilitation may be leveraged to inform the resolution of miscues within meaningful phrases and larger contexts (see Olson & Gee, 1988). Readers’ behavior of self-corrections, while arguably at the level of lexical and post-lexical processing, has been suggested to be indicative of their efforts for online comprehension-monitoring strategies (Bowey, 1986). Van der Schoot and colleagues (2011; 2008) found that good readers rely on both “local” (sentence) and “global” (passages) features of the written text to support their reading strategies as well as online comprehension. In contrast, poor readers were reported to not rely on the more “global” processing for online comprehension-monitoring strategies (Oakhill et al., 2005; Vorstius et al., 2013). This also aligns with Stanovich and West (1979a)’s idea of contextual facilitation that is more readily present in poor readers, whereby they are more inclined to attend to the “local” (sentence and syntax) features to buffer their deficits with word-level skills (e.g., decoding) (see also Jenkins et al., 2003a, b). This “local” underpinning of contextual facilitation is likely related to poor readers’ reliance on the syntactic features versus cohesiveness of the passages to support their reading strategies, such as self-corrections (Bloom, Fletcher, van den Broek, Reitz, & Shapiro, 1990; Vortius et al., 2013; Stanovich, 1980). Overall, our results indicate that text complexity not only influences students’ production of miscues, but also their behavior of self-correcting them.

Behavior of self-correcting miscues, text complexity, and executive functions

In addition to examining text complexity and self-corrections, we also evaluated the link between EF and students’ behavior of self-correcting miscues for these manipulated texts. We found that students with better EF skills exhibited higher probability of self-correcting miscues when reading the experimentally-manipulated passages. In particular, significant predictive effects were found for working memory and cognitive flexibility. While reading complex texts taxes students’ cognitive resources and leads to the production of miscues, EF components such as reasoning, working memory, and cognitive flexibility are likely recruited to detect and correct such reading errors. That is, poorer reading- and language-related abilities as well as less effective EF would explain more miscues (Acheson & MacDonald, 2009). EF, such as working memory, may also be involved in detecting such miscues (Pearlmutter & MacDonald, 1995). Textual and contextual processing in reading arguably operates via two mechanisms: (1) fast-acting, automatic spreading of semantic memory; and (2) slow-acting, attentional control that causes facilitation, response inhibition, and switching (Stanovich, 1980; Stanovich & West, 1979a; see also Perfetti & Roth, 1981). In line with this, cognitive flexibility reportedly taps response inhibition, cognitive switching, and initiation of retrieving from lexical and semantic storage (e.g., working memory and reasoning) to support reading fluency (Berninger & Nagy, 2008). Similarly, the mechanistic role of working memory has been hypothesized to coordinate multiple features of the written text. Thorndike (1971) argued that reading could potentially be perceived as synonymous to reasoning, especially in the case of monitoring and detecting inconsistency (e.g., oral reading errors [miscues]) during sentence and passage reading. With better storage of cues drawn from prior text, individuals are more likely to assess what they read versus what are written in the current paragraph and make sense, which may be indicative of better relational reasoning. As such, working memory and/or reasoning both are likely involved in students’ ability to monitor and detect errors—and in some cases, self-correct them. When a miscue is detected, students likely shift their attention toward and resolve such discrepancy, by tapping cognitive flexibility (Cartwright, Marshall, Huemer, & Payne, 2019; Deak, 2003; Goral et al., 2011). This highlights the potentially unique roles of the ability to plan/organize in production of miscues, and of cognitive flexibility in self-corrections. That is, while reading, these specific EF components likely operate in two ways: (1) planning/organizing word-, sentence-, and passage-level cues (e.g., stored within working memory) and harmonizing multiple reading- and language-related processes, thus reducing chances for potential miscues; and/or (2) via cognitive flexibility (e.g., shifting attention) to recruit relevant cognitive skills and self-correct miscues as they arise (for “dual routes of cognitive control,” see Ramscar, Dye, Gustafson, & Klein, 2013, see also Stanovich, 1980).

Overall, it is possible that myriad EF components are involved in students’ self-corrections, which is in line with neurobiologically-based models of dynamic, dissociable executive processes in monitoring, detecting, and detecting errors (Garavan, Ross, Roche, & Stein, 2002). Moreover, our findings for the role of EF in the ability to self-correct miscues highlight the multi-dimensional nature of ORF, and suggest that EF potentially plays a top-down, interactive-compensatory role when students read difficult texts.

Limitations and future directions

While the current study provides insights for the role of EF in students’ production of miscues and behavior of self-corrections within the paradigm of text complexity, there are several limitations worth mentioning. First, our study leveraged advanced, modern statistical modeling strategies (fully cross-classified multilevel mixed-effects regression) to evaluate the association between components of ORF, EF, and text complexity. While these models are robust, future investigations should employ a developmental framework as well as functional methods (e.g., eye-tracking) to hone in the mechanistic roles of EF and move closer to a causal inference. Second, while the current study employed a novel approach of using experimentally-manipulated passage conditions for text complexity, two experimental features deserve attention: (a.) genre-level factor (e.g., narrative versus expository), and (b.) the level of text complexity. In terms of genre-level factor, the current study focused on expository but not narrative texts. This was largely due to previous literature suggesting that, as compared to narrative texts, reading expository texts is more difficult and likely taps into different cognitive skills (Graesser, McNamara, & Louwerse, 2003; Saenz & Fuchs, 2002; see also Aboud, Bailey, Del Tufo, Barquero, & Cutting, 2019; Eason et al., 2012; Swett et al., 2015). Also, the current study aimed to introduce difficult texts by increasing text complexity. Future studies should also examine the role of EF in reading less difficult passages (e.g., via a text-leveling approach; see Compton et al., 2004), or consider a continuum of difficulty as opposed to only easy versus hard levels. Such findings may shed further light on whether EF is recruited uniquely when reading more (versus less) difficult texts. Moreover, experimental passage manipulations at global (e.g., referential cohesion and genre) versus fine-grained (e.g., word imageability and phonological units and patterns) factors may also provide a more comprehensive examination into the link between text complexity and components of ORF (Chinn et al., 1993; Tortorelli, 2019). Finally, the current study aimed to evaluate the interaction between components of ORF, text complexity, and EF. Albeit the insufficient power to detect true significance, our results found a potential interplay between EF and text complexity in P(SC). Future studies should explicitly examine whether the interaction between text complexity and EF in predicting self-corrections varies across reading and language skills. For example, it is possible that low-performing readers may rely on better EF to self-correct while reading more complex texts. Researchers should also consider examining the interaction between manipulated passage conditions and their corresponding reading and language abilities, such as between syntax-manipulated texts and students’ syntactic processing skill. This may capture the unique contributions of students’ EF to their miscues and self-corrections.

Summary

Reading proficiency and overall academic success undoubtedly rely on skilled ORF. Yet, ORF has been shown to be a complex orchestration of multiple reading- and language-related skills. As such, EF has been hypothesized to be key in skilled ORF. To this end, our findings lend empirical support for the potentially interactive-compensatory role of EF in the components of ORF—namely, production of miscues and students’ behavior of self-correcting them. Specifically, while poorer EF is associated with production of more miscues, having better EF is related to a higher probability of self-correcting such reading errors. In addition, given the widespread adoption of the CCSS and the growing attention on the impacts of text complexity on reading outcomes, our results also highlight the relation between EF and components of ORF when students read experimentally-manipulated passages for text complexity. Reading difficult texts likely taxes students’ cognitive resources and negatively influences the reading- and language-related skills needed to support ORF. Yet, our results imply that, with strong EF, students recruit EF to interactively assess reading errors and (re-)integrate reading- and language-supporting skills to maintain fluency and potentially restore comprehension through self-corrections.

Finally, in terms of remediation for reading difficulties, especially ORF, it is important to note that most interventions typically focus on domain-specific (e.g., reading and language) skills. Outcomes and findings from such interventions have been mixed (Wolf & Katzir-Cohen, 2001), especially adolescent readers who are resilient learners (Fletcher & Vaughn, 2009; Scammaca et al., 2007; for the “decoding threshold hypothesis,” see Wang et al., 2018). To this end, understanding the relative contributions of both domain-specific and domain-general skills to reading outcomes may help tailor and potentially individualize remediation strategies for students with reading difficulties.

Appendix 1: Passage topics and manipulations

		Topics	Manipulations	# of passages
1		Toads	Baseline only	1
2	MO	Moths	Baseline or decoding manipulation	2
3	WB	The West Branch Flood	Baseline or decoding manipulation	2
4	SS	Sap and Syrup	Baseline or vocabulary manipulation	2
5	OC	Octopuses	Baseline or vocabulary manipulation	2
6	IG	Igloos	Baseline or syntax manipulation	2
7	BA	Bugs of the Amazon	Baseline or syntax manipulation	2
8	MU	Mustangs (animal)	Baseline or cohesion manipulation	2
9	HA	Hot Air Balloons	Baseline or cohesion manipulation	2

Appendix 2: Random group assignment for counterbalancing passage administration order

Group	Manipulation	Decoding	Vocabulary	Syntax	Cohesion
A	Baseline	MO	OC	BA	MU
	Manipulated	WB	SS	IG	HA
B	Baseline	MO	OC	IG	MU
	Manipulated	WB	SS	BA	HA
C	Baseline	MO	OC	IG	HA
	Manipulated	WB	SS	BA	MU
D	Baseline	MO	OC	IG	HA
	Manipulated	WB	SS	BA	MU
E	Baseline	WB	OC	IG	HA
	Manipulated	MO	SS	BA	MU
F	Baseline	WB	SS	IG	HA
	Manipulated	MO	OC	BA	MU
G	Baseline	WB	SS	BA	HA
	Manipulated	MO	OC	IG	MU
H	Baseline	WB	SS	BA	MU
	Manipulated	MO	OC	IG	HA
I	Baseline	WB	SS	BA	MU
	Manipulated	MO	OC	IG	HA
J	Baseline	MO	SS	BA	MU
	Manipulated	WB	OC	IG	HA

Appendix 3: Examples of baseline and experimentally-manipulated passages

Baseline versus decoding-manipulated passages

Baseline passage: Moths

Most folks do not grasp the difference between moths and butterflies. Both insects have six legs and are the same size. Both come out of cocoons. You may think that butterflies are more appealing, but that is not always correct. In fact, several butterflies are a tawny or pale color. By contrast, the wings of some moths have nice patterns or bright colors. A more dependable method to identify them is to note when they appear. Butterflies are awake during the daytime. Most moths only wake up after sunset. There is also a visible difference in their feelers, which are used for sensing their surroundings. Moths have thick feelers that look like feathers. A butterfly’s feelers are skinny with a minor bump at the tip.

Most moths use their noses like straws for drinking nectar from flowers. In fact, one has a huge nose that is three times as long as the rest of its body! Another moth has a nose that is as sharp as a blade. It uses its nose to stab an animal’s skin and suck out the blood. It is called a “vampire” moth because it likes to drink blood. Some odd names have been given to others, too. There are “silk,” “wax,” and “hawk” moths.

Everyone is aware of the fact that flying insects are enticed by lamps. Yet for hundreds of years, nobody knew why. However, scientists now think they can explain this fact. They have seen that the light from the moon and stars enable moths to find their way in the dark. When moths see a light bulb’s glow, they get confused because they think the light is up in the sky. The moths you see swirling around lamps at night are lost. By turning off the lights, you will assist them to find their way again.

Decoding-manipulated passage: Moths

Most folks do not realize the difference between moths and butterflies. Both insects have six legs and are the same size. Both come out of cocoons. You may think that butterflies are more colorful, but that is not always accurate. In fact, various butterflies are a beige or neutral color. By contrast, the wings of some moths have pretty designs or beautiful colors. A more reliable way to distinguish them is to observe when they appear. Butterflies are awake during the daytime. Most moths only wake up after the sun sets. There is also an obvious difference in their feelers, which are used for sensing their surroundings. Moths have thick feelers that look like feathers. A butterfly’s feelers are wiry with a miniature knob at the tip.

Most moths use their noses like straws for drinking nectar from flowers. In fact, one has a giant nose that is three times as long as the rest of its body! Another moth has a nose that is as sharp as a knife. It uses its nose to pierce an animal’s skin and suck out the blood. It is called a “vampire” moth because it likes to drink blood. Some unusual names have been given to other ones, too. There are “emperor,” “rough,” and “gypsy” moths.

Everyone is familiar with the knowledge that flying insects are beguiled by lamps. Yet for hundreds of years, nobody knew why. However, scientists now know they can explain this curiosity. They have learned that the light from the moon and stars enable moths to find their way in the dark. When moths see a light bulb’s glow, they get confused because they believe the light is up in the sky. The moths you see circling around lamps at night are lost. By turning off the lights, you will encourage them to find their way again.

Baseline versus vocabulary-manipulated passages

Baseline passage: Octopuses

It is hard to say exactly what an octopus looks like. The body is like a soft floppy bag. An octopus is a very flexible animal. That is because there are no bones to give the octopus a shape. In fact, one of its best skills is changing how it looks. Sometimes an octopus will lie flat like a pancake. At other times the octopus will puff up like a balloon. When making itself look bigger, it hopes to scare away an enemy. All octopuses can also control the color and pattern of their skin. Because they can change color patterns, they can hide anywhere at all. An octopus is always trying to make itself blend in with the things by it.

An octopus has eight long arms. Each arm reaches out in all directions to touch and feel things. Unless there is a hurry, the octopus inches around on its arms. Whenever it needs to go faster, it can also shoot through the water. There is a valve by its head that sucks in a big drink of water. Then the water gets shot out all at once. This helps the octopus move forward quickly.

A hungry octopus is a sneaky hunter. While hiding behind a rock, he watches for a kill with his big eyes. He waits and waits until he finally sees something good to eat. Suddenly, the octopus’s arms dart out to grab and carry the food to his mouth. The mouth has a sharp beak like a bird’s. The octopus uses the beak to crack open the shells of favorite treats, like clams and crabs. There is a long tongue inside the beak with little points on it like a fork. Next, he uses his tongue to scoop the meat out of the shells. Finally, the octopus enjoys the yummy snack that he caught.

Vocabulary-manipulated passage: Octopuses

It is tricky to say exactly what an octopus looks like. The body is like a soft floppy bag. An octopus is a very supple animal. That is because there are no bones to give the octopus a shape. In fact, one of its best talents is altering how it looks. Sometimes an octopus will lie flat like a pancake. At other times the octopus will puff up like a balloon. When making itself look bigger, it hopes to scare away an enemy. All octopuses can manage the hue and pattern of their skin. Because they can alter color patterns, they can hide anyplace at all. An octopus is always trying to make itself blend in with the objects nearby it.

An octopus possesses eight long arms. Each arm reaches out in all directions to touch and feel items. Unless there is a hurry, the octopus crawls around on its arms. Whenever it wishes to go faster, it can also zip through the water. There is a valve by its head that sucks in a big gulp of water. Then the water gets spat out all at once. This helps the octopus move forward swiftly.

A famished octopus is a sneaky predator. While hiding behind a reef, he watches for a target with his big eyes. He waits and waits until he finally glimpses something yummy to eat. Suddenly, the octopus’s arms dart out to grab and transmit the food to his mouth. The mouth has a sharp beak like a bird’s. The octopus uses the beak for cracking open the shells of favorite treats, like scallops and prawns. There is a long tongue inside the beak with little points on it like a fork. Next, he uses his tongue to scoop the meat out of the shells. Finally, the octopus enjoys the yummy snack that he caught.

Baseline versus syntax-manipulated passages

Baseline passage: Bugs of the Amazon

The animals of the Amazon rain forest make it a rare kind of place. There are more types of bugs living there than any other place on earth. In the Amazon rain forest, many bugs have clever ways of killing other animals. Their victims may be bitten or even eaten alive. Some killers attack their prey directly. However, others are more sneaky hunters. Some bugs hide in plain sight to make their victims feel safe. The most deadly killers are the smallest animals in the rain forest.

Spiders can be deadly hunters. The venom from a “black widow” spider is fifteen times stronger than a snake’s. The black widow can kill large animals with just one bite. However, other spiders have to work harder for their meals. The “net-casting spider” spins a small web that it holds like a net. Then the spider waits and waits until a bug comes by. Next, the spider jumps on the bug and covers it with the net very quickly. Finally, when the bug stops moving, the spider gives it a fatal bite.

Bugs from the mantis family are also expert killers. They make a trap to trick their prey. A “dead leaf” mantis sways in the breeze to make itself look like a dead leaf. The “praying” mantis looks like a twig when it sits very still. When a bug lands near, it jumps on its victim for the kill. A mantis will only hunt alone. But other bugs, such as “army ants,” hunt together. They march in big groups, just like human soldiers. These ants attack and eat anything that gets in the way. Every living thing is food to these hunters. Bigger animals have to run away when a colony of army ants attacks. The good news is the Amazon rain forest and all of its scary bugs are far away.

Syntax-manipulated passage: Bugs of the Amazon

The animals of the Amazon rain forest make it a rare kind of place. More types of bugs live there than any other place on earth. In the Amazon rain forest, many bugs have clever ways of killing other animals. Their victims may be bitten or eaten alive. Some killers attack their prey directly. However, others are more sneaky hunters. Even in plain sight, some bugs can hide to make their victims feel safe. By far, the smallest animals in the rain forest are the deadliest.

Spiders and other bugs in the Amazon can be deadly hunters. The venom from a “black widow” spider is fifteen times stronger than a snake’s. With just one bite, a black widow can kill large animals. However, other spiders work harder for their meals. The “net-casting spider” spins a small web that it holds like a net. Then, with its net ready, the spider waits for a bug to come by. Next, in a flash, the spider jumps on the bug and covers it. Finally, as soon as the bug stops moving, the spider gives it a fatal bite.

Bugs from the mantis family are also expert killers. They make a trap to trick their prey. A “dead leaf” mantis sways in the breeze to make itself look like a dead leaf. The “praying” mantis looks like a twig when it sits very still. When a bug lands near, it jumps on its victim for the kill. Lone hunters such as the mantis do not hunt in groups. But other bugs, such as “army ants,” hunt together. Just like human soldiers, they march in big groups. Anything that gets in the way will be eaten. For these hunters, every living thing is food. When a colony of army ants attacks, bigger animals run away. However, the Amazon rain forest and all of its scary bugs are very far away.

Baseline versus cohesion-manipulated passages

Baseline passage: Mustangs

Spanish soldiers took their horses to America over five hundred years ago. Many of the horses escaped over time, and today their offspring are called “mustangs.” They live in big herds in the open lands of the far west. In the past, mustangs were not protected by any law. It was not against the law for humans to hunt them. A hundred years ago, mustangs had been hunted to make dog food. But Congress finally changed the law thirty-five years ago. In fact, harming a mustang is now a crime that could send a person to prison. Today, state ranchers protect these animals from harm.

Mustang herds are booming these days. State ranchers use airplanes to locate and survey the herds each year. When herds grow too big for their land, some horses are moved to new lands. Other mustangs are put up for adoption, but first the wild horses must be tamed. One of the main places where they tame them is in jails. The inmates are trained to break in horses. To win the trust of the horses, they must be calm and kind. At first, the wild horses do not trust humans. But the mustangs start to enjoy working with their handlers in the end.

The tamed horses are finally ready for new human handlers after a few months of training. The inmates get the horses ready for their new owners. Then, families who want to adopt a mustang visit the jail. They meet the inmates and talk to them about the horses. Next, the inmates explain about the personality of each mustang. Horses differ in how calm, kind, and playful they are. It is important for each owner to choose the right horse. Finally, signing a contract is the last step. It says that the state will take the mustang back if it is hurt.

Cohesion-manipulated passage: Hot Air Balloons

Spanish soldiers took their horses to America over five hundred years ago. Over time, many of them have escaped. Today their offspring are called “mustangs.” Now, big herds live in the open plains way out in the far west. A hundred years ago, humans hunted them to make dog food. In the past, there were not any laws to help mustangs. It was not illegal to hunt them. Congress finally changed the law thirty-five years ago. Now, harming a horse is a crime that could send a person to prison. Park rangers protect these animals from danger.

Using airplanes, state ranchers locate and survey mustang herds each year. These days, they are booming. When the herds get too spread out for their area, some horses are moved to new lands. The others must be tamed before they can be adopted. One of the places where that happens is in jails. The inmates take a lot of classes about how to break in horses. To win their trust, prisoners must be kind and calm all the time. The broncos do not like their handlers at first. The mustangs start to enjoy working with them, in the end. Needless to say, it is a very difficult job.

The tamed steeds are ready to be adopted after a few months of hard work. The inmates wash and brush their coats to get them ready to meet their new owners. Families that want to adopt visit the jail. They meet the trainers and talk to them about the horses. Next, the inmates explain about the personality of each mustang. They differ in how calm, kind, and playful they are. It is important to choose the right one. Signing a contract is the last step. It says that the state will take the mustang back if it is hurt.

Appendix 4: Summary of regression results predicting WCPM from EF and reading and language skills for baseline passages only

	WCPM
	B	se
Fixed effects
Intercept	−0.039	0.115
+ Executive functioning
Reasoning	0.007	0.067
Cognitive flexibility	0.017	0.072
Planning/organization	0.148*	0.072
Working memory	0.171**	0.065
+ Reading + language
Decoding	0.248**	0.087
Vocabulary	0.185*	0.084
Morphology	0.073	0.062
+ Demographics
Sex	0.062	0.109
Age	0.154**	0.048
Socioeconomic status	−0.027	0.057
Random effects	σ	sd
Participant	0.226	0.476
Topic	0.034	0.186
Group	0.000	0.000
Residual	0.641	0.476

^*p < 0.05,

^**p < 0.01

Appendix 5: Summary of regression results predicting WCPM

	Step 1		Step 2		Step 3
	B	se	B	se	B	se
Fixed effects
Intercept	−0.723**	0.233	−0.709**	0.233	−0.705**	0.241
+ Passage covariates
Decoding	−0.291*	0.114	−0.292*	0.115	−0.291**	0.114
Vocabulary	−0.429**	0.100	−0.423**	0.099	−0.423**	0.099
Syntax	−0.297*	0.134	−0.280*	0.135	−0.284*	0.136
Cohesion	−0.030	0.111	−0.029	0.110	−0.029	0.110
+ Executive functioning
Reasoning			0.035	0.072	0.010	0.071
Cognitive flexibility			0.056	0.077	0.095	0.075
Planning/organization			0.146†	0.077	0.210**	0.075
Working memory			0.214**	0.070	0.206**	0.071
+ Reading + language
Decoding					0.305**	0.091
Vocabulary					0.214*	0.088
Morphology					0.075	0.065
+ Demographics
Sex	0.151	0.102	−0.107	0.096	−0.125	0.080
Age	0.177**	0.039	0.084*	0.040	0.088*	0.038
Socioeconomic status	0.121†	0.049	0.048	0.050	0.035	0.043
Random effects	σ	sd	σ	sd	σ	sd
Participant	0.523	0.723	0.172	0.415	0.082	0.287
Topic	0.076	0.275	0.000	0.000	0.000	0.000
Group	0.009	0.094	0.000	0.000	0.008	0.092
Residual	0.483	0.695	0.721	0.849	0.724	0.851

^†p < 0.10,

^*p < 0.05,

^**p < 0.01