Reading comprehension involves multiple components, from the identification of words to meaning and integration processes, and the updating of mental models that result in global comprehension, all supported by multiple sources of linguistic and conceptual knowledge. These processes are linked in systematic ways within the structure of a cognitive system (e.g., the Reading Systems Framework; Stafura & Perfetti, 2014). Ignored in a purely cognitive process analysis, however, are the differences among readers in how they approach reading situations—their approaches to text. Some research has addressed readers’ motivation (Davis et al., 2018; Schutte & Malouff, 2007; Wigfield & Guthrie, 1997) and need for cognition (Cacioppo et al., 1984; Cohen et al., 1955).

Notably, questionnaires targeting adults’ reading motivation are scarce (Davis et al., 2018; Schiefele et al., 2012). In a review by Davis et al. (2018), out of the 16 motivation scales reported since 1990, only one was specifically designed for adults (i.e., Schutte & Malouff, 2007). Furthermore, noncognitive reader factors have rarely penetrated the cognitive analysis of adults’ reading comprehension. Our aim is to establish a link between readers’ approaches to text and cognitive aspects of reading. We do this by reporting the development of an instrument to assess readers’ approaches to text, demonstrating its measurement usefulness, and testing its relation to reading experience and comprehension skill.

Cognitive components of reading

Fluid reading comprehension is the result of word identification and meaning retrieval, integrative processes (e.g., Perfetti & Helder, 2021), and the building and updating of a mental model of the text meaning (e.g., Broek & Gustafson, 1999; Graesser et al., 1994; Kintsch, 1988)—driven by sources of linguistic and world knowledge. The Reading Systems Framework (RSF) captures this complexity in three core systems: knowledge (e.g., orthographic, linguistic, and general knowledge), word identification, and comprehension (Perfetti & Helder, in press; Stafura & Perfetti, 2014).

The lexicon has a central role in the RSF in linking the word identification and comprehension systems. A high-quality, reading-accessible vocabulary enables fluid access to word meanings and their integration with text meanings. It also supports inferences (Cain & Oakhill, 1999, 2014; Currie & Cain, 2015) and facilitates the learning of new words through reading (Cunningham & Stanovich, 1997; Martin-Chang & Gould, 2008; Stanovich & Cunningham, 1992).

One hypothesis concerning relations among reading systems is that reading experience (i.e., readers’ cumulative encounters with texts and written material throughout their lifetime) affects reading comprehension by increasing the reader’s vocabulary: More reading experience leads to more words learned and thus more support for comprehension. This causal chain is consistent with research on vocabulary growth through reading (e.g., Duff et al., 2015; Echols et al., 1996; Martin-Chang & Gould, 2008; Nagy et al., 1985; Nation, 2017) and the results of Cunningham and Stanovich (1997) who found that 11th grade print exposure predicted 11th grade vocabulary knowledge—but not reading comprehensionwhen first-grade vocabulary knowledge was controlled. Reading experience predicted reading comprehension only when vocabulary knowledge was not controlled, suggesting that vocabulary knowledge learned through reading experiences plays a role in linking reading experience to reading comprehension skill (Cunningham & Stanovich, 1997).Footnote 1 Accordingly, one aim of our studies is to provide more evidence concerning interrelationships among cognitive components of reading by examining the effect of reading experience on reading comprehension, with vocabulary knowledge as a mediator. A second aim is to examine our assessment of readers’ approaches to text within this cognitive reading model: Readers’ approaches to text may be associated directly with comprehension or more indirectly by influencing one’s reading experience and the vocabulary knowledge gained from reading.

Among the comprehension system components of the RSF, comprehension monitoring is relevant for how a reader approaches a text. Readers may be more or less inclined to monitor their level of comprehension and to use comprehension repair strategies, two components of comprehension monitoring (Baker, 1979; Wagoner, 1983). Relatedly, readers may become more or less engaged with a text, affecting comprehension monitoring and other high-level comprehension processes. Text engagement and comprehension monitoring vary both among readers and situations for a given reader (Guthrie et al., 2004; van den Broek et al., 2011). Some situations demand deeper engagement (reading legal documents to create a court brief) than others (reading the same texts to search for names of people).

Individual differences in readers’ approaches to text

External to the cognitive systems of comprehension but linked to their functioning are the approaches that the reader brings to reading a text. To examine how readers’ approaches are linked to reading comprehension, we developed The Readers’ Approaches to Text Questionnaire (TReAT-Q). Because our assessment is specifically about reading, we first review relevant concepts from reading research and distinguish our reading-specific construct from related instruments designed to relate to more general cognitive functioning, specifically, the need for cognition (Cacioppo et al., 1984; Cohen et al., 1955).

Standards for comprehension and regulatory processes

Readers may differ in their standards for comprehensiona generalization of van den Broek et al.’s (1995) concept of standards for coherencethe reader’s need to establish coherence across text elements to maintain text understanding (van den Broek et al., 2011; Kintsch & van Dijk, 1978; O’Brien & Cook, 2016; Perfetti et al., 2005; van Dijk & Kintsch, 1983; Wigfield et al., 2016). A standard for comprehension affects how the reader handles the continuous updating of text comprehension, which depends on effort at comprehension, including comprehension monitoring, and repair of comprehension failures. To engage in active repair processes, the reader may reread a portion of the text, generate a missing inference, or figure out (or look up) the meaning of an unknown word (e.g., Graesser, 2007; van den Broek & Helder, 2017). The degree to which a reader engages in these repair strategies in turn depends on a standard for coherence the reader applies to text reading (van den Broek & Helder, 2017; van den Broek et al., 1995). There is little repair to do if the reader has a low standard for coherence (or if the text itself has high cohesion; e.g., McNamara et al., 1996).

These regulatory processes are where the reader’s approach to text meets the cognitive components of comprehension. As a reader characteristic, one reader may have both a high standard for coherence and a willingness to expend effort in overcoming a breakdown in coherence, while another may have only the first (thus giving up when coherence breaks down) or neither. Further, some readers may seek reading situations that require more effort, whereas other readers may avoid these more challenging text situations. It is these reader variations that we aim to measure in our approaches-to-text assessment. Research on standards for coherence has focused on situation-based factors, e.g., reading goals (van den Broek et al., 2011; Yeari et al., 2015). However, individual differences in standards of coherence or related approaches to text should also affect reading comprehension. Readers who enjoy reading and engage more with their texts may have higher standards than those who do not. It is these interrelated factors—standards for comprehension, motivation, engagement, reading experience, and enjoyment—that are components of a reader’s approach to text. Currently, there is no comprehensive measure that combines adult readers’ motivational and cognitive aspects of reading and the strategies readers adopt to reach their standards for comprehension, which our readers’ approaches to text questionnaire aims to address.

Reading motivation

Reading motivation is generally understood as an inclination to seek reading activities. It is often defined as (a variation of) “the individual’s goals and beliefs with regard to reading” (Guthrie & Wigfield, 1999, p. 199) and arises from both intrinsic and extrinsic sources. For example, a person may read because they enjoy reading (intrinsically; Wigfield & Guthrie, 1995, 1997) or because the reading activity serves an external goal, e.g., reading to succeed on an exam (Deci & Ryan, 1985; Ryan et al., 1990). Investigations of reading motivation led to the development of the Motivations for Reading Questionnaire (MRQ; Wigfield & Guthrie, 1995, 1997) and the related Reading Motivation Questionnaire (RMQ; Schiefele & Schaffner, 2016) for children, which capture more fine-grain facets of reading motivation. Although some items focused on schoolwork avoidance and reading challenge, the MRQ was designed for children and many questions focused on classroom-based settings that may not pertain to adults.

In one of the few motivation measures for adults (Davis et al., 2018), Schutte and Malouff (2007) found that their multi-dimensional adult reading motivation survey (AMRS) predicted reading enjoyment and self-reported reading frequency. Readers with more intrinsic motivation were more likely to report higher reading enjoyment and spent more time reading recreationally, whereas reading to do well in other realms was more related to time spent on required reading (Schutte & Malouff, 2007). The AMRS focused on four dimensions: reading efficacy, reading as part of self, reading for recognition, and reading to do well in other realms, none of which specifically focus on whether readers avoid reading certain texts, the use of reading strategies, and reading effort, which we aim to address in the present study.

Nonetheless, motivation to read has been shown to more fully engage the cognitive processes that lead to comprehension in children (Wigfield et al., 2008; Zentall & Lee, 2012). For example, in a 12-week intervention study, Wigfield et al. (2008) found that combining motivation and reading strategy instruction resulted in higher reading engagement and reading comprehension, assessed by standardized and experimenter-generated tests, than reading strategy instruction alone. Importantly, reading engagement mediated the effect of the motivation and comprehension strategies intervention; readers with greater increases in reading engagement showed greater gains on reading comprehension measures.

In general, research on reading motivation has focused on why readers read texts and their relation to reading experience and/or comprehension (e.g., Schutte & Malouff, 2007; Wigfield & Guthrie, 1997). This focus has produced important results linking motivation to outcomes; more intrinsic aspects (e.g., curiosity, emotional regulation) are associated with better comprehension, whereas more extrinsic motivations (e.g., social, grades) are associated with worse comprehension (Schaffner et al., 2013; Schiefele & Schaffner, 2016; Wang & Guthrie, 2004). In addition to these links to why readers read (i.e., motivations) is how readers generally approach the texts they read. Our readers’-approaches-to-text construct uses the distinction between intrinsic and extrinsic motivation on the assumption that both relate to how readers’ approach texts. How engaged readers are when they read and the way in which they approach texts will likely affect readers’ reading experiences and comprehension.

Engagement

Engagement generally can be conceptualized as consisting of four dimensions: behavioral (e.g., persisting through a task), emotional (affect related to an activity), cognitive (e.g., strategies), and agentic (acting on a self-derived motivation to pursue an activity; Reeve et al., 2020; Veiga, 2016; Veiga et al., 2014). The extent to which motivated readers become “engaged” with a text regulates their reading processes and thus their comprehension, enjoyment, and other outcomes (Wigfield & Guthrie, 2000). Reading engagement and reading motivation are closely related; however, they are distinct in their control of reading processes. Motivation can prepare a reader to engage in a text, but it is the engagement itself—because it entails attention and the recruitment of cognitive resources (Guthrie & Cox, 2001; Wang & Guthrie, 2004)—that has a direct effect on reading processes. Engagement leads to more positive affect, more interest, more use of appropriate reading strategies, and more knowledge-building through reading (Guthrie et al., 2004). Less-skilled readers report lower enjoyment during reading—a component of engagement—during reading (Crossley et al., 2017), which negatively affects comprehension. Beyond skill-based individual differences, engagement is situational. When readers find texts more interesting, they engage more with the text and thus are less likely to be distracted and more likely to achieve comprehension (Unsworth & McMillan, 2013), with engagement showing a positive correlation to literacy (Brozo et al., 2007).

Experience

The accumulated encounters with texts add up to reading experience. Thus, both reading frequency and amount should contribute to a reader’s overall reading experience. Variations in reading experience lead to variations in reading-relevant knowledge (familiarity with word forms, vocabulary, syntax, text conventions) that is critical for fluid reading comprehension (Duff et al., 2015; Echols et al., 1996; Martin-Chang & Gould, 2008; Nation, 2017; Torppa et al., 2020). We should expect these reading experiences, both their amount and their quality, to be influenced by readers’ motivations and engagement (e.g., Fulmer & Frijters, 2011). For example, intrinsic motivation shows direct (Wang & Guthrie, 2004) and sometimes indirect effects on reading comprehension via reading amount (Schaffner et al., 2013; c.f. Becker et al., 2010), whereas extrinsic motivation negatively effects reading comprehension and reading amount (Schaffner et al., 2013; Wang & Guthrie, 2004). Schaffner et al. (2013) demonstrated that both intrinsic and extrinsic motivation had indirect effects (positive and negative, respectively) on reading comprehension via reading amount, whereas reading amount did not directly predict reading comprehension. This suggests that reading motivation, which we argue is part of readers’ approaches to text, influences what readers gain from their reading experiences, which in turn influence reading comprehension. Although Schaffner et al. (2013) also showed that intrinsic motivation related to comprehension at the word and sentence levels, not examined was how reading experience may affect gains in vocabulary knowledge.

In the present study, we address this gap by examining how readers’ approaches to text relate to reading experiences and how those reading experiences relate to vocabulary knowledge and reading comprehension. Related to readers’ approaches to text, readers’ reading enjoyment is important for facilitating reading experiences; readers who report enjoying reading more difficult or challenging texts have more reading experiences (Schutte & Malouff, 2007; Wigfield & Guthrie, 1997). Experience with challenging texts may allow more opportunities to engage in comprehension repair or prompt the use of contextual knowledge to support inferences (2011b; McNamara et al., 1996, 2012; Schmitt et al., 2011; van den Broek et al., 2011). Furthermore, experience with more challenging texts may contribute to increased vocabulary and reading comprehension gains. Because of the importance of reading experience in the accumulation of knowledge that drives reading processes, we use the readers’ approaches to text framework to assesses whether a reader’s approaches to text influence reading experiences and reading comprehension.

Need for cognition

Because reading can involve texts that vary in complexity and challenge, the reader’s approach to text may be influenced by the more general construct of need for cognition. As identified by Cohen et al. (1955), individuals vary in the extent to which they enjoy thinking about or understanding complex problems and situations. Similar to our conceptualization of readers’ approaches to text, the Need for Cognition questionnaire includes items relating to how individuals approach complex problems (e.g., enjoyment of complex tasks or avoidance of difficult cognitive problems; Cacioppo et al., 1984). A need for cognition has been found to predict comprehension of expository and narrative texts (Dai & Wang, 2007) and is associated with reading frequency and number of books read (Juric, 2017). We expect that a need for cognition could influence a reader’s approach to challenging texts. We aimed to determine whether our assessment of readers’ approaches to text, with its focus on reading behaviors, would add a distinctive explanation beyond the need for cognition questionnaire developed by Cacioppo et al. (1984).

Present studies

To examine how a reader’s approaches to text relate to reading comprehension, we carried out two studies to assess adult readers’ self-reports of their motivation, desire to understand what they read, embrace or avoidance of text difficulty, and use of reading strategies and related these components of readers’ approaches to text to reading comprehension. The core assumption of the assessment is that the reader’s approaches to text influence the extent to which readers engage in strategies and procedures that produce variation in text comprehension.

Study 1 describes the development of The Reader’s Approaches to Text Questionnaire (TReAT-Q), a self-report measure for adults administered to a community sample, which goes beyond why adults read (i.e., reading motivation) to include how they approach texts. Additionally, because an approach to text may partly reflect a need for cognition, Study 1 also tests whether readers’ approaches to text still relate to readers’ reading experiences when need for cognition is accounted for. We expected (1) TReAT-Q to consist of at least three factors relating to readers’ goals/motivation, desire to understand, and strategies implemented to meet a desired level of understanding (i.e., comprehension monitoring) and (2) Readers’ approaches to text to predict readers’ reading experiences when controlling for need for cognition, which has previously been linked to reading experiences (e.g., Juric, 2017).

Study 2, conducted with a university sample, had two main goals. The first goal was to test a cognitive model of reading comprehension, based on the RSF, by implementing a structural equation model (SEM) to test direct and indirect effects of reading experience on reading comprehension with vocabulary knowledge as a mediator in a university sample. The second goal was to extend the cognitive model of reading comprehension by including readers’ approaches to text and to examine (1) their prediction of reading experience as a confirmation of findings from of Study 1 and (2) their possible direct and indirect effects on reading comprehension with reading experience and vocabulary knowledge as mediators. We also included an assessment of test–retest reliability for TReAT-Q by administering the questionnaire twice to the same university sample with an interval of 2 weeks. Study 2 hypotheses are as follows: (1) Vocabulary knowledge and reading experience will be positively associated with reading comprehension, with vocabulary as a mediator, (2) Readers’ approaches to text, as measured by TReAT-Q, will be positively associated with reading experience and comprehension, with direct and indirect effects on both, and (3) Readers’ approaches to text are stable and TReAT-Q will show good test–retest reliability.

Study 1: Readers’ approaches to text questionnaire

The aim of Study 1 was to develop a measure of readers’ approaches to text and to test its relation to adults’ self-reported reading experience and to test whether scores on this measure predict readers’ reading experiences when controlling for need for cognition. To measure individual differences in approaches to text, a questionnaire was developed and refined to an exemplar stage using an exploratory factor analysis (EFA).

Method

Participants

Study 1 included 205 native English speakers (20 self-reported speaking a second language) recruited via Amazon Mechanical Turk. Participants self-identified as having no psychological or learning disabilities and normal or corrected-to-normal vision. Participants received $2.00 as compensation and gave informed consent to participate in the study. The responses of one participant who responded in a patterned manner (i.e., visually detected zig-zag pattern) were excluded. Of the remaining 204 participants, 112 were male and 92 were female (no participant selected “Other” as an option). Participants’ ages ranged from 18 to 71 (Mage = 31.96, SD = 8.58); 20 had a high school diploma or GED equivalent, 20 had an associate degree, 53 completed some college, 81 had a bachelor’s degree, and 30 had a master’s or doctoral degree. One hundred sixty-three (79.9%) participants were White, 17 (8.3%) were Black or African American, nine (4.4%) were Asian, 11 (5.4%) were Hispanic or Latino, two (1.0%) were Native American or Alaska Native, and two (1.0%) selected Other. No participant selected “Native Hawaiian or Pacific Islander.”

Measures

Participants completed three measures, TReAT-Q, Need for Cognition questionnaire (NFC; Cacioppo et al., 1984), and a subset of the Adult Reading History Questionnaire (ARHQ), based on Lefly and Pennington’s (2000) adaptation of Finucci et al.ʼs, (1982; 1984 questionnaire.

The reader’s approaches to text questionnaire (TReAT-Q)

Items for TReAT-Q were generated by the first author based on open-ended responses from 11 undergraduate students, graduate students, and post-doctoral researchers on the following three questions: (1) Why do you generally read? (i.e., reading goals/motivation), (2) What motivates you to understand a text? (i.e., desire to understand), and (3) What reading strategies do you use to understand a text, and in what situation? (i.e., comprehension monitoring). For example, a response to “What reading strategies do you use to understand a text, and in what situation?”, was “If I can’t infer what a word means based on context, I will look up its definition online.”, and was used as separate items in the measure as: (1) “I often look up concepts I do not know.” and (2) “If I cannot guess what a word means based on its context, I will look it up.” Based on responses to the “Why do you read?” question, items related to both intrinsic and extrinsic motivation were included.

The questionnaire included 87 items across the three hypothesized categories: reading goals/motivation (30 items), desire to understand (30 items), and comprehension monitoring (27 items). Five of the 87 items were adapted from the AMRS (Schutte & Malouff, 2007; indicated in Appendix A). Of the 87 items, 24 were reverse scaled. Participants rated whether they agreed that a statement was characteristic of them on a Likert scale ranging from strongly disagree (– 3) to strongly agree (3).

Need for Cognition (NFC)

The NFC questionnaire consists of 18 items (Cacioppo et al., 1984; Appendix B) related to preferences for problem solving, including complex and abstract problems. Nine items were reverse scaled. Participants rated whether they agreed that a statement is characteristic of them on a Likert scale ranging from strongly disagree (– 3) to strongly agree (3). The score range of this measure is – 54 to 54.

Reading experience: Adult Reading History Questionnaire (ARHQ) subset

As an indirect measure of their reading experience, participants also completed a subset of seven items from the Adult Reading History Questionnaire (ARHQ; Lefly & Pennington, 2000) and one question from the Reading Activity Questionnaire (Parault & Williams, 2009; Appendix C). The seven questions fit into three broad categories: Participants’ reading frequency as (1) time spent reading for pleasure and (2) time spent reading for work on a seven-point Likert scale (0–3 with halves), Article reading as (3) frequency of reading daily newspapers, (4) reading a newspaper on Sundays, and (5) reading newspaper and/or magazine articles on the internet on a four-point Likert scale (0–3) and Estimated reading amount measured as, (6) books read for pleasure each year (none, 1–2, 2–5, 6–10, more than 10), and (7) number of hours read in the previous week (less than 5, 5–10, 11–20, 21–30, 30 +). Many of these self-report items do not indicate a time period for which these experiences occur; however, we expect responses to relate to readers’ general reading behavior. Scores ranged from 0–23.

An additional question from the Reading Activity Questionnaire asked participants to indicate the types of reading materials they read (e.g., fantasy, science fiction, etc.); however, because this item was not used in analyses, its results are not presented here.

Procedure

Participants completed TReAT-Q, the NFC questionnaire, and the subset of the ARHQ, in that order. All questionnaires were administered via Qualtrics (Qualtrics, Provo, UT) and took 15 min to complete.

Analysis procedure

The analyses were carried out in two steps. First, a principal axis factor (PAF) analysis, a type of exploratory factor analysis (EFA), was conducted to extract the items of TReAT-Q that significantly loaded onto the factors identified based on a Horn’s parallel analysis. The Horn’s parallel analysis estimated the number of factors identified from a randomly sampled set of the data and a randomly generated set of data from a normal distribution (Revelle, 2020). The number of factors to retain should be fewer than the number identified by the parallel analysis. A PAF analysis was chosen because it is useful for non-normal data (Bartholomew, 1980; Fabrigar et al., 1999), which was the case in the present study. For the PAF analysis, factor loadings greater than 0.5 were considered significant (Kline, 2011; Schutte & Malouff, 2007). Multiple fit indices assessed model fit. Non-significant χ2 values indicate good fit. The root mean square error of approximation (RMSEA) and Tucker–Lewis Index (TLI) were also included to assess model fit. TLI values > 0.90 indicate acceptable fit (Sharma et al., 2005) and values > 0.95 indicate good model fit (Hu & Bentler, 1999). RMSEA values < 0.08 and < 0.05 indicate acceptable and good fit, respectively (Hu & Bentler, 1999).

The second step implemented hierarchical regressions using the trimmed TReAT-Q to account for scores on the self-reported reading experience questionnaire (ARHQ subset), controlling for need for cognition. Analyses with NFC and TReAT-Q were conducted separately, then both were entered into a single model. If the change in variance accounted for is significant when TReAT-Q is entered second into the combined model, this would indicate that the readers’ approach to texts questionnaire accounts for additional variance not explained by need for cognition. Further, a non-significant change in variance when NFC is entered second would suggest that the new TReAT-Q alone is a sufficient predictor of reading experience (indexed via self-reported reading amount and frequency).

Results

Before assessing whether TReAT-Q related to need for cognition (NFC) and readers’ reading experience (ARHQ), parallel and PAF analyses estimated the number of factors and items to retain in the TReAT-Q.

Principal axis factor analysis on TReAT-Q

TReAT-Q was designed to reveal three hypothesized factors (reading goals/motivation, desire for understanding, and comprehension monitoring); however, it is possible that the data reflect a different factor structure than hypothesized. Therefore, a parallel analysis using the lavaan package (Rosseel, 2012) in R (R v. 4.0.3) estimated the number of factors to retain. Although this parallel analysis suggested seven factors, the scree plot showed that three of these seven factors were close to an eigen value of 1 and close to the eigenvalues generated by resampled random data and simulated data, thus, four factors were retainedFootnote 2 (Fig. 1).

Fig. 1
figure 1

Scree plot of parallel analysis. The black horizontal line indicates an eigen value of 1

Full size image

Next, four factors were estimated with a PAF, using the oblique, oblimin rotation from the lavaan package in R (Rosseel, 2012). Forty-five of the 87 items had factor loadings greater than 0.5 (Appendix A). If there were two or more similarly worded items (e.g., items 63 and 67 in Appendix A), the item with the highest factor loading was retained, resulting in 38 items. The PAF analysis was re-run, removing items with factor loadings < 0.5, until all factor loadings were ≥ 0.5 and cross-loading differences were greater than 0.15. The final questionnaire contained 31 items (Table 1). Based on the shared characteristics of items that loaded onto a given factor, the four factors were defined as follows: (1) intrinsic goals (e.g., “I read because I enjoy it.”, “It is important for me to read regularly.”), (2) extrinsic goals and strategies (e.g., “I read to study for exams or a job promotion.” and “I often make summaries of what I read.”), (3) effort at understanding (e.g., “I try to really understand a text, so I can form my own opinions about the content of the text.”, and (4) avoidance of text difficulty (e.g., “I’d rather read easy texts than difficult ones.”). These four factors accounted for 51.95% of the variance in the measure. Intrinsic goals accounted for 14.49% of variance (seven items), extrinsic goals and strategies accounted for 12.29% of the variance (eight items), effort at understanding accounted for 13.20% of the variance (nine items), and avoidance of text difficulty accounted for 12.0% of the variance (seven items). This model had a good fit to the data, χ2 = 277.3, df = 347, p > 0.999; TLI = 0.902; RMSEA = 0.055 (90% CI = 0.05 to 0.06).

Table 1 PAF analysis factor loadings of retained TReAT-Q items
Full size table

The factors of intrinsic goals (α = 0.91, 95% CI = 0.89 to 0.93), extrinsic goals and strategies (α = 0.85, 95% CI = 0.82 to 0.88), effort for understanding (α = 0.86, 95% CI = 0.82 to 0.88), and avoidance of text difficulty (α = 0.86, 95% CI = 0.84 to 0.89) all had good internal consistency. Internal consistency for the entire questionnaire (31 items) was good (α = 0.88, 95% CI = 0.86 to 0.90). The possible score range of the overall measure was – 93 to 93 and, within the dataset, scores ranged from – 41 to 85 (M = 29.72, SD = 22.90).

Descriptive statistics

Scores on the ARHQ subset ranged from 2–23 and the average score was 11.39 (SD = 3.86), with a Cronbach’s alpha of α = 0.68, 95% CI = 0.61 to 0.75. NFC scores ranged from – 51 to 54 and the average score was 13.55 (SD = 23.91), with a Cronbach’s alpha of α = 0.95, 95% CI = 0.94 to 0.96. Table 2 shows correlations among TReAT-Q subscales, the ARHQ subset (z-scored), and the NFC questionnaire. All measures were correlated, except extrinsic goals and strategies were not significantly correlated with intrinsic goals, effort at understanding, or avoidance of text difficulty (ps > 0.47).

Table 2 Study 1 correlations
Full size table

Demographic factors related to TReAT-Q scores

Table 3 shows the scores on each subscale of TReAT-Q by Education, Gender, and Race/Ethnicity (White vs. non-White). To examine which demographic factors related to each of the subscales, separate regressions were conducted on each score for Age, Education, Sex, and Race/Ethnicity, with the mean across groups as the intercept. Demographic factors were compared using simple coding and Education levels in the model were compared to “High school diploma (or GED)”. Age was a significant predictor for intrinsic goals (β = 0.15, t(196) = 2.17, p = 0.031), avoidance of text difficulty (β = 0.24, t(196) = 3.88, p < 0.001), and extrinsic goals and strategies (β = – 0.33, t(196) = – 4.10, p < 0.001), but not effort at understanding (β = 0.02, t(196) = 0.33, p = 0.74). Females had higher scores for intrinsic goals than males (β = 4.06, t(196) = 3.50, p < 0.001). Participants who identified as White had lower scores on the extrinsic goals and strategies subscale than participants who identified as non-White (β = – 4.36, t(196) = – 2.57, p = 0.011). Participants with a bachelor’s degree (β = 5.98, t(196) = 2.92, p = 0.004) and master’s/doctoral degree (β = 6.35, t(196) = 2.69, p = 0.008) had higher intrinsic goals scores than those with a high school diploma (or GED). Participants with a bachelor’s degree had higher avoidance of text difficulty scores (i.e., less avoidance of difficult text situations; [β = 4.56, t(196) = 2.46, p = 0.015]) than participants with a high school diploma (or GED). Regarding effort at understanding, participants with an associate’s degree ((β = 5.74, t(196) = 2.04, p = 0.043), some college (β = 4.92, t(196) = 2.11, p = 0.036), bachelor’s degree (β = 7.75, t(196) = 3.48, p < 0.001), and master’s/doctoral degree (β = 6.87, t(196) = 2.68, p = 0.008) scored higher than those with a high school diploma (or GED).

Table 3 TReAT-Q subscale scores by education, gender, and race
Full size table

TReAT-Q as a predictor of reading experience (ARHQ subset)

Hierarchical regressions tested whether the four TReAT-Q subscales accounted for variance in adults’ self-reported reading experiences (ARHQ subset) after accounting for need for cognition (NFC). Each TReAT-Q subscale score was calculated as the sum of scores across items within a given factor identified by the PAF (e.g., intrinsic goals score = sum of the seven items that loaded onto the factor). ARHQ subset scores were z-scored to account for scale differences between items. In all regression analyses, Age, Gender, Ethnicity (White vs. Non-White), and Education served as control variables, with the average across groups serving as the intercept. Demographic factors were compared using simple effects coding and Education levels in the model were compared to “High school diploma (or GED)”.

Three regression analyses were conducted: (1) NFC model, with NFC predicting ARHQ along with the control variables, (2) TReAT-Q model, with each subscale of TReAT-Q predicting ARHQ, and (3) a combined model with NFC + TReAT-Q subscales predicting ARHQ (see Table 4). In the first NFC model, NFC significantly predicted ARHQ scores and the model accounted for a significant amount of variance in ARHQ (F(8, 195) = 7.04, R2 = 0.19, p < 0.001). In the TReAT-Q model, with the four TReAT-Q subscales predicting ARHQ, intrinsic goals, extrinsic goals and strategies, and effort at understanding predicted ARHQ (Table 4). However, avoidance of text difficulty was not a significant predictor of ARHQ (p = 0.504). This TReAT-Q model also accounted for a significant amount of variance (F(11, 192) = 12.24, R2 = 0.38, p < 0.001). The combined NFC + TReAT-Q model, demonstrated that NFC was no longer a significant predictor of ARHQ when the four TReAT-Q subscales were included in the model (Table 4). This combined model accounted for more variance in ARHQ than the NFC model (F(4, 191) = 15.46, p < 0.001; ΔR2 = 0.19). There was no significant difference between the TReAT-Q model compared to the combined NFC + TReAT-Q model (F(1, 191) = 0.57, p = 0.45; ΔR2 < 0.01).

Table 4 Regression models predicting ARHQ subset scores
Full size table

Summary

Study 1 developed and tested a measure of readers’ approaches to text, TReAT-Q. Our initial hypothesis was that three categories related to readers’ approaches to text would emerge as TReAT-Q factors: reading goals, desire to understand, and comprehension monitoring strategies. Instead, analyses suggested that four factors would best describe the data, labeled intrinsic goals, extrinsic goals and strategies, effort at understanding, and avoidance of text difficulty. The results demonstrated that three of the four TReAT-Q subscales, i.e., all but avoidance of text difficulty were good predictors of self-reported reading experience, measured via a subset of the Adult Reading History Questionnaire (ARHQ.) TReAT-Q’s predictive value was beyond that provided by the need for cognition (NFC), a more general self-report measure of preference for engaging in complex cognitive tasks. Additionally, all subscales had good internal consistency (Cronbach’s alpha between 0.85 and 0.91).

Scores on TReAT-Q generally predicted readers’ self-reported reading experiences for a heterogenous adult sample, with participants spanning a variety of educational and racial backgrounds. An additional question is whether TReAT-Q predicts reading comprehension skill. The second study addresses this question with a university sample and assesses the test–retest reliability of TReAT-Q.

Study 2

Study 2 had two goals. First, we tested a cognitive model of reading comprehension by examining the interrelations among reading experience, vocabulary knowledge, and reading comprehension. Following the implications of the RSF, we propose that vocabulary knowledge plays a central role in linking reading experience and comprehension and test whether reading experience influences reading comprehension via vocabulary knowledge. Second, we extend the cognitive model to include readers’ approaches to text (as measured by TReAT-Q) and the applicability of TReAT-Q by assessing its relation to reading comprehensions skill and reading experience in a university sample. We used an SEM that includes measures of reading comprehension and experience as dependent measures and TReAT-Q factors—intrinsic goals, extrinsic goals and strategies, effort at understanding, and avoidance of reading difficulty— as predictors of reading experiences and reading comprehension skill. Using the RSF as a guide for examining these interrelationships, we test whether reading experience predicts both vocabulary knowledge and reading comprehension (Cunningham & Stanovich, 1997; Stafura & Perfetti, 2014) and, using mediation analysis, whether any effects of reading experience and readers’ approaches to text on reading comprehension are mediated by vocabulary knowledge.

Third, if TReAT-Q captures a stable individual attribute, we should expect a reasonably high test–retest reliability; accordingly, we assessed the subscales twice at an interval of approximately 2 weeks. Finally, because Study 1 and Study 2 differ in populations (i.e., a more heterogeneous community sample vs. a university sample), we also conducted a post hoc analysis to examine potential population differences on TReAT-Q subscale scores as well as the interrelationships among the subscales.

Method

Participants

One hundred eighty-seven participants were recruited through the University of Pittsburgh’s psychology research participant pool and received course credit for their participation. Informed consent was obtained from all participants. All participants self-identified as native English speakers (23 indicated speaking a second language), with no psychological or learning disabilities, and with normal or corrected-to-normal vision. Ages ranged from 18– 56 (M = 19.09; SD = 3.43) across 64 male and 123 female participants: 177 (77.01%) White, 13 (6.95%) Black or African American, three (1.60%) Hispanic or Latino, and 27 (14.44%) Asian. These participants were included in another study (Calloway, 2019; Studies 2 and 3) in which they completed questionnaires and tasks related to reading comprehension along with six additional tasks related to decoding and listening comprehension. However, because Study 2 focuses on high-level reading comprehension, only the relevant measures are presented here.

Measures

Participants completed the 31 items of TReAT-Q that were retained from the PAF analyses and the same two questionnaires of NFC and ARHQ subset as in Study 1. Participants also completed assessments of reading comprehension, vocabulary, and print exposure (as an additional measure of reading experience).

TReAT-Q

TReAT-Q in Study 2 consists of the 31 out of 87 items that were retained from the factor analysis reported in Study 1, where they loaded on one of four factors: (1) Intrinsic goals; (2) Extrinsic goals and strategies; (3) Effort at understanding; and (4) Avoidance of text difficulty. In the original version, all seven items in the avoidance of text difficulty subscale were reverse-scaled; in order to create a more balanced distribution of reverse-scaled items, we reduced the number of reverse-scaled items to four and modified the other three subscales to have at least two reverse-scaled items (reverse-scaled items = 11; Appendix D). Participants rated whether they agreed that a statement was characteristic of them on a Likert scale ranging from Strongly disagree (– 3) to Strongly agree (3). The second administration of TReAT-Q occurred approximately two weeks after the first administration and, to reduce the likelihood of participants responding based on familiarity from previous responses, the location of the scale anchors was reversed and ranged from Strongly agree to Strongly disagree (3 to – 3).

Reading experience: Author Recognition Test

As a more objective measure of reading experience that is less susceptible to response bias (Moore & Gordon, 2015), participants completed three versions of the author recognition test (ART), a measure of print exposure. Furthermore, whereas the ARHQ subset is a self-report questionnaire that includes recent reading activities, the ART is more sensitive to long-term reading experiences. In the ART, participants viewed a list of names and selected names they recognized as authors. ART-1989 (Stanovich & West, 1989) and contained 50 authors and 30 foils. ART-2002 is a modified version of Stanovich and West’s ART (Perfetti & Hart, 2002) with 40 authors and 40 foils. ART-2008 was created by Acheson et al. (2008) with 65 authors and 65 foils. Across the three tests, six authors that appeared in ART-1989 also appeared in ART-2002, four authors that appeared in ART-2002 appeared in ART-2008, and 23 authors that appeared in ART-1989 also appeared in ART-2008. The three versions were administered as a single assessment; however, scores were calculated separately for each version. Participants were given a list of 261 names, and across the three lists, there were 126 authors and 135 foils. The foils were names of famous singers, actors, politicians, athletes (ART-2002), and psychology researchers (ART-1989 and ART-2008). Scores for each version were calculated as the number of authors identified correctly minus incorrectly. Score ranges for each version are as follows: ART-1989 – 30 to 50; ART-2002 – 40 to 40; ART-2008 – 65 to 65.

Reading comprehension assessments

Participants completed the Gates–MacGinitie reading comprehension test (Level AR; MacGinitie et al., 2000). Two versions of the test (Forms S and T, randomized across participants) each contained 11 passages with 48 multiple-choice questions with four choices. Twenty questions tested literal comprehension and 28 tested inferential comprehension, separately scored as percent correct. Participants had a 35-min time limit, as designated by the test manufacturers, and they could not refer to the passages when answering questions.

Participants also completed the comprehension component of the Nelson–Denny test (Form G; Brown et al., 1993), consisting of seven passages and 38 multiple-choice questions with five answer options. Readers could refer to the passage as they answered questions and had 20 min to complete the task, as designated by the test manufacturers. Half of the questions assessed literal comprehension and half assessed inferential comprehension, separately scored as percent correct.

Vocabulary assessments

Participants also completed three vocabulary measures: the vocabulary portion of the Gates–MacGinitie test (MacGinitie et al., 2000), the Nelson–Denny test (Form G; Brown et al., 1993), and the Peabody Picture Naming Vocabulary Test (PPVT-4; Dunn & Dunn, 1981). The vocabulary section of the Gates–MacGinitie test consisted of 45 multiple-choice items with five options each. Participants selected the word or phrase that best described an underlined word given in a prompt within a 20-min time limit. The maximum score for this test was 45. The Nelson–Denny vocabulary test has 80 multiple-choice items with five options each. Participants read a prompt and selected the word or phrase that described an italicized word within a 15-min time limit. The maximum score for the test was 80. The PPVT-4 required participants to select a picture, from an array of four, that described a spoken word. The task included 19 sets of 12 items each, which increased in difficulty across sets. Participants were assigned a starting point according to their age, 17–18 (set 13) or 19 + (set 14). Raw scores were calculated by subtracting the ceiling item number (i.e., the item number from the most difficult set with fewer than eight errors) from the total number of errors. Raw scores were then converted to standardized scores (M = 100) to account for age and ceiling item differences.

Procedure

Participants completed the assessments over the course of three in-person sessions in the laboratory. Each participant completed tasks separate rooms at computer stations. Session 1 (1.5–2 h): TReAT-Q, ARHQ subset, and ART questionnaires, via Qualtrics (Qualtrics, Provo, UT), and the Gates–MacGinitie comprehension and vocabulary assessments on a computer. Session 2 (approximately 1 h), 1–4 days later: computerized versions of the Nelson–Denny comprehension and vocabulary assessments, the PPVT-4, and four other assessments used for a different study (Calloway, 2019; Study 2). Session 3 (approximately 1 h), 14–21 days after Session 1: TReAT-Q via Qualtrics for the second time and computerized reading and listening tasks assessed for a separate study (Calloway, 2019; Study 3).

Analysis procedure

Of the 187 participants, 178 completed assessments during both Sessions 1 and 2, thus providing data for the main analyses. Cronbach’s alpha indexed reliability for all measures, except the PPVT-4, for which split-half reliability was calculated by the psych package in R (Revelle, 2020). When answers are dichotomous (i.e., correct vs. incorrect), the formula to calculate Cronbach’s alpha defaults to Kuder–Richardson formula 20 (KR-20; Revelle, 2020). Test–retest reliability of TReAT-Q over a 14- to 21-day time span was assessed via Pearson’s correlations between scores at Time 1 and Time 2 (addressing Hypothesis 3), along with the internal consistency of each subscale at Time 1 and Time 2.

Next, confirmatory factor analyses (CFAs) were used to assess a measurement model using the lavaan package in R (Rosseel, 2012). Separate measurement models of each questionnaire (TReAT-Q, NFC, and ARHQ) were conducted to ensure that the questionnaires, especially TReAT-Q, corresponded to their hypothesized constructs. Next an overall measurement model (CFA) was conducted, with latent factors of reading comprehension, vocabulary knowledge, reading experience, need for cognition, and approaches to text (as measured via TReAT-Q). The CFA was implemented to test whether the hypothesized indicator variables loaded significantly onto their latent factors, by testing regressions from latent factors to observed variables and covariances among all latent factors. Importantly, the latent factor of approaches to text will demonstrate that TReAT-Q subscales contribute to an overarching factor of approaches to text for this university sample. For factor analyses, non-significant χ2 values indicate good fit. However, because χ2 is sensitive to sample size and parameter ratios, the root mean square error of approximation (RMSEA) and Tucker–Lewis Index (TLI) were also included to assess model fit. TLI values > 0.90 indicate acceptable fit (Sharma et al., 2005) and values > 0.95 indicate good model fit (Hu & Bentler, 1999). RMSEA values < 0.08 and < 0.05 indicate acceptable and good fit, respectively (Hu & Bentler, 1999).

To test the relationships among TReAT-Q, reading experience, and high-level reading comprehension skills, we tested three SEMs. The first SEM (Cognitive model) addressed Hypothesis 1 and tested whether reading experience predicts vocabulary knowledge (e.g., Cunningham & Stanovich, 1997) and reading comprehension (e.g., Acheson et al., 2008), as predicted by the RSF (Stafura & Perfetti, 2014; Perfetti & Helder, in press). The Cognitive model also included a mediation analysis, assessing whether reading experience had an effect on reading comprehension mediated through vocabulary knowledge. The second model (Cognitive + approaches to text model) built on the first model to test Hypothesis 2 that TReAT-Q predicts reading comprehension and reading experience. Building on findings from Study 1 that TReAT-Q predicted self-reported reading experiences, we used a serial mediation model to test the indirect effects of TReAT-Q on reading comprehension mediated via reading experience and vocabulary knowledge (i.e., both reading experience and vocabulary knowledge serve as mediators). NFC was regressed on reading comprehension as a control variable. The final SEM contained only significant regressions from the Cognitive + approaches to text model. Finally, a post hoc regression analysis comparing scores on each of the TReAT-Q subscales between Study 1 and Study 2 was implemented to examine potential differences between the samples (i.e., a community vs. university sample).

Results

Descriptive statistics

Table 5 shows the descriptive statistics for each measure and correlations among them. Before reporting the structure equation model results, we report reliability statistics for the various measures.

Table 5 Descriptive statistics and correlations for Study 2 measures for participants who completed Sessions 1 and 2
Full size table

Reliability indicators

Cronbach’s alpha indexed the internal consistency of each measure, with values > 0.70 indicating good internal consistency and values > 0.60 as adequate internal consistency. TReAT-Q had good overall internal consistency (α = 0.88, 95% CI = 0.86 to 0.90) and good internal consistency for two of the subscales: intrinsic goals (α = 0.90, 95% CI = 0.88 to 0.92), and avoidance of text difficulty (α = 0.78, 95% CI = 0.73 to 0.83); extrinsic goals and strategies (α = 0.63, 95% CI = 0.54 to 0.71) had adequate internal consistency; effort at understanding (α = 0.59, 95% CI = 0.50 to 0.68) did not reach adequate internal consistency. The test–retest reliability of the entire TReAT-Q with its four subscales for the 157 participants who returned for Session 3 was good (r = 0.78, p < 0.001). The test–retest reliability of each subscale is as follows: intrinsic goals (r = 0.78, p < 0.001), extrinsic goals and strategies, (r = 0.60, p < 0.001), effort at understanding (r = 0.60, p < 0.001), and avoidance of text difficulty (r = 0.71, p < 0.001).

The overall internal consistencies for the NFC questionnaire (α = 0.90, 95% CI = 0.87 to 0.92) and ARHQ subset (α = 0.67, 95% CI = 0.59 to 0.74) were good and acceptable, respectively. Internal consistency was calculated separately for each of the three versions of the Author Recognition Test. ART-1989, had an internal consistency of α = 0.55, 95% CI = 0.46 to 0.64), ART-2002 had an internal consistency of α = 0.72, 95% CI = 0.67 to 0.78, and ART-2008 had an internal consistency of α = 0.81, 95% CI = 0.78 to 0.85.

The internal consistencies for the reading and vocabulary tests were as follows: Gates–MacGinitie reading test: Form S (n = 87), α = 0.70, 95% CI = 0.61 to 0.79; Form T, (n = 91) α = 0.73, 95% CI = 0.65 to 0.81; Nelson–Denny reading test (Form G), α = 0.78, 95% CI = 0.73 to 0.82; Gates–MacGinitie vocabulary test, α = 0.76, 95% CI = 0.71 to 0.81; Nelson–Denny vocabulary test, α = 0.80, 95% CI = 0.75 to 0.84. Because the PPVT-4 test items vary with age and with progress in the item sequence, split-half reliabilities were used to test reliability across non-standardized raw summed scores of even (M = 28.87, SD = 4.57) and odd (M = 28.79, SD = 4.68) items. These tests produced r = 0.78, p < 0.001, standard error of measurement [SEm] = 2.19.

Confirmatory factor analyses (CFA) of questionnaires

Before running a full measurement model, we first ran separate CFAs for each Likert-item questionnaire (i.e., NFC, reading history, and TReAT-Q) to ensure that each questionnaire had good fit to enter into the full measurement model. Modification indices were used to determine if similarly worded items would benefit from having their residuals covary. Model fit statistics are presented in Table 6.

Table 6 CFAs for NFC, ARHQ subset, and TReAT-Q
Full size table

For the NFC CFA, two latent variables were included, one for positively worded items and another for negatively worded items: Previous studies found differences between the two item types (Hevey et al., 2012; Soubelet & Salthouse, 2017). This model had good fit (Table 6). Residuals were allowed to covary between items 11 and 12, 14 and 18, 4 and 11, and 3 and 17 (see, Appendix B for items). The ARHQ subset z-scored responses were used as indicators onto an overall ARHQ latent factor. After allowing residual errors to covary among all newspaper/magazine-related items and between reading amount and books read for pleasure, the model had good fit. For the TReAT-Q CFA, all four subscales were included as latent factors with individual items as indicator variables. The model had poor fit. To investigate the source of poor model fit, separate CFAs were conducted for each TReAT-Q subscale. The intrinsic goals, extrinsic goals and strategies, and avoidance of text difficulty latent factors all had good model fit; however, effort at understanding had poor model fit. Thus, for the full measurement model CFA, effort at understanding was not included.

Confirmatory factor analysis (CFA) of the measurement model

The CFA consisted of five latent factors: (1) Reading Comprehension, (2) Vocabulary Knowledge, (3) Approaches to Text, (4) Reading Experience, and (5) Need for Cognition. The Reading Comprehension latent factor included responses to inferential and literal questions from the Gates–MacGinitie and Nelson–Denny reading tests, as defined by the publishers of the tests. Error terms were allowed to covary because the questions could refer to the same text. The Vocabulary Knowledge latent factor included the Nelson–Denny Vocabulary test, Gates-MacGinitie Vocabulary test, and the PPVT-4. The Reading Experience latent factor included the three versions of the ART (1989, 2002, and 2008) and the ARHQ subset. The Need for Cognition latent factor included negatively and positively worded items.

Although the χ2 was significant, the remaining fit indices suggested that the measurement model estimated by the CFA had good fit to the data χ2(92) = 132.81, p < 0.003, CFI = 0.97, TLI = 0.95, RMSEA = 0.05 (90% CI: 0.03 to 0.07). All indicators significantly loaded onto their latent factors (ps < 0.001; Table 7) and all latent factors were significantly associated with one another (ps < 0.005), with covariances ranging between 0.30 to 0.86.

Table 7 Measurement model
Full size table

Full SEMs

Three SEMs were estimated: Cognitive only, initial Cognitive + approaches to text, and final Cognitive + approaches to text (which retained only significant regressions among latent factors).

The cognitive model (reading experience, vocabulary knowledge, and reading comprehension)

The first SEM tested the regression effects of reading experiences on vocabulary knowledge and vocabulary knowledge on reading comprehension, as suggested by the relation between the word meaning and comprehension subsystems of the RSF. A Reading Experience latent factor was regressed on both Vocabulary Knowledge and Reading Comprehension latent factors. The model also included a mediation pathway, testing an indirect effect of Reading Experience on Reading Comprehension via Vocabulary Knowledge. The overall Cognitive model had good fit to the data, (χ2(39) = 38.27, p = 0.503, CFI > 0.99, TLI > 0.99, RMSEA < 0.001 (90% CI: 0.00 to 0.05). In this model, the regression of Vocabulary Knowledge on Reading Comprehension and the regression of Reading Experience on Vocabulary Knowledge were both significant (Fig. 2). The indirect effect of Reading Experience on Reading Comprehension, mediated through Vocabulary Knowledge, was also significant (indirect effect = 0.68, z = 5.35, p < 0.001); however, the direct effect of Reading Experience on Reading Comprehension was not significant (direct effect = –0.21, z = – 1.79, p = 0.073).

Fig. 2
figure 2

SEM with reading experience, vocabulary, and reading comprehension. ND = Nelson–Denny; GM = Gates–MacGinitie; ART = Author recognition test; ARHQ = Adult reading history questionnaire; Inf = inference questions; Lit = literal questions. Significant factor loadings and regressions are in black and non-significant effects are in grey dashed lines

Full size image

Adding approaches to text to the cognitive model

Two subsequent Cognitive + approaches to text models included latent factors for (1) Approaches to text and (2) NFC. The Initial Cognitive + approaches to text model built on the Cognitive model (Fig. 2) and Approaches to text and NFC were regressed on Reading Comprehension. Approaches to Text was also regressed onto Reading Experience, and indirect effects on Reading Comprehension mediated by Reading Experience and Vocabulary Knowledge were tested. The Final Cognitive + approaches to text model removed any non-significant regressions from the Initial Cognitive + approaches to text model.

The Initial Cognitive + approaches to text model had adequate fit (χ2(95) = 137.80, p = 0.003, CFI = 0.97, TLI = 0.96, RMSEA = 0.05 (90% CI: 0.03 to 0.07)) and only Vocabulary Knowledge had a significant direct relationship with Reading Comprehension (Fig. 3). A serial mediation analysis revealed that Approaches to Text had an indirect effect on Reading Comprehension, mediated via Reading Experience and Vocabulary Knowledge (indirect effect = 0.31, z = 3.78, p < 0.001). The indirect effect of Approaches to Text on Reading Comprehension mediated only via Reading Experience was not significant (indirect effect = – 0.11, z = – 1.80, p = 0.073). As in the Cognitive model, the indirect effect of Reading Experience on Reading Comprehension mediated via Vocabulary Knowledge was significant (indirect effect = 0.68, z = 5.30, p < 0.001). NFC did not significantly predict Reading Comprehension (Fig. 3).

Fig. 3
figure 3

SEM with Approaches to text, NFC, Vocabulary Knowledge, Reading Experience, and Reading Comprehension. ART = Author recognition test. NFC = need for cognition; ND = Nelson–Denny; GM = Gates–MacGinitie; PPVT-4 = Peabody Picture Naming Task, version 4; ARQH = Adult reading history questionnaire; Inf = inference questions; Lit – literal questions. Non-significant factor loadings and regressions are in grey dashed lines

Full size image

In the Final Cognitive + approaches to text model, non-significant direct effects of Approaches to text, Reading Experience, and NFC on Reading comprehension were removed. The removal of the non-significant direct effects produced a good fit to the data (χ2(71) = 100.33, p = 0.013, CFI = 0.97, TLI = 0.97, RMSEA = 0.05 (90% CI: 0.02 to 0.07); Fig. 4). The mediation effect of Reading Experience on Reading Comprehension via Vocabulary Knowledge remained significant (indirect effect = 0.68, z = 5.34, p < 0.001). The serial mediation effect of Approaches to textFootnote 3 on Reading Comprehension, with both Reading Experience and Vocabulary Knowledge as mediators, also remained significant (indirect effect = 0.31, z = 3.76, p < 0.001).

Fig. 4
figure 4

Final Cognitive + approaches to text SEM with Approaches to text, Vocabulary Knowledge, Reading Experience, and Reading Comprehension. ART = Author recognition test; ND = Nelson–Denny; GM = Gates-MacGinitie; PPVT-4 = Peabody Picture Naming Task, version 4; ARQH = Adult reading history questionnaire; Inf = inference questions; Lit – literal questions

Full size image

Summary

Overall, the findings from the CFA (i.e., measurement model) and SEM demonstrate that three of the four TReAT-Q subscales of (intrinsic goals, extrinsic goals and strategies, and avoidance of text difficulty) significantly contribute to the readers’ approaches to text construct, largely supporting findings TReAT-Q results from the community sample in Study 1. Unlike Study 1, however, the effort at understanding subscale did not significantly contribute to the TReAT-Q latent factor. Additionally, the measurement-model CFA showed a positive association between approaches to text and reading comprehension (cov = 0.30, p = 0.005). The final SEM showed that higher TReAT-Q scores were associated with greater reading experience, which in turn predicts higher vocabulary knowledge and reading comprehension ability. Although the more general measure of need for cognition had a positive relationship with reading comprehension in the measurement-model CFA (cov = 0.33, p = 0.002), it did not predict reading comprehension in the SEM. Measures of vocabulary knowledge, readers’ approaches to text, and reading experience overshadowed effects of need for cognition which may have otherwise emerged.

The internal consistencies for TReAT-Q subscales were lower in this university sample compared to the community sample in Study 1; however, university students generally showed good test–retest reliability for the TReAT-Q, suggesting readers’ approaches to text are relatively stable over a two-week period, even when the anchors of the scales are reversed. To better understand potential differences between the two samples, we tested differences between samples on each TReAT-Q subscale in a post hoc analyses.

Population differences on the TReAT-Q

In separate regression analyses comparing scores for each TReAT-Q subscale between the community (Study 1) and university (Study 2; Time 1) samples, Age, Gender, Race/Ethnicity (White vs. non-White), and NFC were included as control variables. The intercept was the mean across groups. To control for multiple comparisons, p-value significance is based on the Bonferroni corrected p-value of p = 0.00208. The regressions revealed that university participants scored lower on subscales of intrinsic goals (β = – 4.75, t(376) = – 4.02, p < 0.001) and avoidance of text difficulty (β = – 17.82, t(376) = – 18.74, p < 0.001). The university sample scored higher on the effort at understanding subscale than the community sample (β = 11.28, t(376) = 12.65, p < 0.001) and there was no difference between the two samples on the extrinsic goals and strategies subscale (β = – 0.66, t(376) = – 0.55, p = 0.581). Table 8 shows the descriptive statistics for each study and indicates whether there were mean and/or variance differences between the two samples.

Table 8 Community (Study 1) and university sample (Study 2) TReAT-Q differences
Full size table

Overall, the findings demonstrate that TReAT-Q is sensitive to population differences in how readers approach texts, suggesting that university students are less intrinsically motivated to read, avoid difficult texts more often, but apply more effort at understanding (although this subscale did not have good internal consistency for the university sample). It should be noted that differences in the interrelationships among the subscales and the scores themselves may be a combination of population differences and differences in the number of items participants responded to. The community sample completed the initial 87-item TReAT-Q that was then pared down to the final 31-item TReAT-Q completed by the university sample.

Discussion

The goals of the present studies were to introduce and measure the construct of approaches to text and to establish a link between approaches to text and cognitive aspects of reading. We developed The Readers’ Approaches to Text Questionnaire, TReAT-Q, and tested it with a community sample. This produced a measure with four subscales of intrinsic reading goals, extrinsic goals and strategies, effort at understanding, and avoidance of text difficulty. TReAT-Q predicted reading experience in this community sample. A second study with university students tested a SEM of reading comprehension that suggested the following conclusions: (1) vocabulary knowledge mediates the relationship between reading experience and reading comprehension, (2) readers’ approaches to text directly influences readers’ reported frequency and amount of reading, along with print exposure, and (3) readers’ approaches to text indirectly influences reading comprehension via vocabulary and reading experience.

Components of readers’ approaches to text

Intrinsic reading goals

The Readers’ Approaches to Text Questionnaire (TReAT-Q) developed for the present studies shares some similarity to existing motivation scales (Henk & Melnick, 1995; Schutte & Malouff, 2007; Wigfield & Guthrie, 1997); the factor analysis with the community sample (Study 1) identified intrinsic reading goals as one of the components of readers’ approaches to text and was confirmed in a university sample (Study 2). Readers who enjoy reading reported engaging in more reading activities and reported having more reading experiences, aligning with more general intrinsic motivation research finding that people tend to seek out activities that they enjoy (Ryan & Deci, 2020). Accordingly, intrinsic reading motivation has been a key component of various existing motivation scales (e.g., Davis et al., 2018; Schutte & Malouff, 2007; Wigfield & Guthrie, 1997) and reading interventions (Wigfield et al., 2008). The other major component of reading motivation, extrinsic motivation, was also revealed as a component of readers’ approaches to text.

Extrinsic reading goals and strategies

The extrinsic reading goals and strategies factor of the TReAT-Q included items about reading strategies associated with extrinsic goals of learning for work or school, such as highlighting, underlining, and making outlines that predicted readers’ reading experiences (Study 1) and indirectly predicted reading comprehension (Study 2). Whereas extrinsic goals and strategies were uncorrelated with the other three components of readers’ approaches to text for the community sample, it was related to all other components in the university sample, despite no differences between the two samples on the extrinsic goals and strategies subscale (Table 7). This aligns with findings from Schutte and Malouff (2007) who also found differences between community and university samples; a university sub-sample, but not a community sub-sample, showed a positive relationship between reading to do well in other realms (e.g., “I read to improve my work or university performance.”) and self-reported hours of required reading. Because much of a university student’s reading activities are determined by course assignments, perhaps how much the student enjoys reading and avoids reading difficult texts is closely linked to how motivated the student is to achieve these external reading goals needed for academic success. If motivation for an activity lies on a continuum (Howard et al., 2017; Ryan & Deci, 2020; Schutte & Malouff, 2007), intrinsically and extrinsically motivated reading goals may be more closely integrated for university students, whereas such motivations may be more crystalized and further separated for adults who have reached the end of their formal education careers. While intrinsically and extrinsically motivated reading goals relate to the why of readers read, the final two subscales, effort at understanding and avoidance of text difficulty relate to how individuals approach texts.

Effort at understanding

The effort at understanding subscale taps into the degree to which readers attempt to adopt relevant reading strategies needed for a desired level of text understanding. Our TReAT-Q ultimately did not include items that directly reflected the comprehension failure detection component of comprehension monitoring; however, the effort at understanding subscale demonstrates that readers do self-report their general use of comprehension monitoring strategies that influence their approach to texts (e.g., “If a text is difficult, I often re-read sections of the text.”). Readers may be better at noticing the strategies that result from comprehension monitoring than noticing the act of comprehension monitoring itself. Indeed, much research on comprehension monitoring assesses it via inconsistency detection, indexed by slowing down reading when encountering a break in coherence break (e.g., Albrecht & O’Brien, 1993; Connor et al., 2015; Kim et al., 2018; O’Brien et al., 1998), asking participants to indicate if there was an inconsistency in a passage (Cain et al., 2004; Oakhill & Cain, 2012), or both (Currie et al., 2021; Helder et al., 2016; Zabrucky & Ratner, 1992). Thus, effort at understanding reflects the deployment of reader-initiated strategies, caused by a comprehension failure, to maintain coherence and aid text comprehension.

Whereas effort at understanding was a subscale for the community sample, it had poor internal consistency and did not significantly load onto the approaches to text construct for the university sample (Table 6). This difference in results between the university sample and the community sample may reflect differences in the variability of their effort at understanding scores. For the university sample, the mean effort at understanding score was higher and the variability was lower compared with the community sample: university sample (M = 14.4, SD = 5.49, range – 4 to 21), community sample (M = 3.37, SD = 9.01, range – 21 to 21). The relative homogeneity of the university sample may have led to reduced ranges and standard deviations, both of which have been shown to reduce Cronbach’s alpha (Amirrudin et al., 2021). Because much of a university student’s reading is related to coursework needed to meet degree requirements and because most students perceive some effort being involved to succeed, the items on the effort at understanding subscale may not be as reliable for distinguishing university readers.

Avoidance of text difficulty

Avoidance of text difficulty taps into whether readers would rather approach or avoid challenging texts (e.g., “I’d rather read easy texts than difficult ones.”), which may influence how readers use reading strategies and increase readers’ familiarity with various syntactic structures and lexical items. Reading difficult texts requires sufficient comprehension monitoring (Connor et al., 2015) and sometimes results in more effortful reading (Clifton et al., 2007; Frazier & Rayner, 1982; van den Broek et al., 2001). Findings from the present studies highlight this link between the reader’s approach to difficulty and enhanced monitoring and comprehension strategies; readers who report reading more difficult texts are more likely to deploy effortful reading strategies to maintain coherence. An additional link may go from accepting challenging texts to growth in vocabulary and comprehension. Consistent with this linkage is evidence that reading texts slightly above one’s reading level leads to better vocabulary and comprehension gains than texts that are too easy (e.g., Carver, 1994; Fulmer et al., 2015). Our text difficulty conclusions align with other studies that included reading challenge or work avoidance as components of motivation assessments (Schutte & Malouff, 2007; Wigfield & Guthrie, 1997).

Approaches to text in a cognitive model of reading comprehension

Given the four components of readers’ approaches to text that emerged, we examined how these approach components might connect to reading components within a cognitive model of reading comprehension. We first tested a cognitive-only model of reading comprehension in an SEM that revealed higher vocabulary knowledge was strongly related to higher reading comprehension ability, consistent with the assumption that vocabulary knowledge is a key component of reading comprehension (Joshi, 2005; National Reading Panel, 2000). For successful comprehension, readers must know a majority of the words they read (Moghadam et al., 2012; Schmitt et al., 2011; Stahl et al., 1991). Furthermore, in the cognitive-only model, vocabulary knowledge fully mediated the relationship between reading experience and reading comprehension. This conclusion, although presumably too simple to capture the full effects of reading experience, is consistent with theory and experimental results: Experiencing a word across multiple contexts helps strengthen its lexical representation at the word-form and semantic level (Braze et al., 2007; Jenkins et al., 1984; Nation, 2017; Nation & Snowling, 1998; Perfetti & Hart, 2001, 2002). In one version of this idea, the lexical legacy hypothesis (Nation, 2017) postulates that lexical items encountered across different reading experiences enhances the quality of their lexical representations (Perfetti & Hart, 2001). Related studies have found that print exposure, a stand-in for reading experience, predicts higher vocabulary knowledge (Martin-Chang & Gould, 2008; Moore & Gordon, 2015).

The RSF suggests that general knowledge—which can be gained through reading experiences (Stanovich & Cunningham, 1993)—has direct, reciprocal relationships with both vocabulary knowledge and reading comprehension. However, in the present study reading experience did not predict reading comprehension after accounting for vocabulary knowledge. One possibility for the lack of a direct relationship between reading experience and reading comprehension is related to the stage of reading development in adults and the inclusion of vocabulary as a mediator. Whereas children are continuing to learn about text structures and develop skills related to reading fluency, which often improve with additional experience (e.g., Nation, 2017; Torppa et al., 2020), older readers tend to have more solidified knowledge of syntactic structures and higher reading fluency (Chall, 1987). Thus, for college-level adult readers, vocabulary knowledge may the best remaining candidate for reading-related gains. Although studies with adults have also shown print exposure to predict reading comprehension (e.g., Acheson et al., 2008; Martin-Chang & Gould, 2008), they generally do not focus on the mediating role of vocabulary knowledge (Cunningham & Stanovich, 1997; Martin-Chang & Gould, 2008; Schiefele & Schaffner, 2016; Torppa et al., 2020). In contrast, our study highlights the central role of vocabulary knowledge and suggests that additional research is needed to further investigate this mediation relationship.

Our final Cognitive + approaches to text SEM showed an indirect effect of approaches to text (with three subscales) on reading comprehension, mediated by reading experience and vocabulary knowledge. Previous studies reporting associations between motivation factors and reading experiences (Davis et al., 2018; Parault & Williams, 2009; Wigfield & Guthrie, 1997) found that children with higher motivation for reading read more frequently and read more books than those with lower reading motivation. Interestingly, in our study, neither readers’ approaches to text nor reading experience directly predicted reading comprehension, aligning with other research using factor analyses to examine the relation of reading motivation to amount of reading (Schutte & Malouff, 2007; Wigfield & Guthrie, 1997) and reading comprehension (e.g., Schaffner et al., 2013; Wang & Guthrie, 2004). Assessing reading comprehension by standardized texts may provide a measure that is too coarse-grain to reflect more specific comprehension components that are affected by experience—for example, the use of global text features (genre cues, structure cues, graphics), writing styles, less common syntactic structures.

Our interpretation of our Cognitive + approaches to text model is that how readers approach texts does not directly influence reading comprehension; instead, the approach to text influences the degree to which readers actively engage with texts they read, the amount of reading experiences they have and, consequently, how much they gain from their reading experiences. In the present studies, these gains are reflected by higher vocabulary knowledge for those with more reading experiences. Examining other elements necessary for comprehension such as syntactic structure, background knowledge, and morphological knowledge may also shed light on how readers’ approaches to text influence the knowledge gained from reading experiences.

Additionally, future studies could more closely examine the longitudinal relationship among approaches to text, reading experiences, vocabulary knowledge, and reading comprehension in adults. Differences among demographic groups might show different trajectories and it is possible that the overall measure represents slightly different constructs for males compared to females. Further, community samples and university samples may differ in the role that reading plays, as suggested by the results that effort at understanding emerged as a factor in the approaches-to-text construct for the community sample, but not the university sample.

Conclusions

In summary, the present studies demonstrated that a 31-item self-report questionnaire of readers’ approaches to text, as measured via TReAT-Q, was positively associated with reading experience, which indirectly related to reading comprehension. Readers who have positive approaches to text appear to be those who enjoy reading and read challenging texts that require the deployment of appropriate reading strategies to reach a criterion level of understanding. Examining readers’ approaches to text within a cognitive model of reading comprehension provides insight into whether readers have greater reading experiences that are linked to increased vocabulary knowledge, and consequently, reading comprehension. Exactly how readers’ approaches to text are predictive of reading comprehension for specific reading situations (e.g., for leisure vs. work) must still be examined. However, TReAT-Q provides a promising outlook for examining readers’ approaches to textsuch as reading challenging material, being intrinsically motivated to read, and using strategies that promote learning. Although sample homogeneity should be considered, TReAT-Q offers measures of adult reading motivation, effort, and avoidance of text difficulty, all of which should assist in studying reading comprehension in adults.