Sage Journals: Discover world-class research

Abstract

This study explores the similarities and differences in reading strategies applied by students to Chinese versus English reading. 842 students responded to a reading strategy inventory in terms of their habits when reading in Chinese versus in English. The result validates the general structure of a reading strategy inventory originally designed for English. The general scale seemed reasonable for Chinese reading as well as English reading. However, gender differences were found on some items. In addition, there was no statistically significant correlation between strategy use and test scores, which is inconsistent with previous research. Besides, we found a strong correlation between English and classical Chinese reading scores, but not between English and modern Chinese reading. Language as L1 or L2, curriculum and restricted test formats may partly explain the low correlations while social and educational environment can contribute to the high correlation observed. This study also serves as a rare example of a complete cycle of invariance test for ordinal data with a repeated design in Mplus. Implications for research and pedagogy were discussed.

Plain language summary

Although there are common underlying structures that can explain reading behavior, practice in context can vary. It is necessary to consider the local milieu and gender differences to better serve students for foreign language education. This study explores the similarities and differences in strategies by students when reading in English as a foreign language versus in Chinese as a native language. We evaluated an established strategy scale in context, validated the scale structure, and then explored the language and gender factors. Results supported the general structure of the scale but also revealed gender differences on some items. No significant correlation was found between strategy use and test scores, which is inconsistent with previous research. The correlation between English reading scores and two types of Chinese reading scores is different. There is no significant relationship between English and modern Chinese reading, but a significant one between English and classical Chinese reading. Cultural and pedagogical traditions may be part of the reason, and attention to individual differences is suggested. This study uses a convenient sample, which limits the generalizability of specific results. However, it provides a rare example with a complete cycle of scale validation procedures for ordinal data based on an up-to-date procedure in Mplus. Example syntax is provided.

Keywords

language education reading strategies measurement invariance strategy transfer gender

Introduction

Reading comprehension requires an array of processes where large amounts of information must be processed at both the word and text level (Stanovich, 1982a, 1982b). When readers’ knowledge is limited, text material is too difficult, or a reading task is complex, readers will have to use strategies to aid comprehension. This is widely acknowledged and true even for experienced readers (Graesser, 2007).

Due to its importance in rescuing non-native reading, strategy use has attracted much attention in foreign/second language teaching and research. Evidence generally supports a strong relationship between strategy use and reading comprehension (Brevik, 2017; Grabe, 2009; Muijselaar et al., 2017). Moderating factors such as gender and language proficiency have also been explored (Denton et al., 2015; Maghsoudi, 2022; Oxford, 1990). However, a consensus is not reached (e.g., Arabmofrad et al., 2021; Habók et al., 2019). There is an implication that not all strategies work equally well for all learners (e.g., Sun et al., 2021).

Learners’ first language and gender have been found to be partly responsible for these discrepancies (Cantrell & Carter, 2009). As Hofer and Sinatra (2010) have put it that theory in one culture may not be “productive” in another. Since concepts of major and influential reading strategies trace back to native English researchers, their applicability and validity in different contexts require more research. In this study, we focus on two factors, language (English as a foreign language vs. Chinese as the native language) and the gender of learners to explore the reading strategy construct.

Literature Review

Language Impact on Reading Strategy Use

There is rich evidence that first language (L1) influences reading in second language (L2). Research shows that experiences and habits in L1 play an important role (Sun et al., 2021; van Gelderen et al., 2004). Not only does proficiency and literacy in L1 matter, so do the reading strategies acquired (Bernhardt, 2011; Koda, 1993; Maghsoudi, 2022). Reading strategies can transfer between languages even when the linguistic distance is great, such as from Arabic, Japanese, or Chinese to the English language (Chiswick & Miller, 2005; Chuang et al., 2012; Nakada et al., 2001).

However, research dispels the unidirectional impact of L1 (Davis & Bistodeau, 1993). Clarke (1979) proposed a “short-circuit” effect to explain that a ceiling in L2 competence can prevent the complete transfer of skills from L1. When challenged with difficult tasks in L2, good L1 readers may revert to similar strategies as poor L1 readers. A recent study again consolidated this proposal between L1 and L2 English speakers (Feller et al., 2020). More interestingly, one study on “outlier” students implies that L1 impact may be less important to L2 reading compared to other factors such as interests and motivation. Thus, a person can be good in L2 reading because of motivation even though they are poor in L1 reading (Brevik & Hellekjær, 2018).

Strategy use across languages is also heterogeneous. For example, Lin and Yu (2015) found that participants used more diverse strategies for English reading than for Chinese reading. Researchers believe that the linguistic features of L1 and the learning experience of individuals can affect L2 strategic decisions (A. D. Cohen, 2011; Grabe, 1991; Sun et al., 2021). So do other factors, such as culturally determined literacy practice through schooling (Davis & Bistodeau, 1993; Taki, 2016). For example, Lau and Chen (2013) found that instruction had different effects on self-regulated learning for students depending on whether they were in Hong Kong versus Beijing.

In sum, reading strategies can vary due to linguistic or cultural features in L1 and L2 languages. When exploring strategy use, local context needs to be taken into consideration.

Gender Impact on Reading Strategy Use

While language reflects a cultural difference more broadly, gender exerts an influence at the individual level. Gender difference in reading strategy has been reported for both L1 and L2 learners, where results tend to support female dominance over males (Cantrell & Carter, 2009; Denton et al., 2015). Females are found to have more knowledge of reading strategies and they also use them more frequently (Bećirović et al., 2018; O’Reilly & McNamara, 2007). This is also true for online reading, contrary to the assumed disadvantage for girls (Cooper, 2006; J. Wu, 2014).

Despite these general patterns in findings, there is also evidence for the opposite. For example, Phakiti (2003) found that boys reported more metacognitive reading strategies than girls, and Tsai (2009) found that boys applied more control and procedural reading strategies than girls when reading online. Divergent results on gender were partly attributed to reading mode (Tsai, 2009) or proficiency (Poole, 2005, 2009). For example, Poole (2005) found no gender difference on reading strategy for advanced learners in the U.S. but a significant difference for low to intermediate learners in Bogotá (Poole, 2009).

Besides quantity difference, there also seems to be a qualitative difference in strategy use. For example, both Martinez (2008) and Bećirović et al. (2018) found that female students differed from male students on supporting strategies when reading, but they did not differ on global strategies. Roy and Chi (2003) found that girls preferred reading in a linear sequence while boys preferred scanning information in a horizontal way. Neuroscience provides evidence that boys prefer a rule-governed approach to reading (Baron-Cohen, 2002) and that the synthetic phonics method at the word level favored boys more than girls (Logan & Johnston, 2010).

In sum, strategy use differs between genders in addition to the languages involved. Understanding of this can contribute to theory building as well as to benefit classroom teaching. However, for both purposes, being able to measure the strategy is a must.

Measuring Reading Strategies

Reading strategies have been measured with different tools, such as self-reporting questionnaires (Feller et al., 2020; Schmitt, 1990), interviews (Craig & Yore, 1995; Maghsoudi, 2022), verbal protocols (Magliano et al., 2019; Smith et al., 2020), observation (McHardy et al., 2021), computer-based log tracking (Umarji et al., 2021), and eye movement (Jian et al., 2019). Every approach has its strengths and weaknesses (Desoete, 2008; Gascoine et al., 2017). A questionnaire is easy to conduct, but may reflect the quantity rather than the quality of strategic behavior; concurrent verbal protocols provide measures of online strategy use, but it is intrusive and may change the reading process; tracing and eye-tracking allow for direct observation of information processing and strategy application, but it is not feasible for daily use by classroom teachers.

Afflerbach et al. (2008) proposed that strategies are compensatory tools and should be taught and monitored regularly so that they can transform into automatic skills in the end. This formative feature speaks to the strength of the questionnaire as a friendly tool for classroom studies, which matches the context of this study.

A quick survey of the literature reveals several popular questionnaires, one of which is the Metacognitive Awareness of Reading Strategies Inventory (MARSI, Mokhtari & Reichard, 2002). It was developed for Grades 6 to 12 and has been revised, updated, and applied actively to various groups including English language learners (ELLs) (Mokhtari et al., 2018; Mokhtari & Sheorey, 2002). Grade six is the beginning age group for strategy competency requirements by China’s Standards of English Language Ability (CSE) (National Education Examinations Authority (NEEA), 2018), which starts from Level 2 (around Grade 6). Thus, MARSI offers a good starting point for measuring reading strategies for this study.

The scale development process of MARSI implies that it aims for general reading strategies and can apply to various populations or learning environments (Mokhtari & Reichard, 2002). The original inventory was field-tested with 825 students between Grades 6 and 12 from 10 different school districts and five different states in the US. Three factors were revealed, including global reading strategies (GLOB), supporting strategies (SUPP), and problem-solving strategies (PROBS). The scale structure was confirmed with a similar, but new sample of students.

However, though the total sample size is big for scale development, individual group size is small. For example, there are only 31 students representing Grade 6. Despite this, Cronbach’s alpha, a reliability statistic, is impressively high between .86 and .93 across grades. In 2002, Mokhtari and Sheorey modified it into the Survey of Reading (SOR) for ELLs, based on 302 university students in the Midwest US. In 2018, MARSI was revised again and was reduced to 15 items (Mokhtari et al., 2018). The reduced MARSI was validated with students between Grade 6 to first-year community college students.

Mokhtari and his colleagues suggested using SOR for students with low English proficiency and either SOR or MARSI for students with advanced proficiency (Mokhtari et al., 2018; Mokhtari & Sheorey, 2002). However, they also cautioned against limitations in its application in consistency with research that has confirmed or challenged this scale with other populations or languages (Alvarado et al., 2011; Guan et al., 2011).

Rationale for the Current Research

Theoretical understanding has to be supported by empirical data before it can help guide practice. Reading strategies are important for English learning in China. However, there is no locally developed theory or strategy scale that meet our expectation. Replication studies are essential for behavioral and psychological science to advance theory (e.g., Koul et al., 2018; Zwaan et al., 2017). Cross-validation of an existing popular tool with different samples and methodology improvements are two pivotal approaches for this purpose.

As explained above, the individual group size that validated the original MARSI tool was not big, thus validating the tool serves not only to test the generalizability of this operationalized theory of reading strategy but also guarantees quality data to explore differences between L1 and L2 reading and the gender differences as reviewed.

However, we have not found an example with a multi-factor model that involves all relevant statistical challenges brought by Likert-type ordinal data, a repeated measurement design, and correlated error residuals together that have to be addressed in our study (to be described under the methodology section). We located a set of up-to-date statistical procedures proposed by H. Wu and Estabrook (2016) and Svetina and her colleagues (Svetina et al., 2020) that are appropriate for these features. Our research, however, may be a rare concrete example available in the literature based on the procedures they recommended.

Although there are studies comparing strategy use in L1 versus L2, many are based on different groups of students rather than the same students to see how the same participants apply the strategies in different contexts (Feller et al., 2020). Repeated measures have stronger statistical power and such research design can better detect the strategy transfer or non-transfer from L1 to L2.

Grade 10 is of primary research interest to us because this age group is in the optimal center range for strategy instruction (Alexander et al., 1989). Their English proficiency is at a medium level where strategy can have a strong effect on reading (Maghsoudi, 2022). In addition, they have learned English as a foreign language long enough to employ various reading strategies but are not so proficient that differences in metacognitive strategic awareness may be “washed out” (Mokhtari & Reichard, 2004). In other words, they can still identify the strategies that they intentionally employ and researchers can detect them.

To control for extraneous confounding issues, we decided to use a homogeneous group with similar L1 and L2 language proficiency and schooling background, we only studied the paper and pencil reading mode and we applied observational study rather than experimental research design. The benefit of these restrictions, however, is that they enable direct comparison of reading strategies between L1 and L2 reading and enrich the validation studies on a critical developmental group that deserves more attention (Alexander et al., 1989).

Research Question

There are three research questions:

How well does MARSI measure the reading strategies of a group of typical high school English learners in China?

Are metacognitive reading strategies the same or different between L1 (Chinese) reading and L2 (English) reading?

Are there gender differences?

Method

Instrument

To avoid the influence of academic English vocabulary that may go beyond students’ language proficiency, the original MARSI (Mokhtari & Reichard, 2002) was translated into Chinese. Some terms were also simplified for easy understanding. The first Chinese draft has been field-tested with a seventh-grade girl in a less-developed area in China and a sixth-grade boy in an advanced city in China. The former received language education in a traditional way and with limited educational resources while the latter had been exposed to rich academic terminologies and complex teacher languages. Interviews were conducted separately to explore how accurately they could understand the translated inventory items. Then the wording in Chinese was modified and refined. Since the whole research is framed as exploring the habits while reading, we clarify the meaning of the Chinese word “” as “will” rather than “can” (this Chinese word can mean both will as a habit and can as the ability to do something). All these are done to ensure that participants with varying language proficiency can respond to the items with correct and consistent understanding as intended.

Participants and Procedures

All Grade 10 students (aged 15–16) in a high school in Southeast China were invited to participate in the study. This is a convenient sample. There were 20 classes with around 1,000 registered students. Most of these students were top graduates from different middle schools in the city and were highly motivated for further academic pursuits. Digital devices, such as cell phones, were not allowed on campus and there was a strict and tight schedule for students to follow. Students had heavy homework after school and teachers rushed between duties and tasks. This school environment ensured good student habits in completing any task they decided to participate but restricted other possible explorations such as additional time for interviews. This is a limitation in this study. Since digital devices were forbidden for students, all questionnaires were handed out as hardcopies and data were manually checked and typed into a data file.

Ten English classroom teachers handed out the questionnaire to the students and collected them the next day. Students were informed of the purpose of the study and were told that the results would not affect their grades in any way. Participation was voluntary. Responses were anonymous except for two classes (see explanations below).

There was one coordinator on-site in the school who managed the logistics. Responses were mailed back to the first author, and eight graduate and undergraduate assistants worked in pairs to enter the data. Suspicious and missing data were marked and the researchers double-checked problematic responses with the corresponding assistants and the on-site teacher.

To balance objectivity and follow-up exploration, we collected four sets of English reading scores from regular quizzes for two classes (n = 106) and four sets of Chinese reading scores from one of them (n = 46). Except for these two classes who provided names so that we can check for response validity and establish a link between questionnaire responses and reading performance, the only demographic data required of all participants is gender.

Nine hundred forty-nine students in total responded to both the English and Chinese reading strategy questionnaires. There is missing data, but the highest missing rate for individual items is just 0.8%. No apparent missing pattern is detected; thus the data is regarded as missing completely at random (MCAR) (Allison, 2003). For a clear presentation of findings and model comparisons, we used all complete cases (N = 842) for scale validation and invariance tests between languages. Due to missing data on gender and some quiz scores, invariance tests between genders and correlational analyses both involve different sample sizes.

Software

We used SPSS 25 and Mplus 8.4 for all analyses, including exploring and validating the scale structure, testing measurement invariance, comparing between languages and genders, and checking the relationship between strategy use and reading scores.

Analyses and Results

Descriptive Statistics

Overall, students who were required to write down their names reported similar strategy use as their anonymous peers (Figure 1). Out of the original 60 items for two languages, only eight show differences with statistical significance, out of which only two reached the level of moderate effect size (>0.3; J. Cohen, 1988). The pattern is also consistent across classes (Appendix 1). In brief, anonymity does not seem to distort reporting behavior very much and the data of all classes reflect the participants’ stable perception in general.

Figure 1.

Strategy Use by Item between Anonymous Classes and Non-Anonymous Classes.

Scale Structure

Since response data is ordinal, a robust estimation method, the weighted least square parameter estimates with standard errors and mean and variance adjusted (WLSMV) was used. Model fit was evaluated based on overall fit indices including root mean squared error of approximation (RMSEA), comparative fit index (CFI), Tucker-Lewis index (TLI) and standardized root mean square residual (SRMR). Item-level indices such as modification index (MI) and expected parameter change (EPC) were also examined.

Table 1 summarizes the confirmatory model results on English reading strategies.

Table 1.

Fit Indices for Confirmatory Factor Analysis (CFA) on MARSI and MARSI-R for English Reading.

	χ²		RMSEA
CFA model	Value	df	Value	90% CI	p	CFI	TLI	SRMR
Model 1: MARSI	2879.717	402	0.086	[0.083, 0.089]	.000	0.797	0.780	0.068
Model 2: MARSI-R	498.383	87	0.075	[0.069, 0.081]	.000	0.897	0.876	0.047
Model 3: MARSI-R(4-point scale)	372.182	87	0.062	[0.056, 0.069]	.001	0.925	0.910	0.048

Note. N = 842. All single model χ² statistics are significant at p < .001, thus omitted from the table. Adequate fit criterion: RMSEA ≤ 0.08, CFI ≥ 0.90, TLI ≥ 0.90, SRMR ≤ 0.06. Good fit criterion: RMSEA ≤ 0.06, CFI ≥ 0.95, TLI ≥ 0.95, SRMR ≤ 0.05.

Model 1 replicates the original MARSI structure with all 30 items. All indices, including RMSEA (>0.08), SRMR (>0.06), CFI (<0.90) and TLI (<0.90) indicate poor fit (Browne & Cudeck, 1993).

Model 2 is based on 15 MARSI-R items, SRMR (<0.05) indicates a good fit, RMSEA (<0.08) indicates an adequate fit but neither CFI (<0.90) nor TLI (<0.90) is adequate.

Model 3 collapses the two lowest response categories and reduces the 5-point scale to a 4-point scale as Mokhtari et al. (2018) did. With this change, all general fit statistics are now adequate, but only SRMR (<0.05) indicates a good fit. In addition, there are many cross-loading items with high (>10) MI values (11 in Model 2 and 9 in Model 3). This is an important issue that has guided the revision of MARSI by Mokhtari et al. (2018) and needs to be solved before measurement invariance can be explored in this study.

In summary, our data do not support the original MARSI or the MARSI-R strongly. Considering that all five categories are actually present in the responses and that collapsing categories may cause criticism in data validity (Rutkowski et al., 2019), we decided to maintain the 5-point Likert scale and explore the best local model for English reading strategies.

All of the 30 strategies in the original MARSI seem reasonable in theory and they all have standardized loadings higher than 0.3 in preliminary analyses. We decided to maintain the general structure and removed cross-loading items with high MIs following common exploratory confirmatory factor analysis (ECFA) procedures (Brown, 2015; MacCallum et al., 1992). Half of the data (n = 421) was used for model-building and the other half (n = 421) for cross-validation. Our best model happened to settle with the same number of items as MARSI-R although specific entries are different (Mokhtari et al., 2018). Model fit is even better when residuals of two items are allowed to covary. Table 2 reports the key modeling results for English reading strategies.

Table 2.

Fit Indices for Exploratory Confirmatory Factor Analysis (ECFA) for English Reading Strategies.

	χ²		RMSEA
ECFA model	Value	df	Value	90% CI	p	CFI	TLI	SRMR
Model 1: MARSI on Even Cases	1459.563	402	0.079	[0.075, 0.083]	.000	0.811	0.796	0.070
Model 2: MARSI-R on Even Cases	300.756	87	0.076	[0.067, 0.086]	.000	0.889	0.867	0.053
Model 3: Local Best Model on Even Cases	198.553	87	0.055	[0.045, 0.065]	.192	0.950	0.940	0.041
Model 4: Local Best Model on Even Cases with Additional Covariance	170.279	86	0.048	[0.038, 0.059]	.592	0.962	0.954	0.039
Model 5: Local Best Model on Complete Cases	369.060	87	0.062	[0.056, 0.069]	.001	0.942	0.930	0.039
Model 6: Local Best Model on Complete Cases with Additional Covariance	273.339	86	0.051	[0.044, 0.058]	.405	0.962	0.953	0.035

Model 5 and 6 include Items 1, 3, 5, 6, and 12 from the original MARSI GLOB factor, Items 1, 2, 4, 7 and 9 from the SUPP factor and Items 1, 2, 4, 7 and 8 from the PROBS factor. All indices support a good fit of the model to data (Table 2). Cronbach’s alpha is .82 for the overall English reading scale, and .69, .63, and .80 for the GLOB, SUPP and PROBS subscales. These are comparable to the original MARSI-R study (Mokhtari & Sheorey, 2002). Figure 2 shows the standardized estimates for Models 5 and 6. Anything that does not change is no longer presented in the figure for Model 6.

Figure 2.

Final Best Models for English Reading Strategies.

Invariance Between Languages

Comparison between groups is meaningful only after measurement equivalence (ME) or measurement invariance (MI) of the tool is established (Horn & McArdle, 1992). To check ME/I, a series of tests are usually conducted starting from the least restrictive model and gradually moving to the most stringent one (Brown, 2015; Vandenberg, 2002; Vandenberg & Lance, 2000). However, Likert-type measures bring more challenges to ME/I studies, because more parameters such as large amounts of thresholds are added to the model yet the ordered responses do not follow a normal distribution as for continuous responses. We decided to follow the approach recommended by H. Wu and Estabrook (2016) and Svetina and her colleagues (Svetina et al., 2020), but with an extension to accommodate the repeated design that leads to a correlation between the pairs of responses to the same item in both languages. In addition, theta parameterization is used to enable tests on residuals.

To simplify the syntax and facilitate the presentation of results, we recoded the final 15 items for the next two sections. Correspondence between the maintained items and the original MARSI items can be seen in Appendix 2.

Separate sample analyses for English and Chinese reading strategies both support the 15-item model (Table 3). RMSEA is 0.062 for the former and 0.057 for the latter, CFI is 0.942 and 0.952, TLI is 0.930 and 0.942, and SRMR is 0.039 and 0.038 respectively.

Table 3.

Fit Indices for Measurement Invariance Tests Between Languages.

	Single model fit^a								Model comparison fit^b
	χ2		RMSEA			CFI	TLI	SRMR	DIFF χ ²
Model	Value	df	Value	90% CI	p				Models	Value	Δdf	p
Separate-sample Results
English Reading	369.06	87	0.062	[0.056, 0.069]	.001	0.942	0.930	0.039
Chinese Reading	322.82	87	0.057	[0.050, 0.063]	.045	0.952	0.942	0.038
Single-sample Measurement Invariance Tests
Model 0: Equal Form without Correlated Responses	2136.840	390	0.073	[0.070, 0.076]	.000	0.861	0.845	0.056	M0-M1	1198.893	15	.000
Model 1: Equal Form	995.593	375	0.044	[0.041, 0.048]	.997	0.951	0.943	0.040
Model 2: Equal Threshold	1041.426	405	0.043	[0.040, 0.046]	1	0.949	0.946	0.040	M2-M1	28.176	30	.561
Model 3: Equal Loading	1041.425	417	0.042	[0.039, 0.045]	1	0.950	0.948	0.041	M3-M2	14.640	12	.262
Model 4: Equal Intercept	1192.146	429	0.046	[0.043, 0.049]	.984	0.939	0.939	0.041	M4-M3	154.384	12	.000
Model 4b: Partial Equal Intercept	1043.053	424	0.042	[0.038, 0.045]	1	0.951	0.950	0.041	M4b-M3	8.036	7	.329
Model 5: Equal Residual	1084.410	439	0.042	[0.039, 0.045]	1	0.949	0.949	0.042	M5-M4b	60.754	15	.000
Model 5b: Partial Equal Residual	1041.141	437	0.041	[0.037, 0.044]	1	0.952	0.952	0.041	M5b-M4b	20.599	13	.081
Structural Invariance Tests
Model 6: Equal Factor Variance	1030.740	440	0.040	[0.037, 0.043]	1	0.953	0.954	0.041	M6-M5b	2.970	3	.396
Model 7: Equal Factor Covariance	1003.030	443	0.039	[0.036, 0.042]	1	0.956	0.956	0.041	M7-M6	2.063	3	.600
Model 8: Equal Factor Mean	1191.109	446	0.045	[0.041, 0.048]	.998	0.941	0.942	0.042	M8-M7	131.624	3	.000
Model 8b: Partial Equal Factor Mean	1001.420	444	0.039	[0.035, 0.042]	1	0.956	0.957	0.041	M8b-M7	0.925	1	.336

Single model fit is evaluated by the fit indices for each model.

Model comparison is evaluated by fit improvements in the current model from the previous one, which is measured by DIFFχ² and Δdf.

Model 0 is the equal form model which does not consider the repeated measures; it does not fit the data, especially according to CFI (<0.90) and TLI (<0.90). Model 1 is the equal form model with correlated responses; it fits the data adequately by all standards (RMSEA = 0.044, CFI = 0.951, TLI = 0.943, SRMR = 0.040), and it significantly improves Model 0 (DIFF χ² = 1198.885, p < .05). Model 2 and 3 support the threshold and loading invariance hypotheses, but Model 4 to 4b support only partial intercept invariance. Five items do not have equivalent intercepts, among which X8 is higher for Chinese reading and X4, X5, X7, and X15 are higher for English reading (Table 4).

Table 4.

Non-invariant Patterns for the 15 Items.

Factor	Item indicator	Intercept	Error residual	Factor mean
Global	X1: I have a purpose in mind when I read.			−
	X2: I preview the text to see what it’s about before reading it.
	X3: I skim the text first by noting characteristics like length and organization.
	X4: I decide what to read closely and what to ignore.	+
	X5: I try to guess what the material is about when I read.	+
Support	X6: I take notes while reading to help me understand what I read.			−
	X7: When text becomes difficult, I read aloud to help me understand what I read.	+
	X8: I discuss what I read with others to check my understanding.	−
	X9: I paraphrase (restate ideas in my own words) to better understand what I read.
	X10: I ask myself questions I like to have answered in the text.		−
Problem-Solving	X11: I read slowly but carefully to be sure I understand what I’m reading.
	X12: I try to get back on track when I lose concentration.
	X13: When text becomes difficult, I pay closer attention to what I’m reading.
	X14: When text becomes difficult, I reread to increase my understanding.
	X15: I try to guess the meaning of unknown words or phrases.	+	−

Note. + means higher/larger in English; – means lower/smaller in English.

Model 5 and 5b show that only two items (X10 and X15) do not have equivalent error residuals between languages and that error residuals for both X10 and X15 are smaller for English reading than for Chinese reading. In other words, most items are measured with equivalent reliability for both languages except for these two. There are more measurement errors with these two strategies in Chinese reading than in English reading.

After taking measurement error into consideration, factor variances and covariances turned out to be equivalent, but two of the three latent means are higher in Chinese reading than in English reading (an estimated 0.601 vs. model specified 0 for GLOB; 0.234 vs. 0 for SUPP). There is no difference for PROBS means (Figure 3).

Figure 3.

Final Model 8b of Measurement Invariance Tests Between Languages with Repeated Measures.

In sum, all 15 items are stable predictors of the corresponding factors and bear equivalent discriminating power in measuring reading strategies between languages. At least partial invariance is supported for both the measurement part and structural part of our model, implying the comparability of reading strategies between languages using this abbreviated inventory. There are some differences in the base level of the strategies reported, indicating more frequent use of the strategies when reading English than in Chinese. However, although reading strategies in both languages consist of the same three dimensions, global and supporting strategies are relied upon less in English than in Chinese. There is no difference in the problem-solving dimension.

Invariance Between Genders

Some participants did not provide gender information thus analyses in this part reduced to 760 cases (378 girls and 382 boys). Separate sample input is appropriate for this analysis and the model with correlated residuals between X2 and X3 now serves as the baseline. RMSEA, CFI, TLI, and SRMR all support a good fit for all of the models listed in Table 5.

Table 5.

Fit Indices for Measurement Invariance Tests Between Genders.

	Single model fit^a								Model comparison fit^b
	χ ²		RMSEA			CFI	TLI	SRMR	DIFF χ ²
Model	Value	df	Value	90% CI	p				Models	Value	Δdf	p
Separate-sample results
Baseline Boys	153.114	86	0.045	[0.033, 0.057]	.730	0.968	0.961	0.039
Baseline Girls	206.095	86	0.060	[0.050, 0.071]	.051	0.945	0.933	0.046
Measurement Invariance
Model 0: Equal Form	359.515	172	0.054	[0.046, 0.061]	.219	0.957	0.947	0.043
Model 1: Equal Thresholds	399.281	202	0.051	[0.043, 0.058]	.428	0.954	0.952	0.043	M1-M0	32.660	30	.338
Model 2: Equal Loadings	421.463	214	0.051	[0.043, 0.058]	.443	0.952	0.953	0.045	M2-M1	27.431	12	.007
Model 2P: Partial Equal Loadings	405.328	213	0.049	[0.041, 0.056]	.603	0.955	0.956	0.044	M2P-M1	16.263	11	.132
Model 3: Equal Intercepts	491.254	225	0.056	[0.049, 0.063]	.077	0.938	0.942	0.046	M3-M2P	67.182	12	.000
Model 3P: Partial Equal Intercepts	411.706	222	0.047	[0.040, 0.055]	.717	0.956	0.958	0.045	M3P-M2P	13.366	9	.147
Model 4: Equal Residuals	426.093	237	0.046	[0.039, 0.053]	.834	0.956	0.961	0.046	M4-M3P	23.736	15	.070
Structural Invariance
Model 5: Equal Factor Variances	406.919	240	0.043	[0.036, 0.050]	.954	0.961	0.966	0.047	M5-M4	4.470	3	.215
Model 6: Equal Factor Covariances	367.891	243	0.037	[0.029, 0.044]	.999	0.971	0.975	0.049	M6-M5	4.052	3	.260
Model 7: Equal Factor Means	365.894	246	0.036	[0.028, 0.043]	.999	0.972	0.976	0.049	M7-M6	4.204	3	.240

Note. N = 760. Those who did not provide gender information were not included for these analyses. All single model χ² statistics are significant at p < .001, thus omitted from the table. Adequate fit criterion: RMSEA ≤ 0.08, CFI ≥ 0.90, TLI ≥ 0.90, SRMR ≤ 0.06. Good fit criterion: RMSEA ≤ 0.06, CFI ≥ 0.95, TLI ≥ 0.95, SRMR ≤ 0.05.

Single model fit is evaluated by the fit indices for each model.

Model comparison is evaluated by fit improvements in the current model from the previous one, which is measured by DIFFχ² and Δdf.

Comparison between the baseline model (Model 0) and Model 1 shows that thresholds are invariant between genders (DIFF χ² = 32.660, p = .338). However, loadings are not all invariant (DIFF χ² = 27.431, p = .007). The highest MI is for X1 (17.972 for boys and 17.978 for girls). Model 2P releases the equivalence constraint on this item and partial loading equivalence is supported (DIFF χ² = 16.263, p = .132). X1 is about the purpose of reading.

The loading is higher for girls than for boys (see Figure 4), which means this item discriminates girls better than boys, or it is a stronger indicator of girls’ global strategy use than that of boys.

Figure 4.

Final Invariance Test Results Between Genders.

Model 3 constrain the intercepts to be equivalent between genders. This model degrades the model fit at a statistically significant level (DIFF χ² = 67.182, p = .000). Based on MI indices, Model 3P releases intercept constraints on X1, X6, and X7, which supports partial intercept equivalence (DIFF χ² = 13.366, p = .147). X1(having a purpose for reading) does not function the same within the global scale between genders according to the previous step, thus intercept difference at this step is not meaningful. However, intercepts of X6 (taking notes while reading), and X7 (reading aloud when text becomes hard) are higher for girls, meaning girls systematically reported these strategies more frequently than did boys. These are also supported by t-tests on the raw responses, where boys reported far less use of X6 (t = −4.094, p = .000) and X7 (t = −3.429, p = .001).

Model 4 tests the equality of residuals of indicators, which is supported by data (DIFF χ² = 23.736, p = .070), implying that strategies are measured with equivalent reliability for both genders.

Model 5-7 tests the structural invariance between genders. Results show the equivalence of factor variances, covariances, and means. For example, all fit indices are good in Model 7 even by the stringent criterion (RMSEA < 0.05, CFI > 0.95, TLI > 0.95, SRMR < 0.05); and model fit does not degrade significantly (DIFF χ² = 4.204, p = .240). Model 7 is the final model with all possible constraints and Figure 4 presents the invariance test results after measurement error is controlled. The graph on the right presents only two parameters that are different between genders.

In sum, girls and boys do not differ on the overall scale structure nor the majority of individual strategies maintained in the model. However, girls take notes more frequently than boys and tend to read aloud more (see Appendix 3), which agrees with previous research (Denton et al., 2015).

Relationship With Test Scores

In order to cross-validate the scale, we check the correlation between the reported strategies (hereafter referred to as MARSI-C10, i.e., MARSI for Chinese 10th-grade students) and mean reading scores from tests. Because modern Chinese is very different from classical Chinese in vocabulary and syntax, we explored the two scores separately. Results show that only two strategies are statistically correlated with all reading scores (Table 6). In addition, the correlation between English reading scores and modern Chinese reading scores is significant, yet the correlation between English and classical Chinese reading is not; neither is the correlation between modern Chinese and classical Chinese reading.

Table 6.

Descriptive Statistics of Reading Strategies and Reading Tests.

	Mean and SD				Correlation
Strategies (items)	Mean (English)	SD	Mean (Chinese)	SD	English	Modern Chinese	Classical Chinese
X1: I have a purpose in mind when I read.	3.79	1.25	3.93	1.20
X2: I preview the text to see what it’s about before reading it.	3.64	1.30	3.94	1.19
X3: I skim the text first by noting characteristics like length and organization.	3.39	1.28	3.72	1.37
X4: I decide what to read closely and what to ignore.	3.44	1.29	3.29	1.25	.234*
X5: I try to guess what the material is about when I read.	3.91	1.23	4.04	1.18
X6: I take notes while reading to help me understand what I read.	2.64	1.30	2.59	1.28
X7: When text becomes difficult, I read aloud to help me understand what I read.	2.83	1.48	2.35	1.33
X8: I discuss what I read with others to check my understanding.	2.37	1.29	2.64	1.31
X9: I paraphrase (restate ideas in my own words) to better understand what I read.	2.98	1.38	3.02	1.40
X10: I ask myself questions I like to have answered in the text.	3.26	1.20	3.32	1.28
X11: I read slowly but carefully to be sure I understand what I’m reading.	4.24	0.92	4.17	0.92
X12: I try to get back on track when I lose concentration.	4.27	0.94	4.27	0.94		.371*
X13: When text becomes difficult, I pay closer attention to what I’m reading.	4.18	0.97	4.29	0.87
X14: When text becomes difficult, I reread to increase my understanding.	4.31	0.95	4.31	0.83
X15: I try to guess the meaning of unknown words or phrases.	4.27	1.02	3.82	1.19
Test Score	Mean	SD			Correlation
English (n = 95)	25.29	1.84
Modern Chinese (n = 46)	21.94	2.21			.407*
Classical Chinese(n = 46)	15.17	1.37

Significant at p < .05. Non-significant correlations are not reported.

A low correlation between Chinese reading scores and MARSI-C10 is acceptable since MARSI was not originally designed for the Chinese language. However, the low correlation between English reading scores and MARSI-C10 is unexpected. Examination of the quiz items quickly reveals several aspects that may lead to these results.

First, MARSI assumes general reading strategies which may involve long materials and a variety of reading environments. However, the quiz passages are much shorter (fewer than 500 words per task) and students have to respond within a limited time. They are not allowed to disturb others, nor can they refer to any resource. In this context, strategies such as “When text becomes difficult, I read aloud to help me understand what I read” (X7) or “I discuss what I read with others to check my understanding” (X8) may be irrelevant.

Second, test items in this study are not designed to measure the reading process, but rather, to quantify the comprehension result. Most of the items are in the multiple-choice (MC) format which further masks the manifestation of strategies. Students may not have the motivation to “ask myself questions I like to have answered in the text” (X10) rather than figure out the best answer to the test item. This can also explain the significant correlation involving X4, that is, students are actually directed by test questions to “decide what to read closely and what to ignore” (X4). When MC items are involved, test-taking strategies to eliminate unlikely answers may also bring a confounding issue that prevents the full use of reading strategies.

The insignificant correlation between classical Chinese reading and modern Chinese reading is expected. The insignificant correlation between classical Chinese reading and English reading is not surprising either since they are different languages in the end. The significant correlation between modern Chinese reading and English reading scores may seem unreasonable at first sight, but the commonality in the pedagogical and social environment may explain this. Both modern Chinese and English are lively and vibrant languages in China. Modern Chinese is the language of instruction in school and is the official language on public and social media. English is taught in school starting from Grade 3 and is widely accessible through numerous apps and public media. There may be significant overlap in text genre, topic, and reading purpose in the two languages, which can contribute to the correlation between the two. This speculation is consistent with observations in the previous section where model fit improves significantly once the correlation between responses to the same strategy is built into the statistical models (see Table 3 and Appendix 4).

In sum, our data do not support a strong correlation between MARSI-C10 and reading test scores.

Discussion

Value of Local Validation and Usage of Tools

In this study the general factor structure of MARSI is supported, confirming a reasonable construct that applies to our English learners. However, the final model does not involve the same items as the aforementioned MARSI-R. Similar clues can be found in previous studies where researchers sometimes select only a subset of the items from the complete inventory to measure reading strategy, which reflects an implicit belief that manifestation of the construct can vary (e.g., Gnaedinger et al., 2016; Karimi, 2015; Lau & Chen, 2013; Law, 2009).

Our participants report the problem-solving strategies with the highest frequency means, followed by global strategies, then supporting strategies. The priority of problem-solving strategies is consistent with the majority of previous research involving high school English learners (Arabmofrad et al., 2021; Dabarera et al., 2014). However, the order of the other two dimensions is different. For example, Stopar and Ilc (2017) found the same order with their 10th-grade Slovenian students as we did, but Ghaith and El-Sanyoura (2019) found that their 10th-grade Lebanese students used more supporting strategies than global ones. In universities, the order of these dimensions also varies where problem-solving strategies sometimes are not the priority strategies (Barrot, 2016; Chen, 2017). Thus, although a common structure seems reasonable in theory, a local imbalance between subscale dimensions is likely.

This leads to an important note of validation studies: Psychometric validation is not the same as practice validation. In this study, although 15 strategies are deleted from our final model, they all seem reasonable components in theory, eight of which even have high reported means (>3.5; bold numbers in Appendix 2) according to Oxford (1990). This means that students agree with the value of these strategies and have used them a lot. Some items have smaller standard deviations which may be part of the reason why they are not maintained in the model statistically. We speculate that if teachers plan for intervention, they may find little change in these strategies as found in previous studies. For example, Cantrell et al. (2010) found significant changes only in problem-solving strategies by the sixth graders in their study but no change for the ninth graders. They explained that older adolescents already mastered these basic strategies (mean = 3.14), thus growth after the intervention is not detectable. The effects of instruction on these strategies may be obvious for younger and less proficient students, but they may function for older students—such as our participants—only when tuned to more complex tasks or more challenging materials. Based on all this, we caution that statistical insignificance in scale validation cannot be taken as evidence that individual strategies are irrelevant in practice. They serve different purposes.

Universal or Language-Specific Strategy Use?

The repeated design enables a direct comparison of reading strategies between two languages in this study. We found strong correlations between Chinese and English reading strategies. We also found parallel factor structures.

However, there are also language-specific differences in individual strategies. For example, getting back on track when losing concentration is significantly related to Chinese reading but not to English reading. On the contrary, deciding what to ignore is related to the latter, but not to the former. The proficiency gap between L1 and L2 may be a factor since our participants are proficient in Chinese, but not as so in English. We are also suspicious that getting back on track may be a frequent strategy due to demotivating topics or material content more than linguistic challenges. On the other hand, unknown vocabulary with sparse clues in context can pose insurmountable obstacles in foreign language reading. We speculate that that “guessing the meaning of unknown words” is not frequently used because it is a low-efficiency strategy. As a result, ignoring them is a logical, though involuntary, decision for L2 learners in practice.

Whether or not language itself is a factor cannot be settled with existing evidence in this study. However, other research has also shown an insignificant relationship between MARSI strategies and L2 reading comprehension (e.g., Arabmofrad et al., 2021; Fitrisia et al., 2015; Habók et al., 2019). One hypothesis can be that L2 reading comprehension may be achieved through heterogeneous strategies that are shaped by a variety of equally effective factors, rather than by a homogeneous set. Still, more research needs to be done to see how a universal construct operationalizes into contextually different indicators.

Gender Difference

The result of this study supports gender differences in some strategies in favor of girls such as taking notes and reading aloud. This is consistent with previous research (e.g., Poole, 2009). The authors of this paper all have rich experience teaching English and have also noticed boys’ reluctance to use these in practice.

However, we also noticed other strategies that were excluded from our best model but showed statistically significant gender differences in both English and Chinese reading (see Appendix 3). For example, boys reported to “critically evaluate what is read” more, while girls reported to “underline information in text” more. Again, this pattern is consistent with previous research (e.g., Poole, 2009).

The fact that there is no difference in reading scores between genders in our sample suggests that some strategies may not play an important role in reading comprehension, at least not directly for boys. The fact that boys and girls differ on specific items, as constantly found in previous research, suggests that we may need to refine the strategy theory. These results can be interpreted from a new perspective. For example, do some commonly recommended strategies in school reflect girls’ nature more than boys’? Are there boy-friendly reading strategies that are underrepresented in current literature?

Pedagogical Implications

The Role of Local Curriculum

For our participants, problem-solving strategy means are high (>3.5) and supporting strategy means are low (around 2.5). Part of the reason could be that these students may have little chance to read long pieces other than test-type passages. As a result, supporting strategies aiming at “sustaining responses to reading” may not be crucial in local academic life (cf. Mokhtari & Reichard, 2002). Kong (2019) has also found that her Chinese participants took all reading comprehension activities in tests as problem-solving activities regardless of testing formats.

We propose that reducing reading to a kind of problem-solving activity may be true for all language-learning tasks in a non-native environment. This is based on our real-life observation of the overarching cognitive ability of these students in L1 that has been established by this age, the typical reading activities in school, and the instrumental motivation for college admission. For example, none of the reading materials in this study includes a graph or a table, revealing that the participants have little experience with visual cues in daily English reading. This can explain why “using tables, figures and pictures in the text to increase understanding” is reported with the lowest frequency in the study. In addition to this, all passages are narrative, which deviates from the rich target that reading activities can aim for. A detailed study on the linguistic features of the reading test items is in preparation, but preliminary analyses reveal that non-fictional reading materials may be underrepresented in school materials, which restricts the range of useful reading strategies that MARSI has intended for.

Shifting the Focus to Individual Difference

While large samples as used in this study can reveal general patterns of strategy use, relying on general models may restrict our understanding of individuals. Gender is taken as a grouping variable much smaller in scale compared to native language but the findings already show differences on this scale. Individuals can be even more varied. For example, Dabarera et al. (2014) used MARSI before and after an instructional intervention and found that the mean scores increased on each strategy subscale for the experimental group. This was reasonably interpreted as evidence that instruction helped to increase strategy use. However, a closer look at the summary tables shows that the variances of almost all subscale scores have also increased for the experimental group. Even within the control group, whose subscale means did not change, variances are larger for two of the three subscales in the post-survey. Is this increase in variability a good thing or a bad thing? Could it reflect the different yet equally efficient routes to reading comprehension growth? If meaning is constructed by readers through interaction with text, then exploring the atypical cases or various combinations of strategies in different situations may direct us to another valuable perspective, that people can successfully process the same textual information through different approaches (Bjorklund, 1990; Gui et al., 2021).

Limitations

This study used only student-reported data and the sample size for test scores was limited. A follow-up study should also include other data sources such as eye-tracking and reading activities designed specifically for detecting strategy-in-use. Internet-based reading strategy is another category that needs more exploration, although feasibility in schools where digital devices are not allowed limits its value for daily instruction. This study uses a convenient sample which may prevent the generalizability of some results. Students from different schools and grades will provide a complete picture for a possibly dynamic reading strategy construct, dynamic with culture, language proficiency, or simply with age.

Footnotes

Appendix 1

Appendix 2

ID Linkage and Raw Data Statistics for English Reading Strategy (N = 842).

Original MARSI scales	ID of items in the final model in Section 5	Original MARSI items in abbreviation (Mokhtari & Reichard, 2002)	Mean	SD
Global Strategies	X1	G1: Having a purpose for reading	3.49	1.27
		G2: Using prior knowledge	4.13	1.02
	X2	G3: Previewing text before reading	3.45	1.30
		G4: Checking how text content fits the purpose	3.01	1.33
	X3	G5: Skimming to note text characteristics	3.29	1.34
	X4	G6: Determining what to read	3.37	1.28
		G7: Using text features (e.g., tables)	3.72	1.21
		G8: Using context clues	3.92	1.04
		G9: Using typographical aids (e.g., italics)	3.12	1.41
		G10: Critically evaluating what is read	3.23	1.14
		G11: Resolving conflicting information	4.18	1.01
	X5	G12: Predicting or guessing text meaning	3.72	1.23
		G13: Confirming predictions	3.55	1.19
Supporting Strategies	X6	S1: Taking notes while reading	2.66	1.27
	X7	S2: Reading aloud when text becomes hard	2.74	1.48
		S3: Summarizing text information	3.39	1.21
	X8	S4: Discussing with others	2.58	1.28
		S5: Underlining information in text	3.67	1.29
		S6: Using reference materials	3.28	1.33
	X9	S7: Paraphrasing for better understanding	3.07	1.33
		S8: Going back and forth in the text	3.73	1.16
	X10	S9: Asking oneself questions	3.23	1.27
Problem-Solving Strategies	X11	P1: Reading slowly and carefully	4.07	1.04
	X12	P2: Trying to stay focused on reading	4.07	1.02
		P3: Adjusting the reading rate	3.91	1.10
	X13	P4: Paying close attention to reading	4.10	1.05
		P5: Pausing and thinking about reading	3.25	1.23
		P6: Visualizing information read	2.31	1.32
	X14	P7: Re-reading for better understanding	4.15	1.02
	X15	P8: Guessing the meaning of unknown words	4.18	1.00

Appendix 3

Items with Significant Gender Difference (N = 760).

Item ID in Section 5	Original MARSI items in abbreviation (Mokhtari & Reichard, 2002)	English reading		Chinese reading
		t	p	t	p
	G10: Critically evaluating what is read	2.499	0.013	3.436	.001
X6	S1: Taking notes while reading	−4.094	0.000	−3.676	.000
X7	S2: Reading aloud when text becomes hard	−3.429	0.001	−2.629	.009
	S5: Underlining information in text	−6.249	0.000	−5.063	.000
X12	P2: Trying to stay focused on reading	−2.130	0.033	−1.542	.124
	P3: Adjusting the reading rate	−1.957	0.051	−2.610	.009

Appendix 4 Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Fang Chen

Data Availability Statement

The dataset for the current study is available from the corresponding author upon reasonable request.

References

Afflerbach

Pearson

P. D.

Paris

S. G.

(2008). Clarifying differences between reading skills and reading strategies. The Reading Teacher, 61(5), 364–373. https://doi.org/10.1598/rt.61.5.1

Alexander

P. A.

Pate

P. E.

Kulikowich

J. M.

Farrell

D. M.

Wright

N. L.

(1989). Domain-specific and strategic knowledge: Effects of training on students of differing ages or competence levels. Learning and Individual Differences, 1(3), 283–325. https://doi.org/10.1016/1041-6080(89)90014-9

Allison

P. D.

(2003). Missing data techniques for structural equation modeling. Journal of Abnormal Psychology, 112(4), 545–557. https://doi.org/10.1037/0021-843X.112.4.545

Alvarado

J. M.

Puente

Jiménez

Arrebillaga

(2011). Evaluating reading and metacognitive deficits in children and adolescents with attention deficit hyperactivity disorder. The Spanish Journal of Psychology, 14(1), 62–73. https://doi.org/10.5209/rev_SJOP.2011.v14.n1.5

Arabmofrad

Badi

Rajaee Pitehnoee

(2021). The relationship among elementary English as a foreign language learners’ hemispheric dominance, metacognitive reading strategies preferences, and reading comprehension. Reading and Writing Quarterly, 37(5), 413–424. https://doi.org/10.1080/10573569.2020.1846005

Baron-Cohen

(2002). The extreme male brain theory of autism. Trends in Cognitive Sciences, 6(6), 248–254. https://doi.org/10.1016/S1364-6613(02)01904-6

Barrot

J. S.

(2016). ESL learners’ use of reading strategies across different text types. The Asia-Pacific Education Researcher, 25(5-6), 883–892. https://doi.org/10.1007/s40299-016-0313-2

Bećirović

Brdarević-čeljo

Dubravac

(2018). The effect of nationality, gender, and GPA on the use of reading strategies among EFL university students. Sage Open, 8(4), 1–12. https://doi.org/10.1177/2158244018809286

Bernhardt

E. B.

(2011). Understanding advanced second-language reading. Routledge.

10.

Bjorklund

D. F.

(1990). Children’s strategies: Contemporary views of cognitive development. Lawrence Erlbaum Associates.

11.

Brevik

L. M.

(2017). Strategies and shoes: Can we ever have enough? Teaching and using reading comprehension strategies in general and vocational programmes. Scandinavian Journal of Educational Research, 61(1), 76–794. https://doi.org/10.1080/00313831.2015.1075310

12.

Brevik

L. M.

Hellekjær

G. O.

(2018). Outliers: Upper secondary school students who read better in the L2 than in L1. International Journal of Educational Research, 89, 80–91. https://doi.org/10.1016/j.ijer.2017.10.001

13.

Browne

M. W.

Cudeck

(1993). Alternative ways of assessing model fit. In Bollen

K. A.

Long

J. S.

(Eds.), Testing structural equation models (pp. 136–162). Sage.

14.

Brown

(Ed.). (2015). Confirmatory factor analysis for applied research (2nd ed.). The Guilford Press.

15.

Cantrell

S. C.

Almasi

J. F.

Carter

J. C.

Rintamaa

Madden

(2010). The impact of a strategy-based intervention on the comprehension and strategy use of struggling adolescent readers. Journal of Educational Psychology, 102(2), 257–280. https://doi.org/10.1037/a0018212

16.

Cantrell

S. C.

Carter

J. C.

(2009). Relationships among learner characteristics and adolescents’ perceptions about reading strategy use. Reading Psychology, 30(3), 195–224. https://doi.org/10.1080/02702710802275397

17.

Chen

K. T.

(2017). An exploratory study of NNES graduate students’ reading comprehension of English journal articles. Reading in a Foreign Language, 29(1), 20–35. http://hdl.handle.net/10125/66726

18.

Chiswick

B. R.

Miller

P. W.

(2005). Linguistic distance: A quantitative measure of the distance between English and other languages. Journal of Multilingual and Multicultural Development, 26(1), 1–11. https://doi.org/10.1080/14790710508668395

19.

Chuang

H. K.

Joshi

R. M.

Dixon

L. Q.

(2012). Cross-language transfer of reading ability: Evidence from Taiwanese ninth-grade adolescents. Journal of Literacy Research, 44(1), 97–119. https://doi.org/10.1177/1086296x11431157

20.

Clarke

M. A.

(1979). Reading in Spanish and English: Evidence from adult ESL students. Language Learning, 29(1), 121–150. https://doi.org/10.1111/j.1467-1770.1979.tb01055.x

21.

Cohen

A. D.

(Ed.). (2011). Strategies in learning and using a second language (2nd ed.). Routledge.

22.

Cohen

(1988). Statistical power analysis for the behavioral sciences. Lawrence Erlbaum Associates.

23.

Cooper

(2006). The digital divide: The special case of gender. Journal of Computer Assisted Learning, 22(5), 320–334. https://doi.org/10.1111/j.1365-2729.2006.00185.x

24.

Craig

Yore

(1995). Middle school students’ metacognitive knowledge about science reading and science text: An interview study. Reading Psychology, 16(2), 169–213. https://doi.org/10.1080/0270271950160203

25.

Dabarera

Renandya

W. A.

Zhang

L. J.

(2014). The impact of metacognitive scaffolding and monitoring on reading comprehension. System, 42, 462–473. https://doi.org/10.1016/j.system.2013.12.020

26.

Davis

J. N.

Bistodeau

(1993). How do L1 and L2 reading differ? Evidence from Think Aloud protocols. Modern Language Journal, 77(4), 459–472. https://doi.org/10.2307/329671

27.

Denton

C. A.

Wolters

C. A.

York

M. J.

Swanson

Kulesz

P. A.

Francis

D. J.

(2015). Adolescents’ use of reading comprehension strategies: Differences related to reading proficiency, grade level, and gender. Learning and Individual Differences, 37, 81–95. https://doi.org/10.1016/j.lindif.2014.11.016

28.

Desoete

(2008). Multi-method assessment of metacognitive skills in elementary school children: How you test is what you get. Metacognition and Learning, 3(3), 189–206. https://doi.org/10.1007/s11409-008-9026-0

29.

Feller

D. P.

Kopatich

R. D.

Lech

Higgs

(2020). Exploring reading strategy use in native and L2 readers. Discourse Processes, 57(7), 590–608. https://doi.org/10.1080/0163853x.2020.1735282

30.

Fitrisia

Tan

K. E.

Yusuf

Y. Q.

(2015). Investigating metacognitive awareness of reading strategies to strengthen students’ performance in reading comprehension. Asia Pacific Journal of Educators and Education, 30(1), 15–30.

31.

Gascoine

Higgins

Wall

(2017). The assessment of metacognition in children aged 4–16 years: A systematic review. The Review of Education, 5(1), 3–57. https://doi.org/10.1002/rev3.3077

32.

Ghaith

El-Sanyoura

(2019). Reading comprehension: The mediating role of metacognitive strategies. Reading in a Foreign Language, 31(1), 19–43.

33.

Gnaedinger

E. K.

Hund

A. M.

Hesson-McInnis

M. S.

(2016). Reading-specific flexibility moderates the relation between reading strategy use and reading comprehension during the elementary years. Mind Brain and Education, 10(4), 233–246. https://doi.org/10.1111/mbe.12125

34.

Grabe

(1991). Current developments in second language reading research. TESOL Quarterly, 25(3), 375–406. https://doi.org/10.2307/3586977

35.

Grabe

(2009). Reading in a second language: Moving from theory to practice. Cambridge University Press.

36.

Graesser

A. C.

(2007). An introduction to strategic reading comprehension. In McNamara

D. S.

(Ed.), Reading comprehension strategies: Theories, interventions, and technologies (pp. 3–26). Lawrence Erlbaum Associates.

37.

Guan

C. Q.

Roehrig

A. D.

Mason

R. S.

Meng

(2011). Psychometric properties of meta-cognitive awareness of Reading Strategy Inventory. Journal of Educational and Developmental Psychology, 1(1), 3–17. https://doi.org/10.5539/jedp.v1n1p3

38.

Gui

Chen

Verspoor

(2021). The dynamics of reading development in L2 English for academic purposes. System, 100, 102546. https://doi.org/10.1016/j.system.2021.102546

39.

Habók

Magyar

Hui

S. K. F.

(2019). The effects of EFL reading comprehension and certain learning-related factors on EFL learners’ reading strategy use. Cogent Education, 6(1), 1–19. https://doi.org/10.1080/2331186x.2019.1616522

40.

Hofer

B. K.

Sinatra

G. M.

(2010). Epistemology, metacognition, and self-regulation: Musings on an emerging field. Metacognition and Learning, 5(1), 113–120. https://doi.org/10.1007/s11409-009-9051-7

41.

Horn

J. L.

McArdle

J. J.

(1992). A practical and theoretical guide to measurement invariance in aging research. Experimental Aging Research, 18(3-4), 117–144. https://doi.org/10.1080/03610739208253916

42.

Jian

Y. C.

J. H.

Hsiao

Y. R.

(2019). Differentiated processing strategies for science reading among sixth-grade students: Exploration of eye movements using cluster analysis. Computers & Education, 142, 1–14. https://doi.org/10.1016/j.compedu.2019.103652

43.

Karimi

M. N.

(2015). L2 multiple-documents comprehension: Exploring the contributions of L1 reading ability and strategic processing. System, 52, 14–25. https://doi.org/10.1016/j.system.2015.04.019

44.

Koda

(1993). Transferred L1 strategies and L2 syntactic structure in L2 sentence comprehension. Modern Language Journal, 77(4), 490–499. https://doi.org/10.2307/329674

45.

Kong

(2019). Investigating the role of test methods in testing reading comprehension: A process-focused perspective. Springer.

46.

Koul

Becchio

Cavallo

(2018). Cross-validation approaches for replicability in psychology. Frontiers in Psychology, 9, 1117. https://doi.org/10.3389/fpsyg.2018.01117

47.

Lau

K. L.

Chen

X. B.

(2013). Perception of reading instruction and self-regulated learning: A comparison between Chinese students in Hong Kong and Beijing. Instructional Science, 41(6), 1083–1101. https://doi.org/10.1007/s11251-013-9265-6

48.

Law

Y. K.

(2009). The role of attribution beliefs, motivation and strategy use in Chinese fifth-graders’ reading comprehension. Educational Researcher, 51(1), 77–95. https://doi.org/10.1080/00131880802704764

49.

Lin

(2015). A think-aloud study of strategy use by EFL college readers reading Chinese and English texts. Journal of Research in Reading, 38(3), 286–306. https://doi.org/10.1111/1467-9817.12012

50.

Logan

Johnston

(2010). Investigating gender differences in reading. Educational Review, 62(2), 175–187. https://doi.org/10.1080/00131911003637006

51.

MacCallum

R. C.

Roznowski

Necowitz

L. B.

(1992). Model modifications in covariance structure analysis: The problem of capitalization on chance. Psychological Bulletin, 111(3), 490–504. https://doi.org/10.1037//0033-2909.111.3.490

52.

Maghsoudi

(2022). Contributions of motivation to read in L2, proficiency and L1 reading strategy awareness to L2 reading. Reading and Writing Quarterly, 38(3), 215–232. https://doi.org/10.1080/10573569.2021.1931591

53.

Magliano

J. P.

Higgs

Millis

(2019). Deep comprehension of text revealed by talking and writing while reading. In K. Millis, D. L. Long, J. P. Magliano, & K. Wiemer (Eds.), Deep comprehension (1st ed., pp. 253–264). Routledge.

54.

Martinez

A. C. L.

(2008). Analysis of ESP university students’ reading strategy awareness. IBÉRICA, 15(1), 165–176.

55.

McHardy

Chapman

O’Neill

(2021). What strategies do less-skilled adult readers use to read words, and how aware are they of these strategies? Adult Education Quarterly, 71(1), 73–89. https://doi.org/10.1177/0741713620952348

56.

Mokhtari

Dimitrov

D. M.

Reichard

C. A.

(2018). Revising the Metacognitive Awareness of Reading Strategies Inventory (MARSI) and testing for factorial invariance. Studies in Second Language Learning and Teaching, 8(2), 219–246. https://doi.org/10.14746/ssllt.2018.8.2.3

57.

Mokhtari

Reichard

(2004). Investigating the strategic reading processes of first and second language readers in two different cultural contexts. System, 32(3), 379–394. https://doi.org/10.1016/j.system.2004.04.005

58.

Mokhtari

Reichard

C. A.

(2002). Assessing students’ metacognitive awareness of reading strategies. Journal of Educational Psychology, 94(2), 249–259. https://doi.org/10.1037//0022-0663.94.2.249

59.

Mokhtari

Sheorey

(2002). Measuring ESL students’ awareness of reading strategies. Journal of Developmental Education, 25(2), 2–10. http://www.jstor.org/stable/42784357

60.

Muijselaar

M. M. L.

Swart

N. M.

Steenbeek-Planting

E. G.

Droop

Verhoeven

de Jong

P. F.

(2017). Developmental relations between reading comprehension and reading strategies. Scientific Studies of Reading, 21(3), 194–209. https://doi.org/10.1080/10888438.2017.1278763

61.

Nakada

Fujii

Kwee

I. L.

(2001). Brain strategies for reading in the second language are determined by the first language. Neurosciences Research, 40(4), 351–358. https://doi.org/10.1016/S0168-0102(01)00247-4

62.

National Education Examinations Authority (NEEA). (2018). China’s standards of English language ability. Higher Education Press & Shanghai Foreign Language Education Press.

63.

Oxford

(1990). Language learning strategies: What every teacher should know. Heinle.

64.

O’Reilly

McNamara

D. S.

(2007). The impact of science knowledge, reading skill, and reading strategy knowledge on more traditional “high-stakes” measures of high school students’ science achievement. American Educational Research Journal, 44(1), 161–196. https://doi.org/10.3102/0002831206298171

65.

Phakiti

(2003). A closer look at gender and strategy use in L2 reading. Language Learning, 53(4), 649–702. https://doi.org/10.1046/j.1467-9922.2003.00239.x

66.

Poole

(2005). Gender differences in reading strategy use among ESL college students. Journal of College Reading and Learning, 36(1), 7–20. https://doi.org/10.1080/10790195.2005.10850177

67.

Poole

(2009). The reading strategies used by male and female Colombian university students. PROFILE: Issues in Teachers Professional Development, 11(1), 29–40.

68.

Roy

Chi

M. T. H.

(2003). Gender differences in patterns of searching the web. Journal of Educational Computing Research, 29(3), 335–348. https://doi.org/10.2190/7br8-vxa0-07a7-8avn

69.

Rutkowski

Svetina

Liaw

Y.-L.

(2019). Collapsing categorical variables and measurement invariance. Structural Equation Modeling A Multidisciplinary Journal, 26(5), 790–802. https://doi.org/10.1080/10705511.2018.1547640

70.

Schmitt

M. C.

(1990). A questionnaire to measure children’s awareness of strategic reading processes. The Reading Teacher, 43(7), 454–461. https://www.jstor.org/stable/20200439

71.

Smith

Kim

Vorobel

King

J. R.

(2020). Verbal reports in the reading processes of language learners: A methodological review. The Review of Education, 8(1), 37–114. https://doi.org/10.1002/rev3.3170

72.

Stanovich

K. E.

(1982a). Individual differences in the cognitive processes of reading: I. Word decoding. Journal of Learning Disabilities, 15(8), 485–493. https://doi.org/10.1177/002221948201500809

73.

Stanovich

K. E.

(1982b). Individual differences in the cognitive processes of reading: II. Text-level processes. Journal of Learning Disabilities, 15(9), 549–554. https://doi.org/10.1177/002221948201500908

74.

Stopar

Ilc

(2017). Reading for a test: The effect of high-stakes exams on reading strategies. Porta Linguarum Revista Interuniversitaria De Didáctica De Las Lenguas Extranjeras. 2, 103–115. https://doi.org/10.30827/digibug.54115

75.

Sun

Wang

Dong

Zheng

Yang

Zhao

Dong

(2021). The relationship between reading strategy and reading comprehension: A meta-analysis. Frontiers in Psychology, 12, 1–11. https://doi.org/10.3389/fpsyg.2021.635289

76.

Svetina

Rutkowski

(2020). Multiple-group invariance with categorical outcomes using updated guidelines: An illustration using mmplus and the lavaan/semTools packages and the lavaan/semTools Packages. Structural Equation Modeling A Multidisciplinary Journal, 27(1), 111–130. https://doi.org/10.1080/10705511.2019.1602776

77.

Taki

(2016). Metacognitive online reading strategy use: Readers’ perceptions in L1 and L2. Journal of Research in Reading, 39(4), 409–427. https://doi.org/10.1111/1467-9817.12048

78.

Tsai

M. J.

(2009). Online Information Searching Strategy Inventory (OISSI): A quick version and a complete version. Computers & Education, 53(2), 473–483. https://doi.org/10.1016/j.compedu.2009.03.006

79.

Umarji

Day

Zargar

Connor

(2021). Opening the black box: User-log analyses of children’s e-book reading and associations with word knowledge. Reading and Writing, 34(3), 627–657. https://doi.org/10.1007/s11145-020-10081-x

80.

Vandenberg

R. J.

(2002). Toward a further understanding of and improvement in measurement invariance methods and procedures. Organizational Research Methods, 5(2), 139–158. https://doi.org/10.1177/1094428102005002001

81.

Vandenberg

R. J.

Lance

C. E.

(2000). A review and synthesis of the measurement invariance literature: Suggestions, practices, and recommendations for organizational research. Organizational Research Methods, 3(1), 4–70. https://doi.org/10.1177/109442810031002

82.

van Gelderen

Schoonen

de Glopper

Hulstijn

Simis

Snellings

Stevenson

. (2004). Linguistic knowledge, processing speed, and metacognitive knowledge in first- and second-language reading comprehension: A componential analysis. Journal of Educational Psychology, 96(1), 19–30. https://doi.org/10.1037/0022-0663.96.1.19

83.

Estabrook

(2016). Identification of confirmatory factor analysis models of different levels of invariance for ordered categorical outcomes. Psychometrika, 81(4), 1014–1045. https://doi.org/10.1007/s11336-016-9506-0

84.

(2014). Gender differences in online reading engagement, metacognitive strategies, navigation skills and reading literacy. Journal of Computer Assisted Learning, 30(3), 252–271. https://doi.org/10.1111/jcal.12054

85.

Zwaan

R. A.

Etz

Lucas

R. E.

Donnellan

M. B.

(2017). Making replication mainstream. The Behavioral and Brain Sciences, 41, e120–e150. https://doi.org/10.1017/S0140525X17001972

Invariant (or Not) Reading Strategies Between Languages and Genders

Abstract

Plain language summary

Keywords

Introduction

Literature Review

Language Impact on Reading Strategy Use

Gender Impact on Reading Strategy Use

Measuring Reading Strategies

Rationale for the Current Research

Research Question

Method

Instrument

Participants and Procedures

Software

Analyses and Results

Descriptive Statistics

Scale Structure

Invariance Between Languages

Invariance Between Genders

Relationship With Test Scores

Discussion

Value of Local Validation and Usage of Tools

Universal or Language-Specific Strategy Use?

Gender Difference

Pedagogical Implications

The Role of Local Curriculum

Shifting the Focus to Individual Difference

Limitations

Footnotes

Appendix 1

Appendix 2

Appendix 3

Appendix 4

Declaration of Conflicting Interests

Funding

ORCID iD

Data Availability Statement

References