Preschool pathways to reading comprehension: A systematic meta-analytic review

Highlights

• Moderate longitudinal correlations between skills in preschool and later reading comprehension.

• Code-related skills and linguistic comprehension are separable predictors of later reading comprehension.

• Two distinct but related pathways from preschool language to reading comprehension in school.

• Linguistic comprehension has a key role in the development of reading comprehension.

Abstract

The ability to construct meaning from texts is the core of reading. We report a meta-analysis and a systematic review of 64 longitudinal studies tracing the development of reading comprehension from preschool. Previous research showed that linguistic comprehension and code-related abilities in preschool correlate moderately with reading comprehension, but the results across studies are inconsistent. Meta-analytic structural equation modelling showed two distinct but related pathways from preschool linguistic comprehension abilities to reading comprehension in school. One pathway consists of code-related skills (letter knowledge and phonological awareness), and it affects reading comprehension through word recognition. A second pathway consists of linguistic comprehension skills (vocabulary and grammar), and it has a direct influence on reading comprehension. Early interventions to facilitate language development appear to provide a promising approach to facilitate the later development of reading comprehension skills.

Introduction

The ability to simultaneously extract and construct meaning through interaction and involvement with texts is the core of reading (RAND Reading Study Group, 2002). Reading comprehension is critical for all aspects of education and for participation in society. It has long been accepted that the foundations of reading comprehension are laid long before children start learning to read (Teale & Sulzby, 1986). To understand this process, it is crucial to conduct longitudinal studies that trace the precursors of reading comprehension from preschool into the school years. Such studies enhance our theoretical understanding of reading comprehension and also provide the basis for developing methods of teaching to improve reading comprehension.

In the last 15 years, the number of longitudinal studies of reading comprehension has increased rapidly. Here, we present a systematic review that summarises these studies. Given how important reading comprehension is for learning outcomes in school, understanding the factors that promote or impede this ability is critical for educational practice. An important contribution of this review is to evaluate the consistency of findings across studies. Summarizing these studies will not only have important theoretical value but could also give direction for educational practice in terms of assessments and interventions. In addition, this review provides convergent robust evidence across different samples, languages, and contexts on the association between linguistic comprehension ability in preschool and reading comprehension in school.

Several theoretical frameworks describe the complexity of reading comprehension (Cromley & Azevedo, 2007; McNamara & Kintsch, 1996; Perfetti & Stafura, 2014). For children in the early stages of reading development, the most influential theoretical framework is the simple view of reading. According to the simple view, reading comprehension is the product of word recognition (decoding) and linguistic (language) comprehension (Gough & Tunmer, 1986). Word recognition refers to the ability to translate printed words into speech, independent of their meaning. Linguistic comprehension refers to the ability to understand the meaning of spoken language. Notably, Gough and Tunmer used listening comprehension as a synonym for linguistic comprehension, which is in line with studies showing that these two constructs are highly related (Lervåg, Hulme, & Melby-Lervåg, 2017; Protopapas, Mouzaki, Sideridis, Kotsolakou, & Simos, 2012). It has also been suggested that listening comprehension and linguistic comprehension are best understood as one construct (Justice et al., 2017). In their original 1986 article, Gough and Tunmer sought to clarify the role of word recognition in reading and reading disability, resulting in several claims that may inform our understanding of reading comprehension development. For instance, Gough and Tunmer (1986) argued that although both word recognition and linguistic comprehension are necessary conditions for reading to occur, their contributions to reading comprehension are independent. Consequently, their contributions are not necessarily equal – their relative importance may change across time, and there may even be a discrepancy between a reader's word recognition skills and linguistic comprehension ability. All of these predictions can be tested empirically at different developmental stages and at various levels of reading proficiency. However, none of the predictions concern the development of reading comprehension before the onset of word recognition. Thus, in the present study we extend the scope of the simple view of reading by investigating the early foundations of reading comprehension. More specifically, we examine the extent to which preschool measures of linguistic comprehension and code-related skills (i.e., precursors of word recognition) predict later reading comprehension. We perform correlation-based meta-analytic structural equation modelling (MASEM) — a method that allows structural equation models to be fitted to meta-analytic datasets (Cheung, 2015).

Previous research has provided strong support for the simple view of reading. Recent studies show that the components of the simple view explain as much as 94–98% of the variance in reading comprehension in early primary school (Foorman, Koon, Petscher, Mitchell, & Truckenmiller, 2015; Lervåg et al., 2017). However, in some longitudinal studies the two components have explained a relatively small percentage of the variance in reading comprehension (e.g., Torppa et al., 2016). The reason for this difference in results may to some extent be explained by different orthographies and the included measures (Florit & Cain, 2011). When it comes to the two components —word recognition and linguistic comprehension—the relative strength changes across development, thus the length of the studies may also be one factor in explaining the differences in studies which may lead to differences in results. In the early stages of learning to read (e.g., Grades 1–3), word recognition is a major constraint on reading comprehension (Lervåg et al., 2017), but later, when children have mastered word recognition, linguistic comprehension becomes a more important influence on reading comprehension (Verhoeven & van Leeuwe, 2012; see also; Geva & Farnia, 2012; Storch & Whitehurst, 2002).

Storch and Whitehurst (2002) reported a seminal study seeking to predict reading comprehension in the 2nd to 4th grade (4–9 years of age) from oral language and code-related measures in preschool. This study provided strong support for the simple view of reading with two distinct pathways from children's early language abilities to later reading comprehension: a direct linguistic comprehension pathway and a code-related pathway that drives reading comprehension via word recognition skills. In the earliest grades, word recognition had the greatest influence on reading comprehension, but from the third grade onwards, language comprehension made a significant contribution. There was also a strong association between children's preschool linguistic comprehension and code-related skills, indicating a close relation between these component skills at an early developmental stage. However, the strength of this association decreased with age and the children's linguistic comprehension abilities showed higher longitudinal stability than did their code-related skills.

In the wake of Storch and Whitehurst's (2002) seminal study, the number of longitudinal studies of reading comprehension has increased rapidly. However, there are large variations between studies in terms of the preschool predictors they include. The most common predictors are, in line with Storch and Whitehurst's take on the simple view of reading, vocabulary, grammar, phoneme awareness, letter knowledge and rapid automatized naming (RAN; Fricke, Szczerbinski, Fox-Boyer, & Stackhouse, 2016; Hulme, Nash, Gooch, Lervåg, & Snowling, 2015). However, although results from a broad range of predictors are reported in the literature, studies that include a combination of measures that adequately assess both code-related and linguistic comprehension skills are rare (Hjetland, Brinchmann, Scherer, & Melby-Lervåg, 2017). Moreover, despite Gough and Tunmer's (1986) suggestion to use listening comprehension as a measure of linguistic comprehension, few studies involving young children have included this type of assessment (Hjetland et al., 2017). On the other hand, a number of studies include additional predictor variables that are not based on the original account of the simple view of reading, such as working memory, socio-economic background, and nonverbal intelligence (Roth, Speece, & Cooper, 2002; Schatschneider, Fletcher, Francis, Carlson, & Foorman, 2004).

The results from previous longitudinal studies of reading comprehension are inconsistent in several respects. For instance, some studies have found a strong predictive relationship between preschool vocabulary and later reading comprehension (Roth et al., 2002), whereas others have only found a weak relationship (Fricke et al., 2016). The variation in the size of the bivariate correlations between measures, coupled with differences in the type of predictors that have been assessed in different studies, has led to variations in the conclusions drawn on the development of reading comprehension.

Several of the inconsistencies in the results of prior studies, may stem from differences in their methodological approach. For instance, one issue that may explain the between-study variation is sample characteristics, including the age of the participants and how long they have been receiving reading instruction at the point at which reading comprehension is assessed (Hjetland et al., 2017). In studies where reading comprehension is measured early, we would expect to find a relatively weak relation between preschool vocabulary and reading comprehension (and a relatively stronger relationship between reading comprehension and preschool precursors of word recognition, such as phoneme awareness and letter-sound knowledge). In contrast, we would expect to find a stronger relation between preschool vocabulary and reading comprehension in older children who have mastered word recognition.

Previous studies have also shown that measurement issues, such as the type of reading comprehension test used influences the strength of the correlations between reading comprehension, word recognition and linguistic comprehension. Keenan and Betjemann (2006) showed that tests using multiple-choice questions are typically not good measures of reading comprehension, as children may be able to answer many questions on such tests using background knowledge without reading the passage. Keenan, Betjemann, and Olson (2008) showed that tests with open-ended questions are more dependent on linguistic comprehension skills than tests with multiple-choice questions or a cloze procedure. Thus, the type of reading comprehension test used may also have led to inconsistencies in the results of previous studies.

Finally, some of the discrepancies between studies may stem from the failure to deal adequately with measurement error. Measurement error attenuates the correlation between variables, and in multivariate studies, differences in reliability can have unpredictable consequences for the estimation of regression scores because a predictor with good reliability will explain more variance than a competing predictor with poor reliability (Cole & Preacher, 2014). Most prior studies have used single measures of language constructs, such as vocabulary, grammar and phonological awareness. Using multiple measures of each construct is far preferable, as it allows the use of latent variables free of measurement error (e.g., Little, 2013). The use of single measures may cause further inconsistencies between studies due to the large variation in the choice of instruments that are used to measure different language constructs. For instance, vocabulary has previously been assessed by different measures, such as word definition tasks, picture naming, and picture pointing (e.g., the Peabody Picture Vocabulary Test [PPVT]). The same is the case for code-related skills (such as word recognition and phonological awareness) and working memory. The assumption that various tasks are equally representative of the higher-order constructs that they are designed to measure has rarely been tested in prior studies.

To sum up, although developmental studies of reading are increasing in number, variation in the results of these studies limits our understanding of the early pathways to reading comprehension. In the present study we use a MASEM-approach to overcome the limitations of individual studies. Conducting a meta-analysis on previous studies will increase statistical power to detect meaningful associations among constructs, and enable us to generalize findings across different samples, settings and assessment types. The use of latent variables in MASEM allows us to deal effectively with measurement error across individual studies.

Although reproducibility in experimental psychology has attracted much attention in recent years (Open Science Network, 2015), much less emphasis has been put on issues of robustness in multivariate observational studies. Our systematic literature review, coupled with a meta-analysis of key structural relations, will pinpoint which findings are robust and how we can refine future studies.

Three prior reviews are of particular interest: Quinn and Wagner (2018) used MASEM to examine the components of the simple view of reading in concurrent correlational studies including both younger students (from 3.5 years) and adults. Their meta-analytic structural equation model included three latent variables: word recognition, linguistic comprehension, and a construct they referred to as cognitive abilities, consisting of working memory and inferencing skills. The model explained 56.8% of the variance in the students’ reading comprehension, but only word recognition and linguistic comprehension had statistically significant independent contributions to reading comprehension. García and Cain (2014) found a sizeable concurrent correlation between word recognition skills and reading comprehension (r = 0.74), with age and listening comprehension moderating this relationship. These reviews were limited to studies conducted with English speaking samples. Finally, the National Early Literacy Panel (NELP, 2008) summarised longitudinal studies of reading comprehension, including studies up to 2004. The average correlations between language-related variables and reading comprehension ranged from r = 0.20 (concept knowledge) to r = 0.59 (reading readiness). However, only few primary studies were included, and this review did not use a model-based meta-analysis to analyse multivariate relations.

In the present study we investigate the early foundations of reading by summarizing preschool predictors of reading comprehension. The selection of preschool predictors in this review was guided by the simple view of reading and included linguistic comprehension and code-related measures such as vocabulary, grammar, phonological awareness, letter knowledge, and RAN. However, we also included predictors such as working memory, nonverbal IQ, and socio-economic background, which by some accounts may represent useful additions to the simple view of reading (Quinn & Wagner, 2018). We will summarise the size of the bivariate relations between language skills in preschool and later reading comprehension and examine factors that possibly moderate them (e.g., age, test type). As type of reading comprehension assessment has been shown to be associated with the strength of correlation between the reading comprehension and the other two main components in the simple view of reading, we hypothesize that tests with open-ended questions are more dependent on language comprehension skills than tests using a multiple-choice or a cloze procedure. Importantly, tests that use similar procedure to tap the student's comprehension of text may differ in other aspect. The current study summarises studies of the correlation between preschool skills and later reading comprehension. However, most studies in this review only include a single reading comprehension test involving open-ended questions and it remains for future studies to clarify whether the type of reading comprehension test has important effects on the relative importance of word recognition and linguistic comprehension as predictors of reading comprehension. Tests with open-ended questions may be more dependent on linguistic comprehension skills than tests using a multiple-choice or a cloze procedure. Importantly, tests that use similar procedures to assess the students' reading comprehension may differ in other aspects (i.e., vocabulary, testing procedure). Our meta-analysis summarises the correlations between abilities in preschool and later reading comprehension across different types of reading comprehension assessment, examining the possible moderating effects of the type of assessment. Explaining variation across different assessment types is one of the key strengths of a meta-analysis that synthesizes several primary studies.

However, to gain knowledge about how reading comprehension develops concurrent bivariate correlations are not sufficient. We therefore model the relationship between preschool skills and reading comprehension and examine which separate measures go together to define common constructs. We control for measurement error and use latent variables when examining how preschool factors influence later reading comprehension. To accomplish this, we utilize correlation-based MASEM (Cheung, 2015). This approach first synthesizes entire correlation matrices and then performs structural equation modelling on the resultant, aggregated correlation matrix. In contrast to performing separate meta-analyses for each individual correlation, this two-step approach accounts for the dependencies between correlations within studies and thus provides more accurate parameter and variance estimates (Cheung & Cheung, 2016). MASEM also handles missing data efficiently since incomplete correlation matrices need not be excluded—the maximum-likelihood estimation procedures account for missing data efficiently (for details, please refer to Cheung & Chan, 2009). Overall, MASEM provides a powerful approach to testing a model that describes the pathways to reading comprehension. At the same time, it requires correlation matrices to be positive definite and may thus limit the selection of primary studies eligible for structural equation modelling.

This meta-analytic review has the following aims:

• We seek to establish accurate estimates of the relationships between preschool code-related skills, and later word recognition and reading comprehension skills, as well as between linguistic comprehension skills in preschool and later reading comprehension.

• We seek to establish accurate estimates of the relationships between domain general skills, such as nonverbal intelligence and working memory and later reading comprehension. In addition, we seek to examine the relationship between background factors related to socio-economic background (SES) and later reading comprehension.

• We formulate and evaluate a structural equation model with two distinct but related pathways comprising code-related skills and linguistic comprehension as predictors of later reading comprehension skills.

• In this model, we expect code-related skills in preschool to have a significant indirect effect on reading comprehension in school through word recognition skills.

• The hypothesised model describing the pathways from code-related skills and linguistic comprehension to reading comprehension should apply to both younger and more experienced readers and to studies using different types of assessment.