Skip to main content
PMC home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2011 Jun 17.
Published in final edited form as: Read Writ. 2010 Jul 1;23(6):701–717. doi: 10.1007/s11145-009-9180-z

An examination of the relationship between reading comprehension, higher-level and lower-level reading sub-skills in adults

Nicole Landi 1,2
PMCID: PMC3117585  NIHMSID: NIHMS302652  PMID: 21691452

Introduction

Reading comprehension is a complex process that requires the coordination of bottom up word level skills and top down meaning processing skills. Much research over the past several decades has focused on the decoding component of this equation, demonstrating strong correlations between low-level decoding skills and reading comprehension (e.g., Shankweiler, 1989). More recent research has examined the unique contribution of higher-level skills to reading comprehension (e.g., Yuill & Oakhill, 1991; Nation & Snowling, 1998; Landi & Perfetti, 2007). The bulk of the research exploring both lower–level and higher-level contributions has been done with school age children, leaving the population of adults relatively ignored. Understanding reading compression and the skills necessary for adequate comprehension in adults will provide a more complete understanding of comprehension ability. This paper will briefly review findings from studies with children and adult poor readers that examine the relationship between comprehension and other relevant skills (e.g., decoding, working memory, and semantic processing) and the few studies that have looked at this relationship in skilled adults. Moreover, I report findings that further specify the relationship between lower-level and higher-level reading skills in skilled adult readers and on the relationship between low-level and high-level sub-skills and reading comprehension in adults.

Relationship between low-level decoding skill and reading comprehension

Comprehension requires accurate word decoding and recognition, thus, it is no surprise that decoding ability and word recognition skills are highly predictive of comprehension ability (Perfetti & Hart, 2001; Shankweiler, Lundquist, Katz, Stuebing, Fletcher, Brady et al. 1999). Although this is particularly true for young children whose word reading skills are still developing (Curtis, 1980), low-level skills such as decoding have also been shown to account for unique variance in comprehension ability in older children, even for skilled readers (Shankweiler et al. 1996). It should be noted the importance of decoding skill in these studies may in part be due to the deep orthography of English, -- a recent study involving Greek speaking 4th graders used hierarchical linear modeling and found word reading skills (spelling, word and pseudoword reading accuracy, and fluency) to predict comprehension ability before verbal ability and vocabulary were taken into account, but not after (Protopapas, Sideridis, Simos, & Mouzaki, 2007).

Although the consensus is that low-level skill such as decoding play a large role in children, the picture is less clear for adults. Bell and Perfetti (1994) found that decoding (assessed by pseudoword reading) predicted a significant amount of variance (11%) after higher-level skills were partialled out for science texts, but for history texts decoding picked up a only a very small amount of the variance (1%), suggesting that decoding may only play an important role in comprehension when word reading is more difficult (i.e., when individuals are encountering many new vocabulary words, taxing both decoding and word recognition skill, as they would in a science text). However, conclusions from this study should be tempered due to the fact that this was a small sample study (< 100 participants) that targeted low ability readers with different profiles rather than representing a wide range of reading skill. Lundquist (2003), again focusing on lower ability adults, found that college students who performed poorly, (75% or less correct on a decoding measure; the pseudohomophone choice task) also read connected text more slowly but were not significantly different from good decoders (95% or greater on the pseudohomophone choice task) in general comprehension ability, suggesting some dissociation between decoding and comprehension. Furthermore, Braze, Tabor, Shankweiler, & Mencl, (2007), also with a population of less-skilled adult readers, found that word reading and pseudohomophone identification tasks failed to account for additional variance in reading comprehension after listening comprehension was taken into account, whereas pseudoword reading still accounted for significant (18%) variance. However, this relationship was greatly reduced (6%) after vocabulary was added to the model.

Research examining the relationship between reading skills in skilled adult readers is less common. Examining skilled readers is critical for providing information about the relationship between comprehension and other reading skills in the absence of any deficit that would put an obvious limit on comprehension ability. Perfetti and Hart (2002) report data from a factor analysis, which sought to examine the relationship among multiple reading skills in adult skilled, average, and poor comprehenders (using a three-way comprehension score mean split). The authors found partial dissociations between low-level skills such as spelling and elision and higher-level skills such as comprehension and vocabulary for all three groups of readers. However, they found that the pattern of factor loadings was different for the less skilled readers such that an additional factor was required to explain a comparable amount of variance—for poor readers, phonological skills (e.g., elision) loaded on a separate factor from lexical skills (e.g., word identification) -- for skilled and average readers, these were combined on one factor. The authors suggest, that for more skilled readers, lexical-level factors and phonological factors are linked, whereas for poor readers they are separate. The authors did not conduct any follow-up analyses to determine the relative contributions of each of the sub-skills after accounting for basic skills such as decoding and IQ, thus it is difficult to determine the degree to which each of the skills tested contributed to reading comprehension ability. More recently, Jackson (2005) examined the relationship between academic skills (e.g., ACT) and a variety of reading measures among a small sample of highly skilled adult university students. She found that reading measures clustered into three independent components – decoding accuracy, reading speed, and text comprehension. This finding is consistent with those of Perfetti and Hart (2002), pointing towards a dissociation between comprehension and decoding abilities in more skilled readers. Interestingly, these findings indicate a unique role for reading speed/fluency that is separate from decoding ability among highly skilled readers. This dissociation has been previously suggested in work with less-skilled readers and individuals with dyslexia (e.g., Wolf, Goldberg, O'Rorke, Gidney, Cirino, & Morris, 2002), but the contribution of fluency to comprehension has not been further investigated for skilled adult readers. One difficulty for interpreting Jackson's (2005) findings, as in the study reported by Perfetti and Hart (2002), is that she did not include any additional analyses (beyond the data reduction) in order to determine the degree to which decoding and fluency were able to predict comprehension ability after taking other measured skills into account.

Taken together, findings from studies that have investigated the relationship between multiple reading skills in children and adults indicate that the relationship between low-level reading skills and comprehension is dependent on the age and skill level of the participants. That is, for children and less-skilled readers low-level skills play a more prominent role in comprehension ability. However, although the studies that have examined skilled adult readers suggest a partial dissociation between comprehension and decoding it is difficult to determine the precise contribution of low-level skills to comprehension skill from these studies because regression models needed to examine partial contributions were not included.

Dissociation between comprehension and decoding in children

Skills other than decoding and word reading have been shown to predict comprehension in both children and adults. For example, skills such as the ability to hold information online in working memory, inference making, and comprehension monitoring have all been shown to be related to overall comprehension ability (see Perfetti, Landi, & Oakhill, 2005 for a review of this research). One complexity for interpreting the relative contribution of these higher-level skills to comprehension ability is that the research that has focused on higher-level contributions to comprehension failure has often failed to include appropriate decoding measures (Perfetti & Lesgold, 1979; Perfetti, 1985; Perfetti, Marron, & Foltz, 1996). For example, studies often include a “decoding” test that requires word reading in context, yet, context has been shown to have a bootstrapping effect on reading ability (Perfetti & Hogaboam, 1975). However, a growing body of work with children now has included appropriate decoding measures (word reading in isolation or non-word reading) and found reading and listening comprehension difficulties in some children with normal decoding abilities (Oakhill & Cain, 2000; Oakhill, Cain, & Bryant, 2003; Stothard & Hulme, 1992). Readers with this discrepancy are sometimes referred to as having Specific Comprehension Deficit (SCD), Specific Comprehension Impairment (SCI), or simply as poor comprehenders (see Perfetti et al. 2005). Researchers estimate that readers with SCD make up as much as 10-15% of readers in the seven to eight year old age range (Stothard & Hulme, 1995). Recent studies have demonstrated that this particular group of children have trouble generating text appropriate inferences (Oakhill & Cain, 2000), monitoring their comprehension progress (Oakhill et al. 2003), using relevant semantic information (Nation & Snowling, 1998), processing syntactically complex sentences (Hagvet, 2003), and holding information online in working memory (Oakhill et al. 2003), all despite adequate single word reading ability (see Nation, 2005 for a review of these findings). In light of the controls for adequate decoding ability in these studies, differences in reading and listening comprehension ability, working memory, and semantic processing must be accounted for by some other mechanism.

It is important to note, when comparing these findings with studies that identify decoding as an important predictor of comprehension skill (e.g., Shankweiler et al. 1996), that these studies test a relatively small percentage of the population of children who are specifically selected as having a discrepancy between decoding and comprehension. Thus, these findings are not inconsistent with findings of strong correlations between decoding and comprehension in school aged children – however, they do suggest that the strong relationship between these two skills does not hold for all children. Of great importance to ongoing research is to further refine the nature of the discrepancy between decoding and comprehension that exists in some children and to determine the prevalence of this discrepancy over development. For example, Catts, Adolf, & Ellis–Weismer (2006) provided a developmental assessment of children with comprehension impairment by specifically comparing reading skills of children identified as poor comprehenders to those identified as poor decoders in kindergarten, second, fourth, and eighth grade. They found that eighth grade poor comprehenders (older children than had previously been studied) had deficits in language comprehension but not phonological processing (with the opposite pattern for poor decoders). Furthermore, they found that children identified as poor comprehenders or poor decoders in the younger grades, including kindergarten had very similar reading skill profiles -- although the discrepancy was larger in the later grades, suggesting that the dissociation becomes more prominent once children have transitioned to fluent decoding.

The primary aim of the current study was to identify whether a similar dissociation between comprehension (and other high-level skills, such as vocabulary) and low-level reading skills (such as decoding) exists in the adult population – as in children with SCD, and if so, to determine the nature and prevalence of this discrepancy. Furthermore, I sought to determine the relative contribution of multiple reading sub-skills, including, decoding, spelling, and print exposure as well as general cognitive skills such as non-verbal IQ to reading comprehension skill in adult skilled readers. Evidence reviewed above from Braze et al. (2007) and Lundquist (2003) suggests that decoding is not a good predictor of comprehension ability in adult less-skilled readers. Moreover, studies by Perfetti & Hart (2002) and Jackson (2005) are suggestive of a partial dissociation between comprehension and decoding. However, these studies a) focused specifically on less-skilled readers, b) had small sample sizes and/or c) failed to include regression models to determine the relative contributions of low-level skills to reading comprehension ability. Moreover, none of these studies explicitly set out to determine whether a dissociation between low-level skills and high-level skills exists in adult readers.

Current study

The current study examined a very large database containing data from over 900 adult participants who took multiple reading tests designed to capture higher-level and lower-level reading abilities. In particular, I examined the factor structure of this database by determining which reading sub-skills (comprehension, vocabulary, decoding, and spelling) share the most variance. Additionally, I looked at the relationship between non-verbal IQ and experience (measured by print exposure) and these reading sub-skills. Furthermore, I report descriptive statistics that characterize the population of adult readers, including the frequency of adults who have discrepancies in their ability profiles. Finally, I also present data from regression models that quantify the contribution of higher-level and lower-level reading sub-skills as well as IQ to reading comprehension.

Methods

Participants

Nine hundred and twenty-eight participants (Mean Age = 20.17, SD = 3.69) at a public university in the northeastern United States took a battery of skill assessments (described below) over the course of the 2004, 2005, and 2006 academic years. This assessment was part of an ongoing screening procedure that was used for subject selection for participation in subsequent studies. Participants were recruited through the introductory psychology pool and through the university newspaper. All participants were native English speakers, with normal or corrected to normal vision. Subjects were screened for epilepsy, history of reading disability, and brain injury. Participants were compensated with credit required for completion of their introductory psychology class or with one payment of seven dollars.

Procedure

The procedure for the study consisted of a series of skill assessments: All tests were given in paper and pencil format and were given in groups of approximately 25 participants. The Nelson-Denny (ND) vocabulary, ND comprehension, and Ravens Matrices tests were timed (see below) and participants were given the rest of a one and a half hour period to complete the other three tests. The tests and respective skills they measure are discussed in detail below.

Measures

Comprehension

The comprehension subtest of the Nelson-Denny (1973) required participants to read a paragraph and answer questions about the content of that paragraph. Participants chose one of five possible answers to each of 36 questions. Participants were given 15 minutes to complete the task. Participants received one point for every question they answered correctly and they received a one fifth point deduction for each question they answered incorrectly or left blank. By using an equal deduction for wrong and blank answers, both speed and accuracy were taken into account in one measure.

Vocabulary

The vocabulary sub-test of the Nelson-Denny (1973) provided a word and asked the participant to select the correct definition from a list of five possible choices (participants repeat this procedure for each of 100 words). Participants were given seven and a half minutes to complete as many items as possible. As in the ND comprehension test, participants received one point for every question they answered correctly they received a one fifth point deduction for each question they answered incorrectly or left blank.

Print exposure

To measure print exposure we used the Author Recognition Test or ART (Stanovich & West, 1989), which consists of a list of names, many of which were names of common authors. The participant's task was to correctly identify the authors and avoid making false alarms to non-author foils. The test contained 80 items and was not timed. Participant performance was measured by d’ (z transform of hits – z transform of false alarms).

Decoding

To measure decoding ability of a large sample we used the pseudohomophone choice task (Olson, Forsberg, Wise, & Rack, 1994) which required participants to sound out written pseudo-words and answer whether or not they had the same pronunciation as a real word (e.g., audishon). The test contained 96 items and participants were not timed. The scoring method used for this test was a standard d’ calculation. This test was used because it can be given in a large group setting. This test has been shown to correlate highly (.71) with the Woodcock Johnson Word Attack measure, one of the most commonly used test of non-word decoding in the literature (Braze et al. 2007).

Spelling

The Baroff spelling test was used (see Perfetti & Hart, 2001; Frishkoff, Collins-Thomas, Perfetti, & Callan, 2008). This is a spelling discrimination task in which participants are presented with one correct and four incorrect spellings of irregular, easily misspelled words (e.g., nuisance, nuisence, newsance, newcense, newsince). Participant performance was measured by d’.

Non-verbal IQ

To obtain a measure of general intelligence that would not rely on verbal skills an abbreviated version of the Ravens Advanced Progressive Matrices was used (Raven & Raven, 2003). Participants were presented with a series of complex patterns and one incomplete pattern and were asked to find the missing part (select one of five choices) of the incomplete pattern. The test contained 18 total items (selected to represent the range of difficulty on the task), and participants were given 20 minutes to complete the test. This test was scored in the same manner as the ND vocabulary and comprehension tests with one point given for correct answers and a one fifth point reduction taken for incorrect or blank answers. Again, as with the decoding measure, this test was chosen because it can be given in a group setting.

Results

Overall Performance

There was a large range in performance for all of the assessments: Table 1 shows descriptive statistics from all assessment measures.

Table 1.

Means, standard deviations and ranges for the age of participants and performance on the six measures.

Minimum Maximum M SD
Age 18.00 49.00 20.17 3.69
Vocabulary -11.60 96.40 41.99 18.50
Comprehension -2.40 36.00 18.61 7.28
ART -0.13 0.93 0.29 0.16
Decoding -0.06 1.02 0.73 0.19
Spelling -0.71 1.14 0.79 0.16
Ravens -2.25 18.00 7.27 3.91
Open in a new tab

Correlations

Correlations between all measures were calculated. All correlations were significant (all p < .05); however, some correlations were stronger that others (Table 2). Significant correlations with comprehension ability, in order of correlation strength were: Vocabulary (.669), ART (.405), Spelling (.226), Ravens Matrices (.159), and Decoding (.092). The correlation with decoding seems particularly low in this analysis; however, it is not inconsistent with other findings that have used the same test with a similar population. For example, Lundquist et al. (under revision), report correlations between this specific test and comprehension that are below .2. This finding is also consistent with the relatively low correlation (.11) observed by Jackson (2005) between nonword reading (measured with the Test of Word Reading Efficiency [TOWRE]) and reading comprehension (Jackson, 2005). Higher correlations between decoding and comprehension are often seen in children for whom decoding is a nascent skill, but reports examining skilled adults suggest that these two skills are not necessarily strongly correlated. Braze et al. (2007) examined less-skilled adults and found a higher correlation between reading comprehension and the pseudohomophone choice task (.42), suggesting that for these individuals, who continue to struggle with reading, decoding may be a more significant factor.

Table 2.

Correlations between tasks

Comprehension ART Decoding Spelling Ravens
Vocabulary .669 ** .463** .181** .298** .167**
Comprehension .405 ** .092 ** .266 ** .159 **
ART .246** .336** .068*
Decoding .379** .105**
Spelling .079*
Open in a new tab
**

p < 0.01 level

*

p < 0.05 level

Principal Components Analysis (PCA)

In order to better characterize the structure of the data, a more complex measure that captures higher order correlations was employed. A PCA was conducted on the data using a Varimax rotation with a Kaiser normalization (orthogonal solution) and was followed up with a Promax rotation (oblique solution). Using Kaiser's (1960) stopping rule, only eigenvalues greater than one were considered for component identification (see Bryant and Yarnold, 1995 for a review of PCA methods).

The PCA, with a Varimax rotation, identified two components that together accounted for 68.2% of the variance (47.5% by the first, and 20.7% by the second). In this analysis a conservative cutoff of .5 (factor loadings of .5 or greater) was used for inclusion in a particular component1. In general, the larger the factor score of a variable (i.e., test), the more likely that variable is to “belong” to that component. However, for factors to be independent it is important that variables have a “large” loading only for one component (Bryant & Yarnold, 1995).

In the first component (C1), Comprehension, Vocabulary, and Author Recognition had factor loadings greater than .5 (.869, .843, and .746, respectively), and Decoding and Spelling had factor scores lower than .5 (.059 and .225, respectively). In the second component (C2), Decoding and Spelling had factor scores greater than .5 (.847 and .743 respectively), and Vocabulary, Comprehension, and Author Recognition had factor scores lower than .5 (.157, .079, and .191, respectively) (see Table 2). These results indicate that similar reading sub-skills underlie performance on comprehension, vocabulary and author recognition tests, and that these sub-skills are different from those reading sub-skills that underlie performance on decoding and spelling tasks. Furthermore, as Table 3 reveals, spelling skills contributed somewhat more to C1 than decoding skills did and that vocabulary, and author recognition contributed somewhat more to C2 than comprehension did. A PCA with a Promax rotation (allows intercorrelations) confirmed the findings of the Varimax PCA. Two factors were identified that together accounted for 68% of the variance (47.5 % by the first factor and 20.7 % by the second). In the first component, (C1), Vocabulary, Comprehension, and Author Recognition all had factor scores above .5 (.879, .866, and .745, respectively) and Decoding and Spelling sub-tests have factor scores below .5 (-.084 and .108, respectively). In the second component, (C2), Decoding and Spelling had factor scores above .5 (.872 and .735, respectively). A low correlation of .27 between the two factors in the Promax rotation supports the orthogonality of the two factors.

Table 3.

Results of a Varimax rotated PCA with Kaiser normalization

Component1: Higher-level skills Component 2: Lower-level skills
ND Comprehension .881 .025
ND Vocabulary .868 .139
Author Recognition .608 .382
Ravens Matrices .297 .057
Decoding .006 .864
Spelling .265 .745
Open in a new tab

Despite the moderate correlations between all of the tests, the separation of test performance into two components suggests that there are two broadly defined “types” of reading skill. This separation also suggests that the tests identified within a component tap somewhat similar skills and that the tests that lie across components tap somewhat different skills. One question that arises is: what are the percentages of participants that show a discrepancy between higher-level skills (C1) and lower-level skills (C2)? To investigate this question, two weighted factor composite scores were made for all participants by multiplying their normalized test scores (individual score- mean/SD) by the corresponding factor loading for each test and then averaging over tests (this was done for component one and for component two, thus providing a C1 score and a C2 score for each individual). For example, for each person's C1 score their vocabulary, comprehension, and ART scores were weighted more highly than their spelling and decoding scores, and vice versa for each person's C2 score. The scores for each person on each component were then plotted, and split by the mean score on both components to create four quadrants: high on both, low on both, high on one and low on two, and high on two and low on one (Figure 1). As would be expected, individuals who scored above average on both or below average on both of the components were in the majority. However a number of people had a discrepancy between their two scores when a simple mean split was considered. In particular, 130 out of 928 people (14%) scored at or above average on C2 but below average on C1 and only 87 (9.3%) of the individuals tested had the reverse discrepancy. This number however is greatly reduced when I impose a harsher cutoff for discrepant characterization – if I only consider individuals who fall beyond .5 standard deviations from the normalized mean on each component, there are only three individuals with better C2 than C1 performance and no individuals with the reverse discrepancy (see figure 1). This simple descriptive analysis suggests that within the adult population there is a significant number of individuals that have discrepant skill profiles (when a simple mean cutoff is used) and that there are somewhat more individuals with a discrepancy associated with poorer high-level skills and weaker low-level skills; however, the number of individuals with “discrepant profiles” in this sample is greatly reduced when a cutoff is used for discrepant classification. In sum, when component scores are used to assess discrepancy, there are not a large number of individuals with highly discrepant scores – in part this may reflect the multifaceted nature of these scores. Despite the relatively orthogonal, two-factor structure, some component skills were still more correlated with each other than others, moreover some skills loaded more clearly on one factor only than others (as can be seen in tables 2 & 3). Examination of the relationship between two scores alone (such as comprehension and decoding) may show more discrepancy; however, for the current analysis the primary concern was the relationship between high-level and low-level skills as defined by the PCA. Regression models in the next section further describe the relationship between individual skills.

Figure 1. Scatter plot of C1 and C2 scores.

Figure 1

Open in a new tab

Figure 1 shows the relationship between Component 1(C1) and Component 2 (C2) scores for each participant. Component scores were created for all participants by multiplying their normalized test scores (individual score- mean/SD) by the corresponding factor loading for each test and then averaging over tests. The solid lines mark .5 standard deviations from the mean on each component. Numbers presented in the corner of each quadrant represent the number of individuals falling in each quadrant.

Regression

A series of hierarchical regression analyses were conducted to determine any remaining contribution of low-level skills after accounting for vocabulary and print exposure. As Table 4a shows, Vocabulary accounted for 45 % of the variance, ART only accounted for another 1.2% of the variance, and the lower-level skills of Decoding and Spelling (as well as the Ravens test) accounted for less than 1% of the variance. A second model was also tested, entering low-level skills first (Table 4b). Even when decoding is entered into the regression model first, it still accounted for less than 1% of the variance. Spelling accounted for 6%, Vocabulary 44%, ART 1%, and Ravens < 1%. These findings show that low-level skills such as decoding ability do not contribute very much to comprehension skill in this adult population. In this sample, this was true whether decoding was entered before or after high-level skills. One final model was analyzed entering IQ into the equation first (Table 4c). In this model IQ made a larger but still small (3%) contribution to comprehension ability. Predictions from low-level and high-level skills were similar to the first two models -- vocabulary remained the strongest predictor of comprehension ability.

Table 4.

Regression analysis exploring the predictors of reading comprehension a) High level skills entered first) b) Low- level skills entered first c) IQ entered first

a)
Variables R2 Adjusted R2 R2 Change Sig. F Change
l.Vocab .45 .45 .45 .000
2.ART .46 .46 .01 .000
3.Decoding .46 .46 .003 .037
4.Spelling .47 .46 .004 .006
5.Ravens .47 .47 .003 .029
b)
Variables R2 Adjusted R2 R2 Change Sig. F Change
1.Decoding .008 .007 .008 .005
2.Spelling .07 .07 .06 .000
3.Vocabulary .46 .45 .39 .000
4.ART .47 .46 .01 .000
5.Ravens .47 .47 .003 .029
c)
Variables R2 Adjusted R2 R2 Change Sig. F Change
1.Ravens .025 .024 .03 .000
2.Vocabulary .450 .449 .43 .000
3.ART .462 .460 .01 .000
4.Decoding .465 .463 .003 .024
5.Spelling .469 .466 .004 .007
Open in a new tab

Discussion

These findings present evidence that adults’ higher-level reading skills (comprehension and vocabulary) are partially dissociated from lower-level skills (spelling and decoding). This finding is consistent with Jackson (2005), Lundquist et al. (under revision), & Perfetti & Hart (2002). Moreover, print exposure clustered with higher-level skills, suggesting a strong relationship between experience and comprehension ability. Non-verbal IQ, however, was not associated with higher-level reading or lower-level reading skills in this sample, suggesting that for skilled readers both lower-level reading skills and higher-level reading skills are independent of IQ.

Finally, the regression analyses suggest that decoding accounts for only a very small amount of the variance in comprehension ability (< 1%) in this sample of adults, even when decoding is entered before higher-level skills. This result differs somewhat from that of Braze et al. (2007) who found decoding to be a moderate predictor (accounting for 11%) of the variance before vocabulary was entered into the equation and a weak predictor (accounting for ~1%) after entering in vocabulary. One difference between the current study and theirs is that they specifically sought out participants that were likely to have poor literacy skills (most from community colleges) and the current population attended a traditional university. Moreover, this finding is also very different from findings with young children who have poor reading difficulty (e.g., Shankweiler et al. 1999), that identified a very high correlation (.79) between decoding and comprehension. This difference is consistent with the hypothesis that decoding has a larger impact on comprehension ability when overall reading ability is low (Bell & Perfetti, 1994).

These results further support the hypothesis that word knowledge is critical for good comprehension. Vocabulary is the single best predictor of comprehension ability (accounting for approximately 40% of the variance in a hierarchical regression); it has the highest correlation with comprehension ability, and it clusters strongly with comprehension in the PCA. This result is consistent with the findings of Braze et al. (2007) and Lundquist (under revision) who also found vocabulary to be a strong independent predictor of comprehension skill in adults and with several studies with children that found high correlations between vocabulary knowledge and comprehension skill (Cunningham et al. 1990; Catts et al. 2006; McKeown et al. 1983; Nation & Snowling, 1998 and in Greek, Protopapas et al. 2007). This relationship has been further refined in a recent study by Wagner, Muse, & Tannenbaum (2006) that specifically identified vocabulary breadth of word knowledge as critical for comprehension.

The current study did not reveal a large contribution from spelling ability (<1%) after decoding was taken into account or a large contribution from nonverbal IQ, which even when entered first only accounted for 3% of the variance in comprehension ability. When IQ was entered after either the low-level or high-level skills it accounted for < 1% of the variance. The fact that IQ was a poor predictor is consistent with other studies that look at predictors of comprehension ability (e.g., Cutting & Scarborough, 2006). However, this sample was relatively homogenous in terms of general cognitive ability—samples that include a broader range of skill level may find IQ to be a better predictor of comprehension ability.

It is important to note that the use of other tests may have produced somewhat different findings – there has been a significant amount of recent research on assessment of reading comprehension that suggests that the choice of a specific comprehension measure and even the general type of comprehension measure can have important implications for comprehension research (see Fletcher, 2006; and Francis, Fletcher, Catts, & Tomblin, 2005 for recent reviews of this literature). For example, Cutting and Scarborough (2006) have shown that the choice of reading comprehension measure has an impact on the measured relationship between comprehension and other skills. Specifically, they find that tests of comprehension do not all tap the same underlying language skills and cognitive processes and thus can be differentially influenced by particular skills. Moreover, Keenan & Betjamman (2006) investigated performance on the Gray Oral Reading Test (GORT), a common multiple choice comprehension measure, and found that children were above chance even when they did not actually read the passages. This suggests that many of the questions can be answered using prior knowledge alone and point out that students are likely to perform above their actual comprehension ability. Thus, the choice of a simple multiple choice comprehension test such as the GORT (and possibly other multiple choice/passage comprehension tests such as the Nelson-Denny) may not be ideal for assessing all aspects of comprehension skill. New models and tests of comprehension that are research based and are able to distinguish more accurately between sub-skills (e.g., the Diagnostic Assessment of Reading Comprehension [DARC]; August, Francis, Hsu, & Snow, 2007) will improve our ability to accurately assesses comprehension, and to distinguish comprehension from other reading skills such as decoding and fluency. The current data strongly support the need to test high-level skills independently of low-level skills. In addition to tests like the DARC, which attempt to isolate sub-skills relevant for comprehension, data presented here support the framework proposed by Catts et al. (2006), based on the simple view of reading that suggests a system that classifies readers based on their relative strengths and weaknesses in both word recognition and comprehension.

Moreover, the decoding measure used in this study, the pseudohomophone choice task, may be a sub-optimal measure of decoding. A one-on-one administered test of difficult non-word reading may have been more challenging, thus, these data may slightly underestimate the contribution of decoding. However, Jackson (2005) used such a difficult nonword-reading task with skilled adult readers and found similar strong dissociations between decoding and comprehension. Furthermore, as previously discussed in this article, the pseudohomophone choice task has been shown to correlate highly with other one-on–one tests such as the TOWRE (Braze et al. 2007).

Also note that the tests of comprehension and vocabulary used in this study are not necessarily pure measures of these respective skills; both require reading and include time as a factor in their analysis leaving open the possibility that fluency may affect performance on these tasks – but, if anything, this would likely have lead to an underestimation of the discrepancy between word decoding skill and comprehension. However, as Jackson (2005) demonstrates, fluency is an important and potentially independent factor that contributes to comprehension skill. Thus, we cannot assume that fluency will pair with or be subsumed by measures of other low-level skills. The relationship between fluency, vocabulary, and comprehension, particularly in skilled readers, remains a complex one and further research is needed to examine the relationship between comprehension and fluency in this population. Furthermore, with respect to the vocabulary assessment used here, significant research suggests that vocabulary can be measured in multiple ways, and that breadth and depth of knowledge are meaningful distinctions that may differentiate reader's vocabulary abilities (e.g., Nagy & Herman, 1987; Wagner et al. 2006)). Certainly these sub-components of vocabulary may make differential contributions to comprehension skill, particularly in terms of the role of general knowledge and inference making as well as semantic processing ability in comprehension. Greater examination of the multidimensional nature of vocabulary and comprehension will improve our understanding of the strong coupling of these skills.

Finally, the tests included in this battery are far from exhaustive; other than IQ, I did not examine additional skills that are likely to be important for reading comprehension, such as working memory, phonological analysis, semantic processing, fluency, and listening comprehension. This absence of other measures was due in part to the method of data collection, which was done in groups, in paper-pencil format in order to ensure a large sample size. These other skills likely contribute significantly to comprehension ability, depending on the particular population being tested (e.g., listening comprehension is one of the best predictors of reading comprehension; Hoover & Gough, 1991; Gough & Tumner, 1986; Cutting & Scarborough, 2006). However, some of these un-tested skills may be subsumed by higher-level skills like vocabulary. For example, recent work shows that working memory does not contribute to comprehension ability once vocabulary and decoding have been accounted for in children (Cutting & Scarborough, 2006) and in adults (Braze et al. 2007). This may be due to the fact that working memory tasks tend to draw heavily on phonological processing (Waters & Caplan, 1996), and this variance is likely to be subsumed by decoding ability. Working memory tasks also draw heavily on semantic skills (Marshal & Nation, 2003) and this variance may be subsumed by vocabulary.

Taken together, these findings join other research showing that reading sub-skills other than decoding affect comprehension – and that in skilled adult readers decoding is not a good predictor of comprehension skill. These findings are consistent with theories that focus on various high-level sources of differences in comprehension skill (Gernsbacher, Varner, & Faust, 1990; Nation & Snowling, 1998; Oakhill et al. 2003). In addition to the demonstration of a dissociation between lower-level and higher-level reading skills, I also show that decoding and other low-level skills such as spelling make only a very small contribution to the prediction of comprehension skill, and that vocabulary is the best predictor of comprehension ability in skilled adult readers. Because of the importance of vocabulary in my analysis, the results are especially consistent with hypotheses that emphasize word meaning (Nation & Snowling, 1998) or lexical quality (Perfetti, 2007; Perfetti & Hart, 2001) in comprehension, reinforcing the importance of lexical-semantic knowledge for adult readers with adequate decoding skills. However, future research is needed to tease apart the relationship between lexical-level semantics and comprehension level semantics, which incorporates pragmatics and syntax as well as the ability to integrate general knowledge with text and form a situation model. These findings are also largely consistent with the simple view of reading (Hoover & Gough, 1990), which posits that reading comprehension is the product of listening comprehension and decoding (see also Catts et al. 2006; Kirby & Savage, 2008 for recent tests and interpretations of the theory). From this simple equation it follows that, factors that contribute to general comprehension will account for the remaining variance in reading comprehension skill after decoding has been accounted for; thus, when skilled readers (with good decoding skills) are the test population, these general comprehension skills will subsume most of the variance in reading comprehension. This paper adds to a growing body of literature that reinforces the need to better understand factors that contribute to comprehension skill. I also want to emphasize that these findings are consistent with the notion that for most adult readers, and many children who have passed the transition to fluency stage, decoding ability is not a limiting factor in reading comprehension skill.

Acknowledgements

I thank Drs. Charles Perfetti, Kenneth Pugh, Stephen Frost, Daniel Mirman, W. Einar Mencl, Leonard Katz & David Braze for their helpful comments, support and editorial assistance. This research was supported in part by the Institute of Education Sciences (IES), Award No. R305G020006A: “Word learning and comprehension: New laboratory approach and classroom studies”,

Footnotes

1

Some researchers consider values of .3 to load on a component; importantly, there were no variables that would have changed components if we had observed cutoff of .3.

References

  1. August D, Francis DJ, Hsu HA, Snow CE. Assessing reading comprehension in bilinguals. The Elementary School Journal. 2006;107:221–238. [Google Scholar]
  2. Bell L, Perfetti CA. Reading skill: Some adult comparisons. Journal of Educational Psychology. 1994;86:244–255. [Google Scholar]
  3. Braze D, Tabor W, Shankweiler DP, Mencl EW. Speaking up for vocabulary: Reading skill differences in young adults. Journal of Learning Disabilities. 2007;40:226–243. doi: 10.1177/00222194070400030401. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Bryant FB, Yarnold PR. Principal-components analysis and exploratory and confirmatory factor analysis. In: Grimm LG, Yarnold PR, editors. Reading and understanding multivariate statistics. American Psychological Association; Washington, DC: 1995. pp. 99–136. [Google Scholar]
  5. Catts HW, Aldolf SM, Ellis- Weismer Language Deficits in Poor Comprehenders: A case for the Simple view of Reading. Journal of Speech language and Hearing. 2006;49:1278–293. doi: 10.1044/1092-4388(2006/023). [DOI] [PubMed] [Google Scholar]
  6. Curtis ME. Development of components of reading skill. Journal of Educational Psychology. 1980;72:656–669. [Google Scholar]
  7. Cutting LE, Scarborough HS. Prediction of reading comprehension: Relative contributions of word recognition, language proficiency, and other cognitive skills can depend on how comprehension is measured. Scientific Studies of Reading. 2006;10:277–299. [Google Scholar]
  8. Dien J, Beal DJ, Berg P. Optimizing principal components analysis of event-related potentials: Matrix type, factor loading weighting, extraction, and rotations. Clinical Neurophysiology. doi: 10.1016/j.clinph.2004.11.025. in press. [DOI] [PubMed] [Google Scholar]
  9. Fletcher JM. Measuring reading comprehension. Scientific Studies of Reading. 2006;10:323–330. [Google Scholar]
  10. Francis DJ, Fletcher JM, Catts H, Tomblin B. Dimensions affecting the assessment of reading comprehension. In: Paris SG, Stahl SA, editors. Children's reading comprehension and assessment. Lawrence Erlbaum Associates, Publishers; Mahwah, NJ: 2005. pp. 369–394. [Google Scholar]
  11. Frishkoff GA, Collins-Thomson K, Perfetti CA, Callan J. Measuring incremental changes in work knowledge; Experimental validation and implications for learning and assessment. Behavior Research Methods. 2008;40:907–925. doi: 10.3758/BRM.40.4.907. [DOI] [PubMed] [Google Scholar]
  12. Gough PB, Tumner WE. Decoding, reading, and reading disability. Remedial and Special Education. 1986;7:6–10. [Google Scholar]
  13. Gernsbacher MA, Varner KR, Faust M. Investigating differences in general comprehension skill. Journal of Experimental Psychology: Learning, Memory, and Cognition. 1990;16:430–445. doi: 10.1037//0278-7393.16.3.430. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Hagtvet BE. Listening comprehension and reading comprehension in poor decoders: Evidence for the importance of syntactic and semantic skills as well as phonological skills. Reading and Writing. 2003;16:505–539. [Google Scholar]
  15. Hendrickson AE, White PO. Promax: A quick method for rotation to oblique simple structure. British Journal of Statistical Psychology. 1964;17:65–70. [Google Scholar]
  16. Hoover WA, Gough PB. The simple view of reading. Reading and writing. 1990;2:127–160. [Google Scholar]
  17. Jackson NE. Are University students’ component skills related to their text comprehension and academic achievement? Learning and Individual Differences. 2005;15:113–139. [Google Scholar]
  18. Kaiser HF. Varimax solution for primary mental abilities. Psychometrika. 1960;25:153–158. [Google Scholar]
  19. Keenan JM, Betjemann RS. Comprehending the Gray Oral Reading Test without reading it: Why comprehension tests should not include passage-independent items. Scientific Studies of Reading. 2006;10:363–380. [Google Scholar]
  20. Kirby JR, Savage RS. Can the simple view deal with the complexities of reading? Literacy. 2008;42:75–82. [Google Scholar]
  21. Landi N, Perfetti CA. An electrophysiological investigation of semantic and phonological processing in skilled and less skilled comprehenders. Brain and Language. 2007;102:30–45. doi: 10.1016/j.bandl.2006.11.001. [DOI] [PubMed] [Google Scholar]
  22. Lundquist E. Phonological complexity, decoding, and text comprehension. under review.
  23. Marshall CM, Nation K. Individual differences in semantic and structural errors in children's memory for sentences. Educational and Child Psychology. 2003;20:7–18. [Google Scholar]
  24. Nagy W, Herman P. Breadth and depth of vocabulary knowledge: Implications for acquisition and instruction. In: McKeown MG, Curtis M, editors. The nature of vocabulary acquisition. Erlbaum; Hillsdale, NJ: 1987. pp. 19–36. [Google Scholar]
  25. Nation K, Snowling MJ. Semantic processing and the development of word-recognition skills: Evidence from children with reading comprehension difficulties. Journal of Memory and Language. 1998;39:85–101. [Google Scholar]
  26. Nation K. Reading comprehension difficulties. In: Snowling MJ, Hulme C, editors. The science of reading. Blackwell Publishing; Oxford: 2005. pp. 248–265. [Google Scholar]
  27. Nelson MJ, Denny EC. The Nelson-Denny Reading Test. Houghton Mifflin Company; 1973. [Google Scholar]
  28. Oakhill J. Individual differences in children's text comprehension. In: Gernsbacher MA, editor. Handbook of psycholinguistics. Academic Press; San Diego, CA: 1994. pp. 821–848. [Google Scholar]
  29. Oakhill J, Cain K. Children's difficulties in text comprehension: Assessing causal issues. Journal of Deaf Studies and Deaf Education. 2000;5:51–59. doi: 10.1093/deafed/5.1.51. [DOI] [PubMed] [Google Scholar]
  30. Oakhill JV, Cain K, Bryant PE. The dissociation of word reading and text comprehension: Evidence from component skills. Language and Cognitive Processes. 2003;18:443–468. [Google Scholar]
  31. Olson R, Forsberg H, Wise B, Rack J. Measurement of word recognition, orthographic, and phonological skills. In: Lyon GR, editor. Frames of reference for the assessment of learning disabilities: New views on measurement issues. Paul H. Brookes Publishing Co.; Baltimore: 1994. pp. 243–277. [Google Scholar]
  32. Perfetti CA. Reading ability. Oxford Press; New York: 1985. [Google Scholar]
  33. Perfetti CA. Reading ability: Lexical quality to comprehension. Scientific Studies of Reading. 2007;11:357–383. [Google Scholar]
  34. Perfetti CA, Hart L. The lexical basis of comprehension skill. In: Gorfien DS, editor. On the consequences of meaning selection: Perspectives on resolving lexical ambiguity. American Psychological Association; Washington, DC: 2001. pp. 67–86. [Google Scholar]
  35. Perfetti CA, Landi N, Oakhill JV. The acquisition of reading comprehension skill. In: Snowling MJ, Seidenberg MS, Hume C, editors. The Science of Reading: Handbook of Reading Research. Blackwell; Oxford: 2005. [Google Scholar]
  36. Perfetti CA, Hogaboam TW. Relationship between single word decoding and reading comprehension skill. Journal of Educational Psychology. 1975;67:461–469. [Google Scholar]
  37. Perfetti CA, Lesgold AM. Discourse comprehension and sources of individual differences. In: Carpenter PA, Just MA, editors. Cognitive processes in comprehension. Erlbaum; Hillsdale, N.J.: 1979. pp. 141–183. [Google Scholar]
  38. Perfetti CA, Marron MA, Foltz PW. Sources of comprehension failure: Theoretical perspectives and case studies. In: Cornoldi C, Oakhill J, editors. Reading comprehension difficulties: Processes and intervention. Lawrence Erlbaum; Mahwah, NJ: 1996. [Google Scholar]
  39. Protopapas GD, Sideridis G, Simos P, Mouzaki A. The development of lexical mediation in the relationship between text comprehension and word reading skills in Greek. Scientific Studies of Reading. 2007;11:165–197. [Google Scholar]
  40. Raven J, Raven J. Raven progressive matrices. In: McCallum RS, editor. Handbook of Nonverbal Assessment. Kluwer Academic/Plenum Publishers; New York: 2003. pp. 223–240. [Google Scholar]
  41. Shankweiler D. How problems of comprehension are related to difficulties in decoding. In: Shankweiler D, Liberman IY, editors. Phonology and reading disability: Solving the reading puzzle. The University of Michigan Press; Ann Arbor, MI: 1989. pp. 35–68. [Google Scholar]
  42. Shankweiler D, Lundquist E, Dreyer LG, Dickenson CC. Reading and spelling difficulties in high school students: Causes and consequences. Reading and Writing: An Interdisciplinary Journal. 1996;8:267–294. [Google Scholar]
  43. Shankweiler D, Lundquist E, Katz L, Stuebing KK, Fletcher J, Brady S, et al. Comprehension and decoding: Patterns of association in children with reading difficulties. Scientific Studies of Reading. 1999;31:69–94. [Google Scholar]
  44. Stanovich KE, West RF. Exposure to print and orthographic processing. Reading Research Quarterly. 1989;24:402–433. [Google Scholar]
  45. Stothard SE, Hulme C. A comparison of phonological skills in children with reading comprehension difficulties and children with decoding difficulties. Journal of Child Psychology & Psychiatry. 1995;36:399–408. doi: 10.1111/j.1469-7610.1995.tb01298.x. [DOI] [PubMed] [Google Scholar]
  46. Waters GS, Caplan D. The capacity theory of sentence comprehension: Critique of Just and Carpenter (1992). Psychological Review. 1996;103:761–77. doi: 10.1037/0033-295x.103.4.761. [DOI] [PubMed] [Google Scholar]
  47. Wagner RK, Muse AE, Tannenbaum KR. Vocabulary acquisition: Implications for reading comprehension. Guilford Press; New York: 2007. [Google Scholar]
  48. Wolf M, Goldberg O'Rourke A, Gidney C, Lovett M, Cirino P, Morris R. The second deficit: An investigation of the independence of phonological and naming-speed deficits in developmental dyslexia. Reading and Writing. 2002;15:43–72. [Google Scholar]
  49. Yuill NM, Oakhill JV. Children's problems in text comprehension: An experimental investigation. Cambridge University Press; 1991. [Google Scholar]

RESOURCES