Genetic and environmental mediation of the prediction from preschool language and nonverbal ability to 7-year reading
Abstract
We use a genetically sensitive design to examine the relationship between language and nonverbal ability at 4½ and reading skills at 7 years of age in a sample of more than 1,000 children participating in the Twins Early Development Study. We find that nonphonological as well as phonological measures of early language make significant contributions towards the prediction of reading at 7, and that nonverbal ability at 4½ is an equally strong predictor. With respect to aetiology, we find substantial genetic contributions towards the relationship between early language skills and reading at 7, as well as a trend towards shared environmental influences. The genetic continuity is not specific to the verbal domain, however, as we also find a substantial genetic relationship between nonverbal ability at 4½ and reading at 7.
It is relatively uncontroversial that early language abilities contribute to the ease with which children learn to read. What is more controversial is which areas of language are most important for literacy. A pervasive view is that the cognitive core underlying the development of reading is phonological processing, in both typical and disordered development: it has even been proposed that dyslexia should be defined in terms of a phonological deficit (Stanovich, 1986; Stanovich & Siegel, 1994).
Phonological awareness
Phonological processing includes several component process skills concerning speech sounds, from their perception, to their articulation, to metalinguistic awareness. This last component, phonological awareness, has received by far the most attention. There is an overwhelming amount of evidence, both correlational and experimental, showing that it is an important skill strongly associated with both normal reading acquisition (Bradley & Bryant, 1983; Wagner, Torgesen & Rashotte, 1994) and reading problems (see Goswami & Bryant, 1990; Wagner & Torgesen, 1987 for reviews). Phonological awareness requires conscious reflection on the organisation of the phonological system (Gombert, 1992), and relies on the ability to segment words according to their phonological structure. Typically, children from around the age of 4 are able to segment a word into syllables, and from the age of 5 or 6 into phonemes (Liberman, Shankweiler, Liberman, Fowler & Fischer, 1977). A large number of tasks aimed at measuring phonological awareness have emerged over the last two decades: these range from requiring the child to tap out the number of syllables in a word, to rhyme judgement and generation, to spoonerisms, in which one has to transpose the initial segments from two words, turning ‘John Lennon’ into ‘Lon Jennon’ (Perin, 1983).
Phonological awareness is important both for the initial development of the alphabetic principle and the formation of phoneme–grapheme correspondences, and thus for word reading (Byrne, 1998; Byrne & Fielding-Barnsley, 1989). It is worth bearing in mind, however, that although phonological awareness is typically regarded as a precursor to reading, it is also possible that reading itself contributes to the fine-tuning of phonological representations. In a study of illiterate adults in Portugal, it was found that participants could successfully complete a phonemic elision task after completing a literacy programme, but not before, suggesting that phonological awareness, especially phonemic awareness, developed as a consequence of learning to read (Morais, Cary, Algeria & Bertelson, 1979). In a recent review of the relevant literature, it has been argued that there is currently no unequivocal evidence that the direction of causation is from early phonological awareness skills to reading and spelling, rather than the other way around (Castles & Coltheart, 2004).
Phonological memory
Another phonological skill that may be implicated in literacy development is phonological memory. The ability to repeat nonwords accurately, which relies heavily on phonological memory (Gathercole & Baddeley, 1990), has been found to be predictive of literacy levels (Gathercole & Baddeley, 1993) and to be deficient in dyslexia (Brady, Poggie & Rapala, 1989; Kamhi & Catts, 1986; Snowling, 1981; Van Daal & Van der Leij, 1999). It seems likely that in order to map phonemes to graphemes, the phonological representation not only needs to be accurate but to be maintained long enough for robust mapping to occur. It may also be that phonological memory is a precursor to phonological awareness: the more robust a phonological representation, and the better maintained in memory, the easier it must be to reflect on and analyse. A deficit in phonological memory might therefore plausibly affect the acquisition of letter–sound correspondences both directly and indirectly through its effects on phonological awareness.
Expressive phonology
A third aspect of phonology that may be involved in reading development is expressive phonology. Infant speech perception and production involve the mapping of articulatory oral motor movements onto the acoustic features of the speech signal, which are then mapped onto developing phonological representations. Although many other factors are also implicated in this process (e.g. increasing vocabulary size forcing sharper phonological representations (Walley, 1993)), it is clear why there might be a link between speech and literacy. Despite this intuitive link, several studies (e.g. Bishop & Adams, 1990; Catts, 1993; Scarborough, 1990) suggest that, at least in the case of disordered speech development, the contribution of speech deficits to literacy problems is limited. On the other hand, at least one study has found that children with expressive speech disorders at 5 went on to have phonological awareness deficits at 6 and 7, and reading problems at 7 (Bird, Bishop & Freeman, 1995), and speech level at 3½ has been found to predict letter knowledge at 6 (Webster, Plante & Couvillion, 1997). In addition, more severe expressive phonological problems, particularly when measured by complex tasks such as word and nonword repetition, seem to be particularly predictive (Larrivee & Catts, 1999). Two recent studies explicitly compared two of the possible reasons for these mixed findings: the persistence of speech difficulties, and their comorbidity with other language deficits (Nathan, Stackhouse, Goulandris & Snowling, 2004; Raitano, Pennington, Tunick, Boada & Shriberg, 2004). Persistence and comorbidity were both found to increase literacy risk, although the studies disagreed on whether speech disorder was predictive on its own. Both sets of authors emphasised the link between early speech and language deficits and phonological awareness skills, which mediate the relationship with literacy.
Nonphonological language skills
The emphasis on phonological processes underlying literacy acquisition has somewhat obscured the role that other areas of language might play (Bishop & Snowling, 2004). Nonetheless, well-replicated findings from prospective investigations, both of typical development and of children with language impairments, show that semantic and syntactic skills are also important factors in accounting for reading outcomes. An early longitudinal investigation of these issues followed a sample of children identified with language impairment at 4 years of age (Bishop & Adams, 1990). Children whose language impairment had resolved by 5½ exhibited normal literacy development, whereas those who still had verbal deficits at 5½ tended to have both oral language and reading deficits at age 8½, particularly in reading comprehension as compared with reading accuracy. Much of the variance in literacy was accounted for by syntactic competence in preschool, with a much smaller contribution from early skills in expressive phonology.
Similar conclusions were reported by Catts (1993), who found that expressive and receptive language skills in kindergarteners (6-year-olds) with speech-language impairments were closely related to reading comprehension in the next 2 years. Phonological awareness and rapid automatised naming tasks were found to contribute independently to word recognition but not comprehension, and pure articulation deficits were unrelated to later reading problems.
A complementary approach to selecting children with early language impairments has been to select children who are considered to be at risk of dyslexia, on the basis of a family history of reading problems, and to follow their linguistic progress from early childhood (Gallagher, Frith & Snowling, 2000; Scarborough, 1990). These studies have generally found that the earliest indicators of future reading problems are quite broad-based, and include early deficits in vocabulary and syntax; less clear-cut results have been reported for the contribution of expressive phonological deficits to reading problems in these samples.
Recent work drawing on a large-scale epidemiological sample rather than a clinical population showed an even wider range of variables relating to reading achievement. Grammatical ability, phonological awareness, nonverbal IQ and rapid naming at 6 years of age all explained unique variance in reading comprehension at 8 and 10 years of age, although letter identification proved to be the strongest predictor. This pattern of results held both for children identified with language impairment in kindergarten (Catts, Fey, Tomblin & Zhang, 2002) as well as the whole epidemiological sample (Catts, Fey, Zhang & Tomblin, 2001).
Similar findings were reported in an investigation focusing on Head Start children: both phonological awareness and receptive vocabulary made independent contributions towards print knowledge (Dickinson, McCabe, Anastasopoulos, Peisner-Feinberg & Poe, 2003). However, Dickinson et al. used receptive vocabulary as their sole measure of nonphonological language. A more comprehensive assessment of oral language was made by the NICHD Early Child Care Research Network (2005); here, 3-year broad language skills contributed to first-grade word-decoding skills, and only part of this prediction was mediated by phonology. Furthermore, the authors emphasise that vocabulary is not an adequate stand-in for oral language skills in general, as other measures of semantics and grammar made additional and unique contributions towards the prediction of both decoding and reading comprehension. These studies did not include nonverbal ability in their analyses.
Whatever the eventual consensus concerning the best predictors of early literacy, it is important to note that prediction is not the same as causality: it is possible that preschool language or nonverbal skills are early indicators of a general underlying resource that is relevant to both the early skills and later reading. Studies looking at the aetiology of these cognitive domains can begin to address the issue of causality.
Genetic influences on language and literacy skills and disorders
Family studies consistently show that individuals with speech and language impairments are more likely than the general population to have relatives with reading impairments (Flax et al., 2003; Lewis, 1992). Although this suggests some shared aetiology, families share both genes and environments, and this type of study cannot disentangle their effects. One method to decompose the variance of a trait into genetic and environmental components is the twin design, which capitalises on the difference in genetic relatedness between identical and fraternal twin pairs. Identical twins are identical genetically, while fraternal twins are only 50% similar; by comparing twin–co-twin similarity across the two types of twin pairs, one can estimate the relative extent of genetic and environmental influence on individual differences in a given trait. These analyses are termed univariate, because of their focus on a single trait.
Twin studies have shown a significant genetic influence on both language and literacy, in the context of both typical and atypical development (see reviews by Bishop, 2002; DeFries & Alarcon, 1996; Stromswold, 2001). Genetic influences seem to play a significant role across many components of language: in two recent studies that included a wide range of language measures in a sample composed of children of the same age, there was consistent evidence of significant genetic effects on diverse areas of preschoolers' language skills, from syntax to phonology (Byrne et al., 2002; Kovas et al., 2005). While univariate analyses are useful for indicating whether genes or environments are more or less important for a given variable, they cannot provide any information on the relationships between variables. Multivariate genetic analyses, by contrast, can directly address the issue of whether traits are related because they share overlapping genetic influences, or because the same environmental factors are important for both traits.
Relatively few multivariate analyses have looked at the aetiological interrelationships between different components of language. The work so far has tended to find that much of the genetic influence on these different aspects of language is shared in common among them. That is, the genes that are important for individual differences in vocabulary are the same genes that are important for individual differences in receptive syntax. This result has been found both in parental reports of vocabulary and grammar in 2- and 3-year-olds (Dionne et al., 2003), and in direct psychometric assessments of 4½-year-olds (Hayiou-Thomas et al., in press). Some genetic specificity does remain, however, most notably in the relationship between expressive phonology and other areas of language (i.e. the genetic influences on these overlap substantially, but are not entirely the same; Hayiou-Thomas et al., in press).
For literacy, converging evidence from a number of independent studies suggests that genetic influences are important for component processes in reading. Univariate analyses have shown that a variety of tasks assessing phonology, fluency and orthographic skills are significantly heritable (Davis, Knopik, Olson, Wadsworth & DeFries, 2001; Gayán & Olson, 2001). For some of these, such as rapid naming, the genetic effects account for nearly all the variance, while for others, such as phonological awareness and letter identification, shared environmental factors also play an important role (Petrill, Deater-Deckard, Thompson, Schatschneider & DeThorne, in press).
Multivariate analyses have shown a pattern of a generally high genetic overlap among these skills (Davis et al., 2001; Gayán & Olson, 2001). More specifically, there appears to be a core genetic factor that is shared among phonological awareness, rapid naming and reading outcomes. In addition to this, there seem to be additional genetic influences that affect rapid naming and reading (but not phonological awareness), and additional shared environmental influences that affect phonological awareness and reading (but not rapid naming; Hohnen & Stevenson, 1999; Petrill, Deater-Deckard, Thompson, DeThorne & Schatschneider, in press). One of the few studies to consider the aetiological relationship between a broader range of language skills and reading assessed 6- and 7-year-old twins concurrently on measures of IQ, language, phonological awareness and literacy, and found that covariance between all four skills was primarily mediated by genetic influences (Hohnen & Stevenson, 1999).
The literature to date suggests an important role for genetic influences on diverse components of language and reading, and for the relationships among these skills. Although there has been some consideration of the aetiological relationship between language and reading, this has focused largely – although not exclusively – on ‘reading-related’ language skills, such as phonological awareness. However, the debate is still open as to how reading is related to broader oral language skills as well as to nonverbal ability, particularly with respect to the aetiology of these interrelationships. Further, previous work has focused on concurrent measurements of reading and phonological skills; in the current study, we are interested in the longitudinal relationship between early language and later reading.
Rationale for the current study
We were interested in examining the issue of which language skills in early childhood, as well as nonverbal ability, would be predictive of later reading ability, using a genetically sensitive design. The Twins Early Development Study (TEDS) is based on a large population-based sample of twins in the UK, and has been following their cognitive, linguistic and behavioural development since the second year of life. A subsample of TEDS was selected at 4 years of age for extensive in-home testing on a broad battery of language and nonverbal tests. Subsequently, the entire TEDS sample was assessed on early reading performance at 7 years of age. The current study focuses on those children for whom we have both language data at 4½ and reading data at 7. The first part of the study is a phenotypic analysis, capitalising on the large sample size to ask the question: which areas of language are most predictive of literacy skill at 7? Secondly, to what extent do genetic and environmental factors underpin the associations between early language, nonverbal ability and reading?
Method
Participants
The sampling frame for the present study was the TEDS, a longitudinal study of twins born in England and Wales in 1994, 1995 and 1996 (Trouton, Spinath & Plomin, 2002). After checking for infant mortality, all families identified by the UK Office for National Statistics (ONS) as having twins born in these years were invited to participate in TEDS when the twins were about 18 months old. The twins were assessed at 2, 3 and 4 years of age using parent questionnaires, which included measures of language, cognitive and behavioural development.
A subset of TEDS twins was tested at home on an extensive battery of language and nonverbal measures, at age 4½. This subsample was selected on the basis of parent report at age 4 on measures of vocabulary and grammar (McArthur–Bates Communicative Development Inventory UK Short Form – MCDI:UKSF; Dale, Price, Bishop & Plomin, 2003), and nonverbal ability (Parent Report of Children's Abilities – PARCA: Oliver et al., 2002). Part of the purpose of this subsample was to investigate impairments in language and cognitive development; with this in mind, the sample is overrepresented for children considered to be ‘at risk’ on the basis of scoring in the lowest 5% on the MCDI or PARCA at 4 years. The dataset used for the current study does not differentiate between the 512 twin pairs containing at least one ‘at-risk’ child, and the 310 ‘control’ twin pairs. We have described this combined sample in previous work, and shown that the measures have either normal or near-normal distributions (Kovas et al., 2005). Twin pairs were excluded where either member of the pair had any major medical or perinatal problems, documented hearing loss or organic brain damage. Participants were selected to be ethnically white, in order to avoid ethnic stratification in molecular genetic studies using DNA from this group; however, over 94% of the population of England and Wales is also white. Maternal education levels were also comparable both to the overall TEDS sample, as well as UK ONS census data. In all selected families, English was the only language spoken at home.
A total of 836 twin pairs participated in the in-home testing, and 3,474 twin pairs participated in the 7-year testing. The current sample includes all twin pairs who had data at both time points, and who did not fulfil any of the exclusionary criteria. The final sample included 1,270 children (221 MZ pairs, 226 same-sex DZ and 188 opposite-sex DZ pairs). A randomly selected member of each twin pair was selected from this sample for all phenotypic analyses (to account for the nonindependence of data in twin pairs). Genetic analyses use data from same-sex twin pairs only, as it is computationally too difficult to incorporate opposite-sex pairs into multivariate models.
Assessment of language and nonverbal skills at 4½ years
Children were seen at home, with each member of a twin pair tested by a different tester. Assessment sessions lasted approximately 1 hour.
Phonology. We used the Goldman–Fristoe test of articulation (Goldman & Fristoe, 1986) to assess expressive phonology: target phonemes are tested as the child names familiar pictures. Phonological working memory was assessed using Children's Test of Nonword Repetition (CNRep: Gathercole, Willis, Baddeley & Emslie, 1994). A 20-item version of the test was used, with 10 items at each of the two and three syllable lengths. Note that the CNRep also makes substantial input and output demands in addition to the phonological memory element (Snowling, Chiat & Hulme, 1991). Phonological awareness was assessed using a task derived from the work of Bird et al. (1995). In this receptive task, the child makes a rhyme judgement by matching a puppet's name to a rhyming object: for example: ‘Which of these things would Dan like? Spoon, Ring, Pan, Key?’ After four practice trials with feedback, a further eight items are administered.
Nonphonological language measures. Expressive semantics was measured using three tasks: (i) Word knowledge: oral vocabulary, from the McCarthy Scales of Children's Ability (MSCA; McCarthy, 1972), in which the child provides an oral definition of spoken words; (ii) MSCA verbal fluency, in which the child names as many examples of items as possible in a given category (e.g. ‘animals’) within 20 s; and (iii) The Renfrew Bus Story Test (Renfrew, 1997a), in which the child retells a story read by the narrator, while looking at the accompanying pictures. We used the bus story information score, which reflects the amount of story content that the child includes in their retelling.
Expressive grammar was measured using the grammar score from The Renfrew Action Picture Test (Renfrew, 1997b), which reflects use of inflectional morphology and function words during utterances elicited for scenes on a set of picture cards.
We assessed receptive language using the British Ability Scales verbal comprehension subtest (BAS; Elliot, Smith & McCulloch, 1996). In this task, the child arranges a set of toys following the examiner's instruction (e.g. ‘Put the house on each side of the car’). We used a subscale consisting of the last 11 items of the subtest, which required comprehension of grammatical morphology and syntax. The scores from the first section of this subtest, which consisted of items requiring only lexical comprehension, showed a clear ceiling effect, and were excluded from further analyses.
Finally, we used the Verbal Memory Words and Sentences subtest from the McCarthy Scales to assess verbal memory for meaningful materials (as opposed to specifically phonological short-term memory).
Nonverbal measures. Our nonverbal measure was a composite derived from four subtests of the MSCA Perceptual-Performance Index: Block Building (copying block arrangements), Puzzle Solving (assembling puzzles), Tapping Sequence (imitation of sequence tapped on a four-key xylophone) and Draw-a-Design (copying geometric designs). These tests were selected to have a minimal language load, based on previous factor analysis of these data (Colledge et al., 2002); the composite was created using the mean of the standardised scores of the four tests.
Assessment of reading at 7 years
Telephone measures of reading abilities. Twins were individually assessed by telephone on Form B of the test of word reading efficiency (TOWRE; Torgesen, Wagner & Rashotte, 1999). The TOWRE is a measure of word recognition ability that comprises two subtests: sight-word efficiency, which assesses fluency and accuracy in sight-word reading, and phonemic decoding efficiency, which assesses nonword reading. Test stimuli were mailed to families in a sealed package prior to the test sessions, with separate instructions that the package should not be opened until the time of testing. Twins in each pair were tested within the same test session and by the same tester, who was blind to zygosity.
Teacher-assessed reading achievement. General reading achievement was assessed by teachers' assessments on UK National Curriculum (NC) Key Stage 1 criteria for reading attainment (QCA Key Stage 1 Handbook, 1999. See Appendix B). These criteria are linked to the National Literacy Strategy, a statutory reading curriculum that provides literacy goals and instruction guidelines for teachers in England and Wales (DfEE, 1998). Teachers are required to rate children's reading ability on a five-point scale based on their knowledge of the child's reading achievement over the academic year. Teacher assessments were obtained by postal questionnaire during the spring semester. Agreement between NC teacher assessments of reading and scores on group-administered NC reading tests at Key Stage 1, obtained for a nationwide sample of 600,000 children, is good (Cohen's κ=.80; Dale, Harlaar & Plomin, 2005).
Reading composite. Previous analysis of the 7-year reading measures from the TEDS sample has found remarkably high correlations between the word and nonword reading subtests of the TOWRE (r=.83; Dale et al., 2005), replicating previous findings based on in-person testing (Torgesen et al., 1999). Furthermore, both phenotypic correlations and genetic correlations (explained below) between scores on the TOWRE- and teacher-rated reading were also found to be substantial (Dale et al., 2005; Harlaar, Dale & Plomin, in press). Given these results, we derived a reading composite by summing the standardised scores on the two TOWRE subtests and the teacher ratings. Although we treat this composite as our main outcome variable of interest in the current paper, we also conducted separate analyses with the word and nonword reading subtests, to examine the possibility that there are different relationships between early language skills with word and nonword reading.
Part 1: Phenotypic results and discussion
Results
The aim of the phenotypic analysis was to investigate whether some areas of oral language skill are more strongly associated with later reading outcomes than others. We performed a simple multiple regression analysis, with the 7-year reading composite as the outcome variable, and the 4½-year language and nonverbal measures as the predictor variables. We used z-scores with a mean of zero and standard deviation of one; standardisation for the 4½-year measures was based on the 310 ‘control’ twin pairs described in the Participants section. Seven-year measures were standardised on the full TEDS sample. Scores were corrected for the linear effects of age and sex, as these can inflate twin similarity (McGue & Bouchard, 1984). Although this is important primarily for the genetic analyses, we used the same scores in the phenotypic regression for the sake of consistency.
The results of the multiple regression are presented in Table 1. The adjusted total R2 is .28, indicating that just over a quarter of the total variance in reading skill at age 7 (as measured by the reading composite) is accounted for by oral language and nonverbal skills at 4½. Although the zero-order correlations are moderate for all measures, three of our nine language measures made significant contributions towards reading, with the largest beta coefficient for verbal fluency, then Action Pictures grammar, and then nonword repetition. Oral vocabulary was marginally significant (p=.07). Nonverbal ability also made a significant contribution towards the total variance in reading skill, which was in fact greater than that of any single language measure.
| Standardised coefficients | Significance | Zero-order correlation | |
|---|---|---|---|
| β | p | r | |
| Constant | .03 | ||
| Nonverbal composite | .19 | .00 | .40 |
| Bus story information | .00 | .94 | .33 |
| AP grammar | .14 | .01 | .39 |
| BAS comprehension | −.01 | .90 | .28 |
| Oral vocabulary | .09 | .07 | .36 |
| Verbal fluency | .14 | .00 | .41 |
| Verbal memory | .06 | .18 | .36 |
| Phonological awareness | .06 | .15 | .31 |
| GF articulation | −.04 | .45 | .26 |
| Nonword repetition | .12 | .03 | .35 |
- Note: Variables in bold represent significant predictors.
The results were similar when the individual reading scores, i.e. teacher-rated reading, TOWRE word reading or TOWRE nonword reading scores, were entered as the dependent variable in the regression, instead of the reading composite. AP grammar and nonverbal ability consistently made significant predictions to reading. Verbal fluency predicted teacher-rated reading and (marginally, at p=.066) nonword reading. There was a nonsignificant trend for phonological awareness to predict teacher-rated reading (β=.08, p=.09) and TOWRE nonword reading (β=.10, p=.08).
Discussion
We found that language and nonverbal ability at 4½ years of age was a moderate predictor of reading skills at 7 years of age, accounting for 28% of the total variance in reading ability in our sample of over 1,000 children. The prediction derived primarily from three out of nine language measures: these included expressive semantics (verbal fluency and oral vocabulary), expressive morphosyntax (AP grammar) and phonology (nonword repetition). The greatest predictive power among the language measures was held by verbal fluency, but at least as large a contribution to the variance in later reading skill was due to nonverbal ability.
With respect to the components of language that are most predictive of later reading skills, we found evidence in support of roles for both phonological processes and broader oral language skills. The phonological measure that we may have expected to play the largest role in developing reading was phonological awareness. However, although its simple correlation with reading was similar to that of other measures, it did not make a unique prediction, with only a nonsignificant trend for a contribution towards the variance in teacher-rated reading and nonword reading (but not TOWRE word reading or the reading composite). It is possible that phonological awareness would have proved a stronger predictor if we had measured this skill at the level of the phoneme as well as at the level of the rhyme, as there is evidence to suggest that it is phoneme-level processes that are particularly relevant for decoding (Muter & Snowling, 1998). Indeed, the question of whether awareness of rhyme per se is important for reading is the subject of some debate, with some research pointing to an important role (Goswami & Bryant, 1990), and other studies finding that the predictive power of phonological awareness for early reading is held by segmentation and not rhyming skills (Hulme, Hatcher, Nation, Brown, Adams & Stuart, 2002); our current results are more consistent with the latter view. Although ideally the current study would have attempted to assess phonemic awareness in addition to rhyme awareness, it is possible that we would have encountered a floor effect, given the very young age of the sample.
Nonword repetition did predict significant variance in reading, unlike our measure of phonological awareness, supporting previous evidence of an important role for phonological short-term memory in reading (Brady, Shankweiler & Mann, 1983; Hulme & Snowling, 1992). However, it is not entirely clear how best to interpret this finding, as previous work on this sample found that nonword repetition and our measure of expressive phonology (Goldman–Fristoe articulation) were very closely related, both phenotypically and aetiologically (Hayiou-Thomas et al., in press). It has been pointed out previously that the nonword repetition task does make demands on expressive phonology (Snowling et al., 1991). However, the finding that Goldman–Fristoe articulation did not make a significant contribution towards reading does suggest that we may be picking up on some differential prediction from the phonological memory aspect of nonword repetition.
An important predictor from early language to later reading was verbal fluency. Although it is primarily a semantic measure, the ‘fluency’ element of the task presumably draws on the speed of information processing; an influential hypothesis in the literacy literature suggests that in addition to phonological skills, the speed of processing, especially for retrieval processes, plays an important role in learning to read (Catts et al., 2001). The association between verbal fluency and reading in our study might therefore reflect a shared reliance on the speed of processing. However, the finding in the current study that oral vocabulary – not a speeded task – also contributed significant variance to the prediction of 7-year reading suggests that early semantic skills do mediate the prediction from both verbal fluency and word knowledge at 4½ to reading at 7. The finding that the grammar measure from the Action Pictures task also contributed significant variance supports the stance that the oral language skills relevant to reading extend beyond vocabulary knowledge (a point recently raised by the NICHD Early Child Care Research Network, 2005). The finding of a strong contribution from grammatical skills and vocabulary is consistent with previous prospective studies (Bishop & Adams, 1990; Catts et al., 2002). However, it is noteworthy that several of the other ‘broad’ language measures from 4½ did not significantly predict reading, including story retelling (bus story) and sentence comprehension (BAS comprehension). It may be that story retelling and sentence comprehension, although they are slightly weaker predictors of early reading in the current study, will turn out to have a more important role to play in later reading comprehension.
The current paper focuses on the longitudinal prediction from early language measures to reading, and we have not discussed the interrelationships among the 4½-year language measures. We have examined this issue in previous work, and found a substantial amount of genetic and shared environmental overlap, despite generally moderate (r∼.3) phenotypic correlations for most measures (with the exception of nonword repetition and articulation, which are strongly correlated; Hayiou-Thomas et al., in press). This is relevant for interpreting the current findings, in that the prediction to reading derives from a broad range of the earlier language measures.
Most of our conclusions about the predictors of reading refer to decoding skills. This is clearly the case for the TOWRE (word and nonword reading); the teacher assessment is broader, and could potentially include information about comprehension, but given the age of the children, and the empirically high correlations – both phenotypic and genetic – that we have previously found between teacher-assessed reading and the TOWRE scores, it seems probable that this too reflects mostly decoding skills. However, the ultimate goal of reading is not decoding print, but understanding written language; although word reading is a prerequisite for reading comprehension, there are additional abilities, such as oral comprehension, that come into play.
One plausible scenario is that phonological skills predict decoding ability, and that wider oral language skills assume importance when the child moves beyond decoding and into reading comprehension. Some of the previous evidence from large prospective studies is consistent with this pattern: Catts et al. (2002) found that second-grade grammar accounted for nearly 4% of the variance in reading comprehension in fourth grade, while word recognition had a unique contribution from phonological awareness over the same time period. Our results are not altogether consistent with this pattern, however, since we found that nonphonological language (and indeed nonverbal) skills were predictive of decoding skills. A possible explanation is that decoding and comprehension are not entirely separable; for example, the opaque orthography of a language like English means that many words cannot be decoded using phonological rules (Nation & Snowling, 1998, 2004). In such circumstances, the ability to use sentence contexts to aid the reading of new words takes on greater importance (Adams, 1990). Nonetheless, it is likely that different components of language will be more directly relevant for decoding versus comprehension. It will be interesting to observe what happens with reading comprehension in our sample: assessment of reading comprehension at 10 years of age is currently under way, and we will be able to examine these issues using the large TEDS sample.
Most of our discussion has focused on the early language skills that are predictive of reading, and this reflects the focus of the literature in this field. However, it is important to note that early nonverbal ability was found to be at least as strong a predictor of later reading skills as any of our language measures. This too is consistent with previous results (Bishop & Adams, 1990; Catts et al., 2001, 2002; Gallagher et al., 2000). However, while it is relatively straightforward to imagine the mechanisms through which early language skills might be useful in learning to read – that awareness of speech sounds makes it possible to connect letters to sounds, for example, or that knowledge of words makes it easier to recognise them in print – it is harder to imagine what paths might lead from block building to reading. It seems more likely that block building acts as a proxy for general cognitive ability, or ‘g’. Finally, even with early nonverbal ability included, most of the variance in reading at 7 is unaccounted for by our measures at 4½. There are almost certainly relevant skills at 4½ that we did not measure in our sample (e.g. phoneme awareness, as discussed earlier, or letter–sounds knowledge).
Part 2: Genetic analysis, results and discussion
Genetic analysis
Genetic analyses were based on the twin design, which capitalises on the fact that identical (MZ for monozygotic) twins share 100% of their varying DNA while fraternal twins (DZ for dizygotic) share on average 50%, just like any other sibling pair (Plomin, DeFries, McClearn & McGuffin, 2001). If the members of an MZ twin pair are more similar to each other on a given trait than the members of a DZ pair, this difference can be attributed to genetic influences. Comparing the members of a twin pair on a single trait yields an estimate of univariate heritability, which is the extent to which individual differences on the trait are caused by genetic as opposed to environmental factors. In multivariate analyses, this model is extended to examine the origins of the covariance between two or more measures by comparing Trait 1 in Twin 1 with Trait 2 in Twin 2 (Martin & Eaves, 1977). If the cross-trait cross-twin correlation is higher in MZ than in DZ pairs, this is evidence for some shared genetic relationship (Dale et al., 2000).
The actual determination of the parameters in the models described below used the structural equation modelling package Mx (Neale, Boker, Xie & Maes, 2002). The basic genetic model uses the maximum-likelihood method to obtain parameter estimates for the effects of additive genetic (A), shared environmental (C) and nonshared environmental (E) influences on a given trait. The additive genetic and shared environmental influences are what make the children within a twin pair similar to each other, while the nonshared – or unique – environmental influences contribute towards differences within the pair. The E parameter also includes the effects of measurement error. The model assumes that there are no effects of nonadditive genetics, nonrandom mating or gene–environment interaction.
We used a Cholesky decomposition approach to model raw twin data; although space does not permit a detailed explanation of the multivariate analysis, such explanations are available elsewhere (Loehlin, 1996; Neale & Maes, 1999). We included a total of 11 variables: the nonverbal composite, nine language measures and the reading composite. This model generates estimates of aetiological factors that influence all 11 measures; then influences on measures 2–11, once the variance shared with measure 1 is accounted for; then influences on measures 3–11, once variance shared with the first two measures is accounted for; and so on until the final measure.
From these estimates, the model derives univariate heritability and environmentality estimates (i.e. the total genetic and environmental influence on each measure), as well as bivariate heritability and environmentality. These are all related constructs: univariate heritability estimates the proportion of the total phenotypic variance of a trait that is attributable to genetic factors, while bivariate heritability indexes the proportion of the covariance between two traits (e.g. vocabulary level and reading) that can be attributed to shared genetic factors. That is, both univariate and bivariate heritability (and environmentality) are concerned with breaking down phenotypic variance into genetic and environmental components. The model also derives estimates of the genetic and environmental correlations between each pair of measures. The genetic correlation is conceptually quite different from heritability estimates, in that it indicates the overlap of the genetic influences on two traits regardless of their heritabilities or the magnitude of the phenotypic correlation between them. That is, the genetic correlation between two traits can be substantial even if their heritabilities or the phenotypic correlation between them are modest. Intuitively, the genetic correlation can be viewed as the likelihood that a gene that affects one trait will affect the second trait. The environmental correlation is an analogous parameter for environmental influences.
Results
The 11-variable Cholesky decomposition fit the data well, in that there was a nonsignificant decrease in fit from the saturated model to the Cholesky model. The −2 log-likelihood measure of model fit for the saturated model was 23924.526, df=9052, AIC=−31653.71. The fit for the Cholesky was −2LL=24276.56, df=9381, AIC=−33321.70. The difference in these was change in −2LL=352.034, change in df=329, p=.183.
The focus of the current paper is on the bivariate relationships between the 4½-year language and nonverbal measures and reading at 7. We therefore present the bivariate genetic and environmental correlations, as well as the bivariate heritability and environmentality estimates in Table 2, and we will limit our discussion to these parameters. The full set of parameter estimates from the model is included in Appendix A, but we have discussed in earlier work the univariate aetiology of the 4½-year measures (Colledge et al., 2002; Kovas et al., 2005) and the 7-year reading measures (Harlaar, Spinath, Dale & Plomin, 2005), as well as the interrelationships among the language variables at 4½ (Hayiou-Thomas et al., in press).
| r p | r g | r c | r e | Biv. h2 | Biv. c2 | Biv. e2 | |
|---|---|---|---|---|---|---|---|
| 1. Nonverbal composite | .35 | .65 (.41–.94) | .00 (.00–.98) | .00 (.00–.12) | 1.00 (.59–1.00) | .00 (.00–.38) | .00 (.00–.10) |
| 2. Bus story | .32 | .23 (.01–.50) | .84 (.32–1.00) | .03 (.00–.16) | .37 (.01–.79) | .61 (.22–.94) | .02 (.00–.12) |
| 3. AP grammar | .32 | .55 (.22–.89) | .30 (.00–.90) | .09 (.00–.22) | .74 (.28–1.00) | .18 (.00–.60) | .09 (.00–.23) |
| 4. BAS comprehension | .24 | .43 (.12–.98) | .45 (.00–1.00) | .00 (.00–.07) | .78 (.20–1.00) | .29 (.00–.81) | .00 (.00–.04) |
| 5. Oral vocabulary | .31 | .47 (.17–.92) | .63 (.13–.99) | .00 (.00–.00) | .60 (.21–1.00) | .52 (.10–.88) | .00 (.00–.02) |
| 6. Verbal fluency | .32 | .65 (.33–.99) | .33 (.00–.99) | .00 (.00–.10) | .86 (.40–1.00) | .17 (.00–.58) | .00 (.00–.13) |
| 7. Verbal memory | .31 | .54 (.19–.94) | .33 (.00–.87) | .09 (.00–.21) | .69 (.22–1.00) | .23 (.00–.65) | .08 (.00–.20) |
| 8. Phonological awareness | .26 | .52 (.18–1.00) | .37 (.00–.97) | .00 (.00–.11) | .79 (.24–1.00) | .22 (.00–.72) | .00 (.00–.12) |
| 9. GF articulation | .20 | .46 (.14–.94) | .02 (.00–.79) | .00 (.00–.11) | 1.00 (.29–1.00) | .02 (.00–.67) | .00 (.00–.14) |
| 10. Nonword repetition | .29 | .44 (.17–.88) | .65 (.00–1.00) | .09 (.00–.21) | .73 (.23–1.00) | .18 (.00–.61) | .09 (.00–.22) |
- Note: r p indicates the phenotypic correlation; rg, genetic correlation; rc, shared environmental correlation; re, nonshared environmental correlation; Biv. h2, bivariate heritability, Biv. c2, bivariate shared environmentality, Biv. e2, bivariate nonshared environmentality.
The phenotypic correlations derived from the model, which are very similar to those reported in Table 1 (the small differences are due to the fact that the current estimates are based on same-sex twins only, and are derived from the model), are approximately .3 for most of the language measures as well as the nonverbal composite.
Genetic and environmental correlations. The genetic correlations between the language measures and reading are substantial – ranging from 0.43 to 0.65 – and are strikingly similar across all measures. The sole exception is the bus story, which has a lower rg at .23 (although it is not significantly different from the other rg estimates, as indicated by the overlapping confidence intervals). These results indicate a strong genetic overlap between the early language variables and the 7-year reading scores: of the genetic influences affecting language ability at 4½ (with the exception of narrative skills measured by the bus story), approximately half continue to play a role at age 7, and form part of the substantial genetic influence on reading skills at this age. The 4½-year nonverbal composite also has a strong genetic correlation with 7-year reading, at 0.65, suggesting that nearly two thirds of the genetic influences on nonverbal ability in the preschool years continue to age 7 and affect literacy acquisition.
The shared environmental correlations are much more variable. Once again, the bus story stands out: it has the highest rc, which, with the oral vocabulary estimate, is the only one that is significantly different from zero (i.e., lower 95% confidence intervals greater than zero). The remainder of the language measures show moderate shared environmental correlations with reading (although these are not significantly different from zero). It should be noted that the lack of significance for these estimates reflects the limited statistical power of the twin method to detect shared environmental influences. The nonverbal composite, as well as GF articulation, stand out with rc estimates of zero. For nearly all of the 4½-year language measures, and particularly two of the measures dealing with expressive semantics – bus story and oral vocabulary – it seems that shared environmental factors are not only important at the univariate level (Kovas et al., 2005), but that these influences continue to play a role nearly 3 years later in the acquisition of literacy skills. By contrast, articulation and nonverbal ability are both relatively unaffected by shared environmental factors at 4½ (Colledge et al., 2002; Kovas et al., 2005), which accordingly show no role in their prediction of reading at 7.
The nonshared environmental correlations are effectively at zero across the board, suggesting that there are no environmental effects unique to one child in the twin pair that contribute to both the 4½- and 7-year measures. In the univariate case, the nonshared environmental term incorporates measurement error, and it is difficult to disentangle this from genuine environmental effects that contribute towards the differences between children in a family. This is less of an issue in multivariate models, however, as measurement error is not shared across different measurement occasions.
Bivariate heritability and environmentality. The bivariate heritability and environmentality estimates complement the story told by the genetic and environmental correlations. Most of the language measures show a consistent picture of substantial bivariate heritability estimates – all of them significantly greater than zero – and at least some shared environmental influence contributing to the relationship with reading – although again, only bus story and oral vocabulary have bivariate c2 estimates significantly different from zero. This means that the moderate phenotypic relationships between 4½-year language skills and 7-year reading are a result of both genetic and shared environmental factors that they have in common. By contrast, the phenotypic relationship between 4½-year nonverbal ability and reading at 7 is entirely because of shared genetic influences, reflected in the bivariate heritability estimate of 1.00.
The odd ones out are once more the nonverbal composite and the articulation measure, which have bivariate c2 estimates of zero. That is, shared environmental factors do not contribute towards the phenotypic relationship between 4½-year articulation and nonverbal ability, and 7-year reading skill; the phenotypic relationship is entirely accounted for by shared genetic influences (consistent with the high bivariate heritability and genetic correlations reported above). Conversely, genetic influences make a relatively modest contribution to the phenotypic relationship between bus story scores at 4½ and reading at 7, as reflected in the bivariate heritability estimate of 0.37; the phenotypic relationship is therefore mostly down to shared environmental influences that affect both early narrative skills and reading.
The bivariate nonshared environmentality estimates are negligible for all measures: none of the phenotypic correlations between 4½- and 7-year scores are accounted for by nonshared environmental factors.
Discussion
Our primary finding from the twin analysis is that the genetic effects that operate on language at 4½ and reading at 7 are substantially correlated: that is, about half of the genetic influences that affect early language skills continue to play a role later on, when children are learning to read. However, this genetic continuity is not specific to the verbal domain, as we found a similarly high genetic correlation between the genetic effects on nonverbal ability at 4½ and reading at 7. Complementing the strong genetic correlations were the high estimates of bivariate heritability, which index the proportion of the phenotypic relationship between early language and reading that is attributable to genetic effects. These findings from individual measures of language ability at 4½ are consistent with ongoing work from TEDS examining the relationship between latent factors of reading and language, which also show substantial genetic relationships (Harlaar, Hayiou-Thomas, Rijsdijk, Dale & Plomin, under review).
We should point out that the analyses presented in this paper are based on a sample that includes the full range of ability, and while it is tempting to generalise to disorders of language and reading, it is possible that a different pattern of results might emerge were one to focus on the lower end of the distribution. This would be a worthwhile direction for future work.
We have previously found that our language measures at 4½ are highly intercorrelated at an aetiological level, with genetic and shared environmental influences acting entirely through the commonalities of these diverse aspects of language (Hayiou-Thomas et al., in press). We had speculated that differentiation among the language measures might emerge in their relationship with reading. It was possible, for example, that we might find that a genetic continuity between early language and reading is held by the phonological measures. Such a finding would have been consistent with previous work suggesting a genetic core shared by phonological and decoding skills (Petrill et al., in press; Gayán & Olson, 2001). This position was only partially supported: although there was evidence of a genetic overlap between all three phonological measures and reading, similar genetic relationships were found across all the other language measures as well. An interesting point did arise in the case of GF articulation and nonword repetition, which although closely related aetiologically in concurrent analyses, did show some differentiation in how they relate to reading. While GF articulation had only genetic and no environmental links to reading, nonword repetition had both genetic and environmental mediation of its relationship with reading. This is particularly intriguing in light of the fact that previous work has emphasised the genetic but not environmental relationship between phonological short-term memory and reading (Bishop, 2001; Bishop, Adams & Norbury, 2004), including in defining a broader phenotype for dyslexia (Pennington & Lefly, 2001), and also in highlighting possible genetic specificity for nonword repetition and nonword reading deficits (Raskind, Hsu, Berninger, Thomson & Wijsman, 2000; Snowling, Gallagher & Frith, 2003).
With respect to the nonverbal composite, we have found in concurrent analyses that there is a substantial genetic overlap (rg∼.60) between language and nonverbal ability at 4½ (Colledge et al., 2002), and also a strong genetic relationship between general cognitive ability (a composite of nonverbal and verbal tests) and reading at 7 (Harlaar, Hayiou-Thomas & Plomin, 2005). In addition, language outcome at 4½ is at least as well predicted by earlier nonverbal ability (at ages 2, 3 and 4) as it is by earlier language skills, and this nonverbal prediction is mediated to a significant extent by genetic effects (Oliver, Dale & Plomin, 2004). Our current finding of strong genetic continuity – and a total lack of an environmental relationship – between nonverbal skills at 4½ and reading at 7 adds to this very consistent picture of a genetic core of cognitive abilities that acts both concurrently and longitudinally.
Our results, showing genetic continuity from both nonverbal and language skills at 4½ to the early stages of reading at 7, fit well with a ‘generalist genes’ perspective that proposes that genetic influences on behaviour – including cognition – are likely to be broad (Plomin & Kovas, 2005). Many genes have pleiotropic effects – that is, they affect many different traits – and consequently genetic relationships may well be found between traits that we would not necessarily expect to be related on phenotypic grounds. A complementary idea to pleiotropy is that complex traits are likely to be influenced by a large number of genetic factors, of varying effect sizes, referred to as quantitative trait loci. From this perspective, it should not be surprising to observe shared genetic effects of the kind we found between nonverbal ability at 4½ and reading at 7, in addition to the more intuitive links between early language and reading. The prediction for molecular genetic work is that when QTLs are identified that contribute to the heritability of oral language, nonverbal ability and reading, most of the QTLs will be the same for the three domains. However, the existence of an association – even at an aetiological level – does not in itself illuminate the mechanisms through which these aetiological factors will affect early language and reading. It may be that language and reading skills are correlated only because they share some genetic substrate that gives rise to a general cognitive resource, useful to both language and reading (rather than contributing to skill-specific processes). However, it is also possible that there are additional direct influences of early language on reading; for example, that the number of words in one's mental lexicon is in itself relevant for how easily one learns to read those words. In recent work, we have compared different models of causality directly in order to address this issue of likely mechanisms that might explain the aetiological association between early language and reading reported in the present paper (Harlaar et al., under review).
We have emphasised the genetic continuity between early cognitive skills and reading, but we did also find some evidence for environmental links between early language and reading, and it is interesting to speculate on the types of environmental factors that might contribute to our shared environmental correlations and bivariate environmentality. Many aspects of the environment have been proposed as facilitators of early language development, including the amount and complexity of verbal stimulation, responsiveness to children's communicative intents, story-telling and bookreading and decontextualised language use. Several of these might plausibly be expected to facilitate reading development as well. Home environments, including factors such as the number of books in the home, have also been shown to be important for reading; moreover, it has previously been suggested that this may be mediated by its effect on verbal abilities (Whitehurst & Lonigan, 1998).
Our data suggest that whatever environments are common to reading and language development are ones that are shared by children growing up in the same family. Our findings of zero re and bivariate e2 are very likely explained in part by the fact that E estimates include measurement error, which is not shared between the 4½- and 7-year assessments. However, we also found very low re estimates between latent factors of language and reading; in latent factor models measurement error is not included in the re estimate (Harlaar et al., under review). The implication is that child-specific experiences – such as a history of middle ear disease – are not likely to be important in explaining continuity between early language ability and reading.
We can also make the prediction that, unlike genes, environmental factors that are found to be important to the language–literacy relationship are not likely to be ones that are important for nonverbal abilities. Thus, while phenotypically we were struck by the similar prediction to reading from diverse language and nonverbal skills, aetiologically it appears that there does exist some differentiation, and that this comes from the environment, while the commonalities across these three domains are mediated genetically.
Acknowledgements
We gratefully acknowledge the ongoing contribution of the parents and children in the Twins Early Development Study (TEDS). TEDS is supported by a programme grant (G0500079) from the UK Medical Research Council and grants from the US National Institutes of Health (HD-44454 and HD-49861).
Appendices
Appendix A
Full set of parameter estimates from the 11-variable Cholesky model (Table A1, Table A2, Table A3a, Table A3b, Table A3c, Table A4).
| 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| (a) Additive genetic latent factors (along the columns) | |||||||||||
| 1. Nonverbal composite | .439 | ||||||||||
| 2. Bus story | .109 | .264 | |||||||||
| 3. AP grammar | .168 | .049 | .047 | ||||||||
| 4. BAS comprehension | .106 | .000 | .037 | .141 | |||||||
| 5. Oral vocabulary | .075 | .039 | .020 | .036 | .063 | ||||||
| 6. Verbal fluency | .096 | .044 | .005 | .001 | .070 | .054 | |||||
| 7. Verbal memory | .048 | .081 | .031 | .001 | .051 | .019 | .003 | ||||
| 8. Phonological awareness | .099 | .003 | .060 | .000 | .012 | .023 | .002 | .037 | |||
| 9. GF articulation | .144 | .000 | .058 | .018 | .013 | .013 | .039 | .000 | .013 | ||
| 10. Nonword repetition | .149 | .025 | .025 | .001 | .010 | .058 | .049 | .000 | .017 | .000 | |
| 11. Reading composite | .291 | .014 | .031 | .003 | .012 | .172 | .100 | .016 | .036 | .002 | .003 |
| (b) Shared environment latent factors (along the columns) | |||||||||||
| 1. Nonverbal composite | .111 | ||||||||||
| 2. Bus story | .028 | .292 | |||||||||
| 3. AP grammar | .001 | .102 | .119 | ||||||||
| 4. BAS comprehension | .047 | .068 | .005 | .029 | |||||||
| 5. Oral vocabulary | .078 | .256 | .028 | .018 | .005 | ||||||
| 6. Verbal fluency | .070 | .055 | .014 | .000 | .007 | .023 | |||||
| 7. Verbal memory | .089 | .087 | .044 | .005 | .043 | .019 | .002 | ||||
| 8. Phonological awareness | .008 | .085 | .061 | .003 | .001 | .000 | .000 | .000 | |||
| 9. GF articulation | .015 | .018 | .015 | .066 | .005 | .061 | .000 | .000 | .000 | ||
| 10. Nonword repetition | .000 | .029 | .001 | .002 | .006 | .002 | .000 | .000 | .000 | .000 | |
| 11. Reading composite | .000 | .130 | .029 | .003 | .000 | .003 | .000 | .000 | .000 | .000 | .000 |
| (c) Nonshared environment latent factors (along the columns) | |||||||||||
| 1. Nonverbal composite | .450 | ||||||||||
| 2. Bus story | .003 | .304 | |||||||||
| 3. AP grammar | .000 | .033 | .481 | ||||||||
| 4. BAS comprehension | .002 | .000 | .002 | .562 | |||||||
| 5. Oral vocabulary | .001 | .009 | .002 | .003 | .367 | ||||||
| 6. Verbal fluency | .010 | .010 | .009 | .004 | .010 | .520 | |||||
| 7. Verbal memory | .003 | .021 | .007 | .002 | .007 | .006 | .432 | ||||
| 8. Phonological awareness | .001 | .002 | .001 | .005 | .002 | .000 | .010 | .586 | |||
| 9. GF articulation | .003 | .002 | .001 | .018 | .007 | .001 | .002 | .002 | .487 | ||
| 10. Nonword repetition | .009 | .001 | .008 | .011 | .000 | .004 | .005 | .005 | .080 | .506 | |
| 11. Reading composite | .000 | .000 | .001 | .001 | .004 | .000 | .002 | .000 | .000 | .001 | .146 |
- Note: These are squared estimates rather than path coefficients, and reflect the percentage of variance explained by each latent factor.
| 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| (a) Bivariate genetic correlations (rg) | |||||||||||
| 1. Nonverbal composite | 1.00 | ||||||||||
| 2. Bus story | .541 | 1.00 | |||||||||
| 3. AP grammar | .798 | .793 | 1.00 | ||||||||
| 4. BAS comprehension | .610 | .350 | .650 | 1.00 | |||||||
| 5. Oral vocabulary | .567 | .652 | .752 | .183 | 1.00 | ||||||
| 6. Verbal fluency | .598 | .662 | .709 | .462 | .789 | 1.00 | |||||
| 7. Verbal memory | .452 | .740 | .768 | .383 | .870 | .921 | 1.00 | ||||
| 8. Phonological awareness | .648 | .444 | .352 | .197 | .393 | .613 | .360 | 1.00 | |||
| 9. GF articulation | .695 | .349 | .727 | .409 | .499 | .441 | .395 | .283 | 1.00 | ||
| 10. Nonword repetition | .669 | .590 | .766 | .543 | .465 | .649 | .558 | .437 | .876 | 1.00 | |
| 11. Reading composite | .654 | .232 | .551 | .429 | .468 | .652 | .539 | .519 | .460 | .436 | 1.00 |
| (b) Bivariate shared environment correlations (rc) | |||||||||||
| 1. Nonverbal composite | 1.00 | ||||||||||
| 2. Bus story | .296 | 1.00 | |||||||||
| 3. AP grammar | .064 | .667 | 1.00 | ||||||||
| 4. BAS comprehension | .562 | .812 | .630 | 1.00 | |||||||
| 5. Oral vocabulary | .449 | .912 | .780 | .759 | 1.00 | ||||||
| 6. Verbal fluency | .646 | .738 | .639 | .819 | .849 | 1.00 | |||||
| 7. Verbal memory | .554 | .688 | .694 | .811 | .730 | .618 | 1.00 | ||||
| 8. Phonological awareness | .220 | .768 | .968 | .792 | .831 | .752 | .798 | 1.00 | |||
| 9. GF articulation | .000 | .220 | .407 | .376 | .059 | .282 | .121 | .471 | 1.00 | ||
| 10. Nonword repetition | .037 | .831 | .681 | .716 | .663 | .567 | .663 | .787 | .622 | 1.00 | |
| 11. Reading composite | .000 | .843 | .298 | .454 | .631 | .326 | .326 | .367 | .017 | .652 | 1.00 |
| (c) Bivariate nonshared environment correlations (re) | |||||||||||
| 1. Nonverbal composite | 1.00 | ||||||||||
| 2. Bus story | .093 | 1.00 | |||||||||
| 3. AP grammar | .024 | .253 | 1.00 | ||||||||
| 4. BAS comprehension | .053 | .029 | .066 | 1.00 | |||||||
| 5. Oral vocabulary | .040 | .157 | .103 | .093 | 1.00 | ||||||
| 6. Verbal fluency | .131 | .143 | .159 | .104 | .169 | 1.00 | |||||
| 7. Verbal memory | .079 | .215 | .175 | .078 | .165 | .185 | 1.00 | ||||
| 8. Phonological awareness | .042 | .059 | .046 | .093 | .070 | .012 | .151 | 1.00 | |||
| 9. GF articulation | .073 | .000 | .031 | .191 | .125 | .076 | .080 | .093 | 1.00 | ||
| 10. Nonword repetition | .121 | .000 | .103 | .141 | .043 | .113 | .126 | .119 | .400 | 1.00 | |
| 11. Reading composite | .000 | .032 | .092 | .000 | .000 | .000 | .087 | .000 | .000 | .086 | 1.00 |
| 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 1. Nonverbal composite | 0.439 | ||||||||||
| 2. Bus story | 0.708 | 0.374 | |||||||||
| 3. AP grammar | 0.927 | 0.473 | 0.264 | ||||||||
| 4. BAS comprehension | 0.686 | 0.376 | 0.543 | 0.284 | |||||||
| 5. Oral vocabulary | 0.625 | 0.340 | 0.406 | 0.173 | 0.234 | ||||||
| 6. Verbal fluency | 0.572 | 0.477 | 0.476 | 0.405 | 0.402 | 0.270 | |||||
| 7. Verbal memory | 0.516 | 0.428 | 0.421 | 0.320 | 0.393 | 0.499 | 0.233 | ||||
| 8. Phonological awareness | 0.804 | 0.400 | 0.299 | 0.225 | 0.279 | 0.545 | 0.252 | 0.236 | |||
| 9. GF articulation | 1.000 | 0.781 | 0.677 | 0.419 | 0.649 | 0.581 | 0.606 | 0.363 | 0.298 | ||
| 10. Nonword repetition | 0.794 | 0.711 | 0.651 | 0.546 | 0.559 | 0.633 | 0.528 | 0.477 | 0.495 | 0.333 | |
| 11. Reading composite | 1.000 | 0.369 | 0.738 | 0.775 | 0.605 | 0.861 | 0.693 | 0.791 | 1.000 | 0.725 | 0.681 |
- Univariate heritabilities are presented on the diagonal, and bivariate heritabilities on the off-diagonal. These are square path coefficients, and represent the proportion of the phenotypic covariance between each pair of variables explained by additive genetic effects.
| 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 1. Nonverbal composite | 0.111 | ||||||||||
| 2. Bus story | 0.180 | 0.32 | |||||||||
| 3. AP grammar | 0.034 | 0.337 | 0.222 | ||||||||
| 4. BAS comprehension | 0.230 | 0.585 | 0.349 | 0.149 | |||||||
| 5. Oral vocabulary | 0.319 | 0.565 | 0.495 | 0.669 | 0.385 | ||||||
| 6. Verbal fluency | 0.245 | 0.388 | 0.310 | 0.410 | 0.439 | 0.168 | |||||
| 7. Verbal memory | 0.354 | 0.410 | 0.388 | 0.547 | 0.471 | 0.294 | 0.289 | ||||
| 8. Phonological awareness | 0.112 | 0.523 | 0.614 | 0.535 | 0.619 | 0.430 | 0.507 | 0.157 | |||
| 9. GF articulation | 0.000 | 0.354 | 0.270 | 0.217 | 0.077 | 0.228 | 0.161 | 0.383 | 0.180 | ||
| 10. Nonword repetition | 0.008 | 0.316 | 0.181 | 0.178 | 0.349 | 0.149 | 0.238 | 0.239 | 0.093 | 0.039 | |
| 11. Reading composite | 0.000 | 0.609 | 0.180 | 0.292 | 0.515 | 0.167 | 0.230 | 0.224 | 0.015 | 0.182 | 0.165 |
- Univariate environmentalities are presented on the diagonal, and bivariate environmentalities on the off-diagonal. These are square path coefficients, and represent the proportion of the phenotypic covariance between each pair of variables explained by shared environmental effects.
| 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 1. Nonverbal composite | 0.450 | ||||||||||
| 2. Bus story | 0.111 | 0.307 | |||||||||
| 3. AP grammar | 0.039 | 0.190 | 0.514 | ||||||||
| 4. BAS comprehension | 0.085 | 0.039 | 0.108 | 0.566 | |||||||
| 5. Oral vocabulary | 0.057 | 0.095 | 0.099 | 0.158 | 0.381 | ||||||
| 6. Verbal fluency | 0.183 | 0.135 | 0.214 | 0.185 | 0.159 | 0.562 | |||||
| 7. Verbal memory | 0.131 | 0.161 | 0.191 | 0.133 | 0.136 | 0.207 | 0.478 | ||||
| 8. Phonological awareness | 0.084 | 0.077 | 0.087 | 0.240 | 0.102 | 0.025 | 0.242 | 0.606 | |||
| 9. GF articulation | 0.144 | 0.000 | 0.053 | 0.364 | 0.275 | 0.191 | 0.233 | 0.253 | 0.522 | ||
| 10. Nonword repetition | 0.199 | 0.000 | 0.168 | 0.276 | 0.091 | 0.218 | 0.234 | 0.284 | 0.411 | 0.628 | |
| 11. Reading composite | 0.000 | 0.022 | 0.082 | 0.000 | 0.000 | 0.000 | 0.077 | 0.000 | 0.000 | 0.094 | 0.155 |
- Univariate environmentalities are presented on the diagonal, and bivariate environmentalities on the off-diagonal. These are square path coefficients, and represent the proportion of the phenotypic covariance between each pair of variables explained by nonshared environmental effects.
| 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 1. Nonverbal composite | 1.00 | ||||||||||
| 2. Bus story | 0.309 | 1.00 | |||||||||
| 3. AP grammar | 0.293 | 0.527 | 1.00 | ||||||||
| 4. BAS comprehension | 0.315 | 0.303 | 0.328 | 1.00 | |||||||
| 5. Oral vocabulary | 0.291 | 0.567 | 0.460 | 0.272 | 1.00 | ||||||
| 6. Verbal fluency | 0.359 | 0.441 | 0.398 | 0.316 | 0.493 | 1.00 | |||||
| 7. Verbal memory | 0.280 | 0.510 | 0.453 | 0.308 | 0.517 | 0.463 | 1.00 | ||||
| 8. Phonological awareness | 0.260 | 0.330 | 0.294 | 0.227 | 0.331 | 0.284 | 0.336 | 1.00 | |||
| 9. GF articulation | 0.247 | 0.149 | 0.301 | 0.284 | 0.203 | 0.215 | 0.172 | 0.207 | 1.00 | ||
| 10. Nonword repetition | 0.322 | 0.293 | 0.349 | 0.306 | 0.232 | 0.308 | 0.294 | 0.257 | 0.557 | 1.00 | |
| 11. Reading composite | 0.353 | 0.317 | 0.316 | 0.243 | 0.309 | 0.325 | 0.309 | 0.263 | 0.204 | 0.286 | 1.00 |
Appendix B
National Curriculum Key Stage 1 (KS1) Teacher Assessment Scale for Reading W Not yet functioning at Level 1
- 1
Pupils recognise familiar words in simple texts. They use their knowledge of letters and sound–symbol relationships in order to read words and to establish meaning when reading aloud. In these activities they sometimes require support. They express their response to poems, stories and nonfiction by identifying aspects they like.
- 2
Pupils' reading of simple texts shows understanding and is generally accurate. They express opinions about major events or ideas in stories, poems and nonfiction. They use more than one strategy, such as phonic, graphic, syntactic and contextual, in reading unfamiliar words and establishing meaning.
- 3
Pupils read a range of texts fluently and accurately. They read independently, using strategies appropriately to establish meaning. In responding to fiction and nonfiction, they show understanding of the main points of express preferences.
- 4+
Reading is substantially more advanced than most pupils at Level 3.
