A number of papers have reported heritability estimates from the Colorado twin study as the sample have increased in size. Table 4 summarises the most recent heritability estimates for word recognition (a combined single word and prose reading measure) and spelling to come out of the Colorado study (Brooks, Fulker and DeFries, 1990; Olson, Forsberg and Wise, 1994) and estimates from the London twin study (Stevenson, Graham, Fredman and McLoughlin,1987;
Stevenson, 1991). The estimates from the London study are 10 adjusted. The table also shows the extreme group heritability estimate for spelling following a
collaborative analysis (DeFries, Stevenson, Gillis and Wadsworth, 1991).
Table 4 Heritability estimates for reading and spelling abiiity
Word recognition Spelling
London study 0.29 0.73 Colorado study 0.46 0.47 h"g London study ---- 0.47 Colorado study 0.47 0.48 Collaborative analysis ---- 0.61
Results from the Colorado study are highly consistent, with similar estimates for word recognition and spelling ability both in the extreme group and when looking at normal variation. Approximately half of the variance is estimated to be due to genetic variation in their sample. These results are now highly reliable with data on approximately 300 pairs of twins. The results from the London study show a noticeably higher heritability for spelling (0.73) compared to that for word
recognition (0.29). This difference in heritability estimates in the London study for the two types of literacy ability is explained by Stevenson, Graham, Fredman and McLoughlin (1987) from a developmental perspective. They suggest that genetic etiology may differ as a function of age and by the age of 13 years large
differences in reading ability between children will have been minimised by reading instruction. Differences are still evident at this age, however, in spelling ability which is more heavily dependent on phonetic knowledge. This suggestion of age
related changes in heritability for reading was directly tested using data from the
Colorado study where subjects were divided into a younger age group (aged 8-11 years) and an older age group (aged 12-20 years) and genetic models fitted to the groups separately (Wadsworth, Gillis, DeFries and Fulker, 1989). Although
differences were found in h^g estimates for reading in younger children (0.60 for 8- 11 year olds) and older children (0.29 for 13 - 20 year olds) the difference between the two estimates was not significant. However, these results provided some support for Stevenson's hypothesis with their finding that genetic etiology differed more as a function of age for reading than for spelling. For spelling heritability estimates were more similar for the different age groups. Another possible explanation for the differences in heritability estimates from the two studies is provided by Olson et al (1989). Their explanation draws on the fact that the causes of individual differences between people will differ for different populations.
Samples with greater variability in educational and social support for reading, they argue, would be likely to yield lower heritability estimates. That is greater variability in environmental causes of differences will act to reduce heritability estimates. This would explain the lower heritability estimate for reading in the London sample where environmental variability is likely to be large.
In order to increase the size of the sample thereby increasing the reliability of the estimates of h^g, a collaborative analysis of the spelling data was carried out with both the London and Colorado samples. The estimates from the two studies from this analysis was highly similar; 0.61 for the London sample and 0.62 for the Colorado sample (DeFries, Stevenson, Gillis and Wadsworth, 1991). This was more striking given the fact that the two studies had recruited subjects in very different ways and had even used different measures in their test batteries. This adds weight to the reliability of the results.
Heritability estimates for aspects of reading other than word recognition, such as reading comprehension, have not been the focus of much discussion in the literature. Stevenson et al (1987) found 51% of the variance in comprehension to be due to genetic factors, but heeds caution in interpretation of this result due to ceiling effects on the Neale at the age of 13 years. Extreme group heritability for reading comprehension appears to be lower. DeFries, Fulker and LaBuda (1987) report an estimate of 0.38 for reading comprehension and Olson et al (1994) an estimate of 0.27. In this latter analysis shared environmental factors were a particularly strong influence on comprehension (c^g = 0.52).
4.2.1.1 Multiple routes
Literacy is clearly a complex ability and so there are likely to be a range of genetic and environmental influences on such an ability. It is therefore probably
inappropriate to look for a single genetic influence. By subscribing to a theory which assumes there are different processes taking place in reading, such as the dual-route theory, it is possible to look for these separate influences.
Both of the above twin studies were grounded in the dual-route theory of reading which assumes that there are two routes to reading; one is via a phonological route, the other via an orthographic route. Assessment of the phonological route can be obtained by a non-word reading task and assessment of the orthographic route by an exception/irregular word reading task or a pseudohomophone judgment task (subjects have to identify the ‘real word’ when the other choice is a
phonologically identical nonword; e.g. rain, rane). Both the London and Colorado study assessed these two reading routes independently enabling identification of separate genetic influences on the two types of reading . Both used a non-word reading task for the phonological route. To assess orthographic route functioning, the London study used an irregular/exception reading word reading task, the Colorado study a pseudohomophone judgment task.
The first sets of analysis to emerge from these two studies (Stevenson 1991 ; Olson et al 1989; Olson, Connors and Rack, 1990) had a clear message. Both hfand h% estimates for phonological coding were relatively high and significant; the estimates for orthographic coding were much lower and not significant (see table 5 below).
Table 5 Heritability estimates for phonological and orthographic reading
1 / Phonological coding Orthographic coding
London study 0.50 0.04
Colorado study 0.52 0.20
London study 0.82 -0.66
Colorado study 0.47 0.28
The negative estimate of h% in the London study for orthographic coding is nonsensical and probably reflects the small numbers used in this analysis. The interpretation of these results at the time was that phonological coding ability is the mediator of genetic influences on reading. Orthographic coding ability is most strongly influenced by environmental factors (Stevenson, 1991; Olson et al, 1989). This pattern of results fitted well with previous research in the developmental literature showing the importance of segmental language skills such as
phonological awareness in determining reading skills. This led both sets of authors to suggest that the heritable precursor to reading ability and disability is segmental language ability which determines later phonological coding ability in reading. Recent results to come out of the Colorado study do indeed show substantial genetic influences on phonological awareness (see section 4.3.1) and a bivariate analysis showed a high genetic correlations between phonological awareness and reading (Olson, Forsberg and Wise, 1994). These data, then, are consistent with there being a genetic influence on reading disability which is a result of an
underlying heritable influence on phonological coding. The results also support a ‘dual-route’ perspective by indicating different developmental etiologies for
phonological and orthographic coding.
This finding also ties in well with Stevenson et al’s (1987) finding of a higher heritability estimate for spelling than for reading ability if we consider the strategies we use for the two tasks. Frith (1980) suggests that we use different cognitive strategies for reading and spelling. When spelling we use a phonological strategy with recourse to orthographically informed information only where necessary. With reading we employ a more orthographic strategy, recognising words on the basis of distinct features. If spelling is carried out by a more phonological strategy and
phonological ability is highly influenced by genetic factors, then this would explain the high heritability estimate for spelling found in the 13 year old data.
Finally, the Colorado group have provided support from their twin data that heritability of reading disability is higher in subjects with low phonological skills relative to their word recognition abilities as compared to a group with high phonological skills relative to word recognition (Olson et al 1991). H% for the former was 0.74 and for the latter was 0.54, the difference between which just reached significance. This is also consistent with the growing belief that a deficit in phonological coding ability is heritable and leads to poor word recognition.
The picture, then, of two routes to reading, one mainly influenced by genetic factors and the other mainly influenced by environmental factors was becoming well
accepted until some more recent analyses which has emerged from both the London and Colorado data. With a larger sample and with the data appropriately ‘double-entered’ in the DePries and Fulker (1988) analysis, the estimates to come out of the recent Colorado data are somewhat different (Olson et al 1994). Deficits in word recognition yielded a similar heritability estimate to previous analyses (h^g = 0.47) and deficits in phonological coding was still quite substantial (h^g = 0.59). However, the estimate of h^g for deficits in orthographic coding were significantly larger in this analysis (h^g = 0.56). Phonological and orthographic processing abilities correlated to the value of 0.43 and in a bivariate extension of the DeFries and Fulker multiple regression analysis, 66% of this covariance was found to be due to shared genetic factors which indicates a substantial common genetic etiology for phonological and orthographic processing.
A recent analysis on the London data looking at individual differences across the range of ability also found a somewhat different picture to the earlier results (Hohnen and Stevenson, in prep). In this analysis in order to get a more reliable measure of phonological and orthographic route processing, many different reading test scores were used including tasks that did not directly assess either the
phonological or orthographic coding ability. This amounted to approximately 15 different assessments including tasks such as reading aloud and silently, a homophone recognition task (subject has to say whether two words sound the same ‘bored’ and ‘board’) and a crosscase matching task (subjects has to say whether two words, one upper and one lower case, are the same). A factor
analysis was then carried out on all 15 tasks (oblique rotation) which yielded three main factors. One was clearly an IQ factor and the other two were separate, yet correlated reading factors. The two reading factors could not be clearly identified as ‘phonological’ and ‘orthographic’ factors since they had loadings from all tasks in the battery. Factor scores were calculated for each subject and behavioural
genetics models fitted to that data using a structural equation modeling technique. Both reading factors showed substantial genetic influences (reading factors 1 = 0.61 ; reading factor 2 = 0.70). More importantly, in a Cholesky decomposition analysis there were found to be two separate genetic influences on reading. The first was shared between the two reading factors and the second loaded just on the second reading factor.
This recent picture to emerge regarding the heritability of literacy is therefore somewhat different to the previous picture. The orthographic route to reading appears to be as heritable as the phonological route (Olson et al 1994) and the genetic covariance between phonological and orthographic deficits is high. Olson et al (1994) argue that a shared genetic influence on phonological and orthographic ability is more supportive of Ehri’s (1989,1992) integrative view of the development of phonological and orthographic coding skills rather than supporting the dual-route conceptualisation where the two routes are considered to be independent of each other. However, they also point out that it is not a direct test of the two theories. The two processes could still have separate mechanisms or anatomical substrates as would be implicated in the dual-route theory but have the same genetic etiology. The other recent finding to come out of both the London study and the Colorado study is that there is more than one genetic influence on reading suggesting that it is correct to continue to study different aspects of reading.
It is possible that this complicated picture is related to developmental changes in the processes causing individual differences in reading. The ages of the children used in the Colorado study varied from 8 to 20 years. In the London study children were 13 years of age. It may be that strategies in reading will be influenced by genetic factors to different degrees according to the stage of reading acquisition. A developmental approach will offer some clarity to this somewhat complicated