Reading self-perceived ability, enjoyment and achievement: A genetically informative study of their reciprocal links over time.

Longitudinal Associations Between Reading Achievement and Reading Motivation

Although several studies report modest to moderate correlations between reading achievement and several aspects of reading motivation, the findings are mixed with respect to the developmental nature of this association (e.g., Baker & Wigfield, 1999 ; Guthrie et al., 2006 ). One unresolved issue is how the motivation–achievement association develops. Several theories have been put forward addressing the causal ordering in the emergence of the motivation–achievement relationship. Early theories of the association between achievement and motivation favored unidirectional approaches. Two contrasting early theoretical frameworks are the Self-Enhancement Model and the Skill Development Model.

According to the Self-Enhancement Model, individual differences in motivation influence subsequent development of academic performance ( Calsyn & Kenny, 1977 ). Confident and interested readers are more invested in learning and mastering reading skills through frequent reading, and this frequent print exposure further results in better reading skills ( Calsyn & Kenny, 1977 ). Support for this model comes from early educational experimental programs, demonstrating that interventions designed to increase motivation lead to significant improvements in children’s reading ability (e.g., Guthrie et al., 1996 , 2006 ; Wigfield, Guthrie, Tonks, & Perencevich, 2004 ). However, most of these studies did not consider the potential link from achievement to motivation.

This influence of achievement on subsequent motivation is central to the Skill Development Model ( Calsyn & Kenny, 1977 ). For example, children at risk of reading failure are more likely to encounter difficulty and frustration in their early reading experiences, which may in turn lead to decreased motivation to read. The support for this model has been inconsistent. For example, one intervention study failed to observe improvements in children’s reading motivation as a consequence of improved reading skills in a group of unskilled readers ( Morgan, Fuchs, Compton, Cordray, & Fuchs, 2008 ). However, several longitudinal studies have supported the temporal precedence of achievement in the reading motivation–achievement relationship in samples of several ages—from early elementary school to middle school ages (e.g., Aunola, Leskinen, Onatsu-Arvilommi, & Nurmi, 2002 ; Chapman & Tunmer, 1997 ; Skaalvik & Valas, 1999 ). These studies utilized cross-lagged longitudinal analyses in which the longitudinal effect of one construct on another is estimated beyond the stability of each construct and the concurrent correlation between constructs. Specifically, these studies demonstrated that individual differences in children’s reading performance predicted subsequent variation in children’s reading motivation, whereas reading motivation failed to predict subsequent reading performance ( Aunola et al., 2002 ; Chapman & Tunmer, 1997 ; Skaalvik & Valas, 1999 ). However, these studies involved relatively small samples and may have been underpowered to detect reciprocal links between reading motivation and achievement.

The reciprocal relationship is central to a third theoretical framework, according to which achievement and motivation have a mutual influence on one another ( Morgan & Fuchs, 2007 ). The reciprocal model has been supported by longitudinal studies that have explored the motivation–achievement relation in several academic domains including literacy and mathematics (e.g., Guay et al., 2003 ; Luo, Haworth, & Plomin, 2010 ; Marsh & Martin, 2011 ; Muijs, 1997 ).

Several methodological differences may explain the inconsistencies found among previous studies with respect to the temporal and causal ordering between reading achievement and reading motivation. Differences in sample size and sample characteristics, study design, and statistical methods could all contribute to the discrepancies in the literature. For example, some studies examined children in the normal range of reading ability (e.g., Guthrie et al., 1996 ), whereas others focused on poor readers (e.g., Morgan et al., 2008 ). Some studies used experimental designs but only examined immediate or short-term outcomes (e.g., Guthrie et al., 1996 ), while others relied on correlational designs to investigate longer-term outcomes (e.g., Marsh & Martin, 2011 ).

Genetic and Environmental Etiology

Examining the genetic and environmental etiology of the longitudinal links between reading motivation and reading achievement can provide new insights into processes through which the two constructs interact. Research exploring factors contributing to variation in academic motivation and its association with achievement has largely focused on the role of environments ( Deci & Ryan, 2008 ; Stipek, 1996 ; Wigfield & Eccles, 2000 ). In particular, family environment, relationships with parents, parents’ and teachers’ educational expectations and attitudes, teachers’ instructional style and quality, and teacher–student and peer relationships have all been found to be important for academic motivation ( Deci & Ryan, 2008 ; Stipek, 1996 ; Wigfield & Eccles, 2000 ). A number of recent studies, using genetically informative approaches, have demonstrated that genetic factors are also involved in explaining individual differences in academic motivation ( Kovas et al., 2015 ).

For example, a recent international twin study of over 13,000 children demonstrated that genetic factors account for approximately 40% of individual differences in self-perceived ability and enjoyment of learning in numerous academic domains, including language, mathematics, and science ( Kovas et al., 2015 ). This was consistent across a wide age range and across six countries that were included in the study. Environmental influences stemmed entirely from unique individual experiences and did not contribute to similarity in academic motivation in children raised in the same family. This study suggests that resemblance among family members in academic motivation is entirely attributable to genetic influences, whereas dissimilarities among family members are largely explained by individual specific environmental factors. Even objectively shared environments, such as family educational resources and classroom environments, seem to be nonshared in terms of the actual experience.

Several studies examined the genetic and environmental etiology of the concurrent and longitudinal relations between academic motivation and academic achievement. For example, in a sample of 13-year-old twins from Germany, the contemporaneous correlations between motivation and academic performance in language and mathematics were mostly explained by genetic factors ( Gottschling, Spengler, Spinath, & Spinath, 2012 ). In the large U.K. Twins Early Development Study (TEDS), academic self-perceived ability and overall academic performance of 9-year old children correlated primarily for genetic reasons ( Greven et al., 2009 ). The study also found that the link from self-perceived ability at age 9 to achievement at age 12 was mostly explained by genetic factors.

Using the same TEDS sample, Luo et al. (2010) examined the longitudinal cross-lagged relations between a domain general composite of self-perceived ability and academic performance between ages of 9 and 12. In line with the reciprocal model, modest mutual links were found between domain general academic motivation and achievement. These cross-lagged reciprocal relations were mediated largely through genetic pathways ( Luo et al., 2010 ). Only one study has examined the etiology of the reciprocal association between motivation and achievement in a domain specific context. This investigation, also using TEDS data, explored the cross-lagged associations between motivation and achievement specific to mathematics ( Luo, Kovas, Haworth, & Plomin, 2011 ). The prediction from teacher-rated mathematics achievement at age 9 to subsequent mathematics motivation at age 12 was attributable to genetic factors, whereas the link from early motivation to subsequent achievement was mediated through both genetic and child-specific environmental pathways ( Luo et al., 2011 ).

Overall, findings from genetically informative twin studies point to the importance of genetic influences and child-specific environmental experiences in the etiology of academic motivation in diverse academic domains. Shared environmental factors are found to have negligible effects on individual differences in academic motivation. Additionally, although the longitudinal association between domain general motivation and achievement is largely mediated by genetic factors, the domain-specific association between mathematics achievement and motivation is affected by both genetic and nonshared environmental factors. These differences in the etiology of longitudinal links in domain-general versus mathematics specific achievement and motivation suggest potential differences in the underlying mechanisms and provide rationale for the study of other specific domains, such as reading. The present study used a genetically sensitive cross-lagged approach to explore the longitudinal association between reading motivation and reading achievement. Based on the existing literature summarized above, we propose the following hypotheses:

Reciprocal longitudinal links of similar strength exist between reading motivation (enjoyment and self-perceived ability) and reading achievement.

Similar to the domain of mathematics, both genetic and nonshared environmental factors contribute to the observed longitudinal cross-lagged associations between reading motivation and reading achievement.

Reading achievement

At age 9/10 and age 12, reading achievement was measured via the Reading Comprehension subtest of the Peabody Individual Achievement Test (PIAT; Markwardt, 1997 ). Children were asked to read a series of sentences and to select the one picture (out of four choices) that best depicts the meaning of the sentence. The PIAT included a total of 89 items arranged in the order of increasing difficulty. For example, one of the initial items was “Some kittens are in the bed”. The test became increasingly more complex and one of the final items was “The verdant countryside is prodigiously arable; however, a squalid domicile sullies the otherwise exquisite panorama.” Children were given up to 20 seconds to read each sentence and another 20 seconds to make their choices. A total reading achievement score was computed by summing the points across all 89 items.

Analytic Strategies

After running descriptive and correlation analyses, we applied structural equation modeling to examine the longitudinal relations between reading achievement and reading motivation, as well as the underlying genetic and environmental etiologies of these longitudinal associations. We conducted these analyses using the OpenMx package for R ( Neale et al., 2016 ; R Core Team, 2015 ).

In order to test our first hypothesis, we fitted a phenotypic cross-lagged model ( Figure 1a ). The cross-lagged model allows for the estimation of the strength of the link from reading motivation at age 9/10 to reading achievement at age 12, and of the opposite link from reading achievement at age 9/10 to reading motivation at age 12. The cross-lagged associations are estimated independently of the stability of the measures and their initial contemporaneous correlations. In order to formally compare the magnitude of the cross-lagged links, we constrained them to be equal. This allowed us to examine whether such constraints would worsen model fit, indicating differences in the magnitude of the paths.

Phenotypic cross-lagged model (panel a) and ACE cross-lagged model (panel b, c, and d). S and A matrices respectively capture symmetric and asymmetric relations. T matrix captures the impact of A, C, and E components on the total phenotypic variance of each variable. In the ACE cross-lagged model, S and A matrices are further decomposed into genetic (A; panel b), shard environmental (C; panel c), and nonshared environmental (E; panel d) components. Achieve = reading achievement; motive = reading motivation; 9/10 = age 9/10; 12 = age 12.

We used the twin design to test our second hypothesis. The twin method allows for the examination of the relative contribution of genetic and environmental factors to the longitudinal relations between reading achievement and reading motivation. The method is based on the comparison of the concordance between monozygotic (MZ) twins, who share 100% of their genetic make-up, and dizygotic (DZ) twins, who share on average 50% of their segregating genes. Genetic and environmental influences can be calculated by comparing correlations for MZ and DZ twins for the same trait (intraclass correlations). A stronger intraclass correlation between MZ twins than between DZ twins indicates that genetic factors are involved in explaining individual differences in that trait. This allows for the decomposition of the total variance of a trait into heritability, shared environmental, and nonshared environmental influences.

Heritability (A) refers to the proportion of the phenotypic (i.e., observed) individual differences attributable to genetic influences. The remaining variance in the trait is further divided into shared and nonshared environmental influences. Shared environment (C) refers to any nongenetic influences that contribute to twin similarities. Nonshared environment (E) refers to any nongenetic influences that contribute to dissimilarities between two twins raised in the same family, and includes measurement error.

The twin method can be extended to examine the etiology of the covariance between multiple traits. Multivariate models are based on the cross-twin cross-trait correlations. Cross-twin cross-trait correlations describe the association between two traits, with twin 1’s score on the first trait correlated with twin 2’s score on the second trait. Cross-twin cross-trait correlations are computed separately for MZ and DZ twins. A higher cross-twin cross-trait correlation for MZ than for DZ twins indicates that genetic factors have a degree of influence on the phenotypic variance shared by two traits. For example, in the present study, the cross-twin cross-trait correlation between reading motivation at age 9/10 and reading achievement at age 9/10 was .22 for MZ twins and .05 for DZ twins. This suggests that genetic factors are implicated in the etiology of the covariance between reading motivation at age 9/10 and reading achievement at age 9/10.

Specifically, to test our second hypothesis we applied the ACE cross-lagged model ( Figure 1b to Figure 1d ). This model allowed us to examine the etiologies of the cross-lagged associations between reading motivation and reading achievement. The limitation of previously used cross-lagged models, using a multivariate Cholesky decomposition approach (see Figure 3 and Figure 4 ), is that the two cross-lagged paths can only be estimated in two separate models, prohibiting direct comparisons of their effects (phenotypically or etiologically). The ACE cross-lagged model used in this study overcomes this limitation by estimating all the paths within the same model.

Cholesky Cross-lagged Model A. This model was used to examine the cross-lagged association between reading achievement at age 9/10 and reading motivation at age 12. See Figure 1 for abbreviations.

Cholesky Cross-lagged Model B. This model was used to examine the cross-lagged association between reading motivation at age 9/10 and reading achievement at age 12. See Figure 1 for abbreviations.

The ACE cross-lagged model is based on the Reticular Action Model definition ( McArdle & McDonald, 1984 ): C = F (I—A) − 1 S (I—A) −1 ′ F ′ where I is the identity matrix, S the matrix defining two-way relationships or symmetric relationships (i.e., variances and covariances), A is the matrix defining one-way relationships or asymmetric relationships (i.e., stability and cross-lagged paths in the case of cross-lagged model), and F is the filter matrix defining observed variables (not used here). The A and S matrices are n × n matrices, where n is the number of observed variables. In the ACE cross-lagged model, the twin design allows us to decompose the variance and covariance into the genetic ( Figure 1b ), shared environmental ( Figure 1c ), and nonshared environmental ( Figure 1d ) components, using the formulae reported below. The formulae were introduced into the model as matrix algebra t

Twin Correlations

Table 3 presents the intraclass correlations between measures of reading achievement and reading motivation separately for MZ and DZ twins. Twin correlations were substantially larger for MZ than for DZ twins for reading motivation at both waves, indicating genetic but negligible shared environmental influences; the same was observed for reading achievement at age 12. The MZ correlation was stronger than the DZ correlation also for reading achievement at age 9/10. However, the correlation between MZ twins did not double that of DZ twins, indicating both genetic and shared environmental influences on reading achievement at age 9/10. MZ correlations for all variables were below 1, indicating nonshared environmental influences on all variables.

Table 3 also reports heritability and shared and nonshared environment estimates from univariate twin model fitting. Reading motivation at age 9/10 and age 12 was moderately heritable, with genetic factors explaining 38% and 51% of the variance, respectively. The remaining variance in reading motivation at both waves was attributable to nonshared environmental influences. Reading achievement at ages 9/10 and 12 was also moderately heritable, with genetic factors explaining 39% and 34% of the phenotypic variance, respectively. Shared environmental influences were modest for reading achievement at age 9/10 (28%), but did not contribute to individual differences in reading achievement at age 12. Nonshared environmental influences, which also include measurement error, were modest for reading achievement at age 9/10 (33%) and large for reading achievement at age 12 (66%).

Table 4 reports cross-twin cross-trait correlations for all pairwise associations. Cross-twin cross-trait correlations were generally moderate for MZ twins and weak for DZ twins, indicating genetic influence on the covariance between each pair of variables. Some of the twin correlations indicated an ADE model—decomposing the variance into additive genetic (A), nonadditive genetic (D), and nonshared environmental effects (E)—as DZ correlations were less than half the MZ correlations. However, fitting an ADE did not improve model fit indices. We therefore reported results of ACE models, as these are in line with analyses presented by previous research.

Phenotypic Cross-Lagged Model

The phenotypic cross-lagged model allows us to explore three main concepts: correlation between variables measured at the same collection wave, stability of the variables, and cross-lagged association between different variables. Results from the phenotypic cross-lagged model are reported in Figure 2a and Table 5 . The phenotypic model showed a positive modest correlation between reading motivation at age 9/10 and reading achievement at age 9/10 ( r = .24). Reading motivation was moderately stable over time (.37), and the same was observed for reading achievement over time (.38). We observed reciprocal longitudinal links between reading motivation and reading achievement. The cross-lagged link from reading motivation at age 9/10 to reading achievement at age 12 was modest (.24). The opposite cross-lagged link from reading achievement at age 9/10 to reading motivation at age 12 was very similar (.26). Constraining the two cross-lagged paths to be equal did not result in worse model fit (χ 2 = 1.76, Δ df = 1, p = .18), suggesting that the two cross-lagged paths are of similar magnitude. Finally, we observed a moderate residual positive correlation between reading motivation and reading achievement at age 12 ( r = .44). Overall the model suggests that reading motivation at age 9/10 contributes to the variance in reading achievement at age 12 beyond the stability of achievement. Similarly and with similar strength, reading achievement at 9/10 contributed to the variance in reading motivation at age 12 beyond its stability.

ACE Cross-Lagged Model

We tested our second hypothesis regarding the etiology of the observed longitudinal associations between reading motivation and reading achievement using the ACE cross-lagged model. The same analyses were run separately for enjoyment and self-perceived ability, and results are presented in Table A2 and A3. Results from the ACE cross-lagged model are shown in Figure 2b , 2c , 2d , and Table 5 . The stability in reading motivation over time was explained by both genetic (around 44%) and nonshared environmental factors (approximately 55%). The stability in reading achievement was attributable to genetic (57%) and shared environmental influences (36%), and only a small portion of variance was explained by nonshared environmental factors (7%). The contemporaneous correlation between reading achievement and reading motivation was explained by both genetic (78%) and nonshared environmental (22%) influences. Importantly, genetic factors explained a substantial proportion of the cross-lagged link from early reading motivation to later reading achievement (58%). The remaining variance in this cross-lagged link was attributable to nonshared environment influences. The cross-lagged link from reading achievement at age 9/10 to reading motivation at age 12 was almost entirely explained by genetic factors (94%), with shared and nonshared environment explaining a negligible part of the covariance (2% and 4%, respectively). Finally, genetic factors, shared environmental factors, and nonshared environmental factors respectively accounted for 37%, 2%, and 61% of the residual contemporaneous correlation between reading motivation and reading achievement at age 12.

Cholesky Decomposition Model

We reanalyzed the data using the traditional Cholesky decomposition approach. Standardized path estimates of Cholesky cross-lag Model A and B are shown in Figure 3b and Figure 4b . Contemporaneous correlations, stability, and cross-lagged prediction derived from the standardized path estimates are shown in Table 6 . Overall, the results obtained fitting the Cholesky decomposition models were consistent with those obtained with the ACE cross-lagged model. Pertinent to our main research questions, reading achievement and reading motivation reciprocally predicted each other longitudinally after accounting for their stabilities and contemporaneous correlations. Similarly to what we observed using the ACE cross-lagged model, the link from reading motivation at age 9/10 to reading achievement at age 12 was explained by both genetic (35%) and nonshared environmental (65%) factors, and the link from reading achievement at age 9/10 to reading motivation at age 12 was almost entirely explained by genetic influences (88%), with the remaining variance explained by nonshared environmental factors (12%).