Development of Lexical Mediation in the Relation Between Reading Comprehension and Word Reading Skills in Greek

Abstract

This study focuses on the shared variance between reading comprehension and word-level reading skills in a population of 534 Greek children in Grades 2 through 4. The correlations between measures of word and pseudoword accuracy and fluency, on the one hand, and vocabulary and comprehension skills, on the other, were sizeable and stable or increasing with grade. However, the unique contribution of word reading to comprehension became negligible after vocabulary measures were entered in hierarchical regression analyses, particularly for higher grades, suggesting that any effects of decoding on comprehension may be mediated by the lexicon, consistent with lexical quality hypothesis. Structural modeling with latent variables revealed an invariant path across grades in which vocabulary was defined by its covariation with reading accuracy and fluency and affected comprehension directly. It is argued that skilled word reading influences comprehension by strengthening lexical representations, at least when phonological decoding can be relatively effortless.

Notes

^a N = 188.

^b N = 176.

^c N = 170.

^a N = 188.

^b N = 176.

^c N = 170.

*p< .01.

**p< .001.

***p< .0001.

**p< .001.

¹In languages with shallow orthography, where ceiling accuracy levels are not uncommon, fluency measures are considered better indicators of word-level skills and more useful for the detection of word-level reading disability (see, e.g., for Finnish: Holopainen, Ahoren, & Lyytinen, 2001; German: Landerl, 2001; Italian: Tressoldi, Stella, & Faggella, 2001; Spanish: Jiménez González & Hernández Valle, 2000). For Greek in particular, see Porpodas (1999) for beginning readers and Protopapas and Skaloumbakas (2007) for seventh-grade children.

^a N = 188.

^b N = 176.

^c N = 170.

²Potential misfits were explored to determine whether the model could be further improved before including structural parameters. Overall model fit in measurement models may appear poor because all factors are tested as orthogonal entities and the absence of modeled covariations reduces overall model fit (Dunn, Everitt, & Pickles, 2002). The multivariate Lagrange test suggested that model fit would be significantly improved only by including covariation paths between latent variables. Given that all measurement paths were significant at p< .05, in both robust and nonrobust estimates (Satorra & Bentler, 1990) and that residual estimates were quite small (i.e., RMSEA = .166), we conclude that the measurement model provided a reliable estimate of the constructs of interest (reliabilities were also high: Cronbach's α = .841; Spearman's ρ = .856).

^a N = 188.

^b N = 176.

^c N = 170.

*Significant χ² statistic at p< .05

³An alternative model was tested in which accuracy and fluency were linked to comprehension with direct paths, whereas vocabulary was linked to comprehension directly as well as indirectly via accuracy and fluency; that is, the order of vocabulary and word-level skills was reversed with respect to comprehension. This alternative model was found to be equivalent to Model 2. However, it was not preferred because (a) it lacked a parsimonious interpretation, and (b) comprehension was more strongly correlated with vocabulary than with accuracy and fluency; thus supporting a closer connection as long as structural models remain equivalent.

^a N = 21.

^b N = 18.

^c N = 20.

^d N = 19.

^e N = 23.

^f N = 17.

**p< .001.

***p< .0001. (by t test with respect to zero).

⁴The equivalence between a model in which lexical skills mediate the effect of decoding on comprehension, and an alternative model in which accuracy and fluency mediate the effect of vocabulary, further strengthens the interpretation in favor of a reciprocal interdependence among lexical and decoding skills.

⁵Interestingly, if decoding ceases to have a direct effect on comprehension and yet poor comprehenders fall behind in decoding, among other skills, due to limited reading practice and reciprocal interactions, one may suggest that the opposite of the Simple View is the case: Instead of poor decoding preventing adequate comprehension, it is poor comprehension that holds back development of fluent decoding.

⁶We thank an anonymous reviewer for pointing this out.