Because the subtests designed to probe the central executive and phonological loop depend heavily ABT-199 concentration on language, it is possible that the observed working memory deficits in the participants with SLI might be due to their language problems rather than to working memory deficits per se. Therefore we performed additional analyses in which we covaried out a measure of language abilities. We computed a single composite variable of language by submitting the four measures of language (expressive and receptive lexical and grammatical abilities; see Table 2) to a principal components analysis, and extracted

a single factor. This approach aims to create a composite variable that maximizes the shared variance of all four language measures, and minimizes the variability that is unique to a single measure or is shared only between two or three of them. The four measures accounted for PLX3397 cost 67.7% of the variance in the language factor. The factor loadings were as follows: Expressive Vocabulary = .853, Receptive Vocabulary = .832, Expressive Language = .769 and Receptive Grammar = .834. The MANCOVAs with the language factor included as covariate yielded significant multivariate group effects both for the central executive (p < .001) and phonological loop subtests (p < .001), although with a reduction of effect sizes in both cases ( Table 3, Covariates: Language

Factor). The post-hoc univariate tests controlling for language abilities revealed significant differences on all the central executive and eltoprazine phonological loop subtests except the Word List Matching subtest, mostly with medium (partial η2 ≥ .059) or large effect sizes ( Table 4, under “Covariate: Language Factor”). The next set of analyses tested SLI-TD group differences on the CMS, to examine declarative memory for verbal and visual information. Results from between-subjects MANOVAs revealed a significant multivariate group effect for the subtests probing verbal information (p < .001), with a large effect size, but not for the subtests of visual information (p = .350),

which yielded a small effect size ( Table 3, Covariates: None). The post-hoc univariate tests ( Table 5, under “No covariates”) yielded significant group differences, with medium to large effect sizes, on all measures designed to assess verbal aspects of declarative memory. In contrast, small effect sizes were found on all visual subtests, only one of which showed a significant group difference. Many of the subtests from the CMS require children to temporarily store information, and thus the observed group differences could in part be explained by working memory deficits rather than problems with declarative memory itself. Group differences on the CMS were therefore examined while controlling for working memory.