Improving the Measures of Phonological Ability in the Russian Language: IRT and CART Modeling Application

Phonological awareness and phonological working memory are essential for successful language acquisition and development of literacy. Although this essence is language-universal, its degree varies for different languages, depending, in part, on language transparency. The current study analyzes the adapted versions of the pseudoword repetition test (assessing phonological working memory) and Rosner's Auditory Segmentation test (assessing phonological awareness) in a typically developing Russian native sample of children (n = 502). As a preparatory step to item analysis, we investigated the effects of grade and gender on performance using a mixed effects model. The initial item analysis was carried out using model comparison within the Item Response Theory model framework and threshold/slope analysis. The majority of the items in both assessments did not differentiate between students with different levels of phonological ability. Further item selection using regression tree models led to the formation of predictive and non-predictive item subsets for each assessment. After comparing the item subsets on various linguistic metrics, the differences were found in number of syllables and subsyllabic complexity for the pseudoword repetition test and elision segment position for the auditory segmentation test. The findings inform test development strategies in the cases of extremely low difficulty/discrimination of the items and outline a blueprint of pseudoword repetition and auditory segmentation test's adaptation for potentially detecting higher levels of phonological ability in transparent languages such as Russian.