The ReadFree tool for the identification of poor readers: a validation study based on a machine learning approach in monolingual and minority-language children

In this study, we validated the “ReadFree tool”: a computerized battery of 12 hierarchically organized tasks in the visual and auditory modalities, which do not imply reading. The tool has been developed to identify poor readers irrespective of their specific language background, thus, to be also suitable for Minority-Language Children (MLC).Each task's discriminant power was tested on 142 Italian-monolingual participants (8-13 years-old) that either presented a reading deficit (i.e., monolingual poor readers (mPR); N = 37) or not (i.e., monolingual good readers (mGR); N = 105). The performances at the discriminant tasks were analysed by means of a multivariate machine learning approach based on a classification and regression tree (CART) model to classify mPR versus mGR.To test the diagnostic accuracy of the ReadFree tool in MLC, we first compared reading and ReadFree tool performances in MLC (N = 68) with those in monolingual readers (mGR + mPR; N = 142). The two groups did not show any significant differences, suggesting that (i) the two samples had the same distribution of good and poor readers and (ii) the ReadFree tool can be used to test MLC without introducing any systematic bias associated with their language use experience and exposure. Secondly, the MLC’s CART classification of good and poor readers was compared to the one obtained by adopting clinical reading tests standardized on Italian monolingual children. Interestingly, the percentage of MLC evaluated as poor readers through the standardized reading tests was higher than the one produced by the ReadFree tool. This evidence supports the idea that reading tasks standardized also on MLC population are needed.