A short version of the MacArthur–Bates Communicative Development Inventories with high validity

Abstract

The MacArthur-Bates Communicative Development Inventories (CDIs) are among the most widely used evaluation tools for early language development. CDIs are filled in by the parents or caregivers of young children by indicating which of a prespecified list of words and/or sentences their child understands and/or produces. Despite the success of these instruments, their administration is time-consuming and can be of limited use in clinical settings, multilingual environments, or when parents possess low literacy skills. We present a new method through which an estimation of the full-CDI score can be obtained, by combining parental responses on a limited set of words sampled randomly from the full CDI with vocabulary information extracted from the WordBank database, sampled from age-, gender-, and language-matched participants. Real-data simulations using versions of the CDI-WS for American English, German, and Norwegian as examples revealed the high validity and reliability of the instrument, even for tests having just 25 words, effectively cutting administration time to a couple of minutes. Empirical validations with new German-speaking participants confirmed the robustness of the test.

A Bayesian-Inspired Item Response Theory–Based Framework to Produce Very Short Versions of MacArthur–Bates Communicative Development Inventories

“…Results for the original model and the baseline measure were averaged over 10 simulations, whereas those for the IRT version were based on single simulations, since the item selection process establishes individual parameters for each child, consequently constraining the selection of words for that child. In line with previous work using real-data simulations (i.e., Makransky et al, 2016;Mayor & Mani, 2019), correlations between the estimates (based on 5, 10, 25, 50, 100, 200, 400, and all items on each CDI) and the full CDI scores, average standard errors, and reliability (1-SE 2 ) were reported across different age groups and genders.…”

“…Recently, Mayor and Mani (2019) presented a languagegeneral approach that takes advantage of the richness of Wordbank (Frank et al, 2017), an open repository for crosslinguistic CDI data from over 75,000 children across 29 languages. Their approach combines a subset of items drawn randomly from the full forms with (prior) CDI data sampled from language-, gender-, and age-matching children on Wordbank.…”

“…The next section details the two main components of our approach, that is, the IRT-based selection of test items coupled with a full CDI score estimation based on Mayor and Mani's (2019) model. We then present the result and discuss the implications of our findings for researchers and practitioners intending to use short forms for quick and cost-effective assessments of young children's vocabulary.…”

“…The method of fitting each of the item-based histograms of full CDI scores to a normal distribution and the estimation of CDI scores closely resembled those described in Mayor and Mani (2019). First, language-, gender-, and age-matched children are retrieved from Wordbank (Frank et al, 2017) using Braginsky's (2018)…”

“…Using real-data simulations, we compared the performance of the present model (the IRT version) with the original model by Mayor and Mani (2019) in estimating full CDI scores. We adopt the following standard for test acceptability, as introduced by Makransky et al (2016): correlations above .95 with the full CDI, average SE below .20, and reliability coefficients (1-SE 2 ) above .96.…”