Predicting Item Difficulty and Response Time Through a CART Approach: Evidence from the INVALSI Test on Digital Competences

In educational measurement, estimating item difficulty traditionally relies on response data and psychometric models, which may be unavailable in early stages of test development. Recent studies have shown that textual character istics of assessment items provide valuable information for predicting difficulty, yet most approaches focus on difficulty as a single outcome. This study uses Ital ian data on student digital competence assessment to investigate how textual and structural features of test items relate to both item difficulty and response times. Using regression tree–based models, univariate analyses are conducted to explore nonlinear relationships and interpretable decision rules linking item content to item-level metrics. The results highlight the central role of textual features in explaining variability across both dimensions and motivate the use of multivariate regression trees for a joint analysis of item functioning.