Educational Poverty (EP) is a concept of increasing importance whose measure, particularly at local level, is of paramount relevance to enforce and monitor policies. Measuring EP is hindered by its latent and possibly multidimensional nature, and researchers have not yet agreed on a set of items that would serve this aim. In this paper, we focus on a set of 33 binary indicators measured with the Activities of Daily Living survey on a sample of 4,382 individuals aged 15–29, and propose to use multidimensional Item Response Theory (IRT) models to extract the latent dimensions of EP. To obtain estimates of these latent dimensions at local (sub-regional) level, we embed Small Area Estimation in the multidimensional IRT model by allowing the values of the latent factors to change with covariates and a area-specific random effects.