Screening for aberrant school performances in high-stakes assessments using in influential analysis

A method is proposed to screen for aberrant school performances in large-scale, high-stakes assessments using influential analysis under a Bayesian approach. Proportions of low and high achievers within a school were modelled via the beta inflated mean regression model (Bayes and Valdivieso, 2016) using school performances in previous years as predictors. The general measure of ϕ-divergence proposed by Peng and Dey (1995) was used to determine aberrancy. A simulation study revealed that the method could recover previously distorted school performances as aberrant. The proposed technique was applied to a Peruvian national reading assessment in grade 4th of primary education for which the government provided a school performance incentive bonus.