Explanation of performance by context. Some proposals from graphic and statistical analysis

Abstract

The assessment of education systems suffers from certain weaknesses, among which is the limited statistical treatment of the information obtained. An improvement could be to pay more attention to explaining the product from the context. This paper aims to illustrate, in a simple way, possible statistical models for linking antecedent variables (context variables) with learning outcomes (performance) using some of the graphical tools that provide the dependency models based on regression. More specifically, an ordinary linear model and a generalized linear model of Poisson regression are carried out. For this objective, the conditions of application for these models are reviewed examining some elements of diagnostic in regression: basic diagnostic plots; unusual data (outliers, hat values, influence measures) or residual plots, for example. R is the software program used in all statistics analyses, because we think it has a good graphic interface and very simple outputs. Then we offer some suggestions for making decisions based on the information provided by graphical representations from model residuals.