DTU CALIDIE Colloquium: Same data, different models

The same large-scale data can be used to answer many different questions. Last year, we discussed large-scale datasets and their potential for secondary analyses. We focussed on the importance of exploring and to deciding which questions could and should be analysed and how. This year we would like to focus even more on the “how,” the method and how important its choice is depending on the purpose of the analyses. Small changes to a model can have large consequences for the results and subsequent conclusions of a study. To guarantee objectivity and replicability, transparency is paramount. We will present and discuss the results we have found using these large-scale datasets by diving into the models we used and the choices we made. Sophie will discuss how using different cut-offs and reference groups to establish math and reading difficulties affect the prevalence of difficulties and the actual cut-off scores below which students are classified as having difficulties. Large-scale datasets are not only informative at the student level but can also be used at teacher or school level. Jessica will investigate different models for the estimation of school effectiveness (value-added models) by using longitudinal data. Additionally, she will discuss how these measures of school effectiveness can enable systematic analyses of pedagogical practices that are effective in a diverse multilingual school setting.