Current challenges for the modelling of educational data
Wednesday, September 12, 2018; 14:30 - 15:30, Room: FORUM
Traditional ways of acquiring data to answer research questions are being challenged in a number of ways. The increasing availability of very large population based administrative datasets provides new and exciting possibilities for more detailed and extensive analyses that can handle the real life complexity of the data and with sufficient power to detect interesting interactions with consequences for increasingly informative interpretations. A major challenge for data methodologists is thus to develop and make widely available statistical models that are able to reflect this complexity. At the same time, studies that are designed to collect more nuanced information than that found in administrative datasets, remain important but are increasingly subject to ‘non-response’ which can lead to potential bias and threats to validity. This too poses a serious challenge to data analysts. The talk will discuss these two trends, their interrelationship and some of the emerging methodologies needed to address the issues. Specifically, the talk will look at new ways for modelling data with missing values, ways for incorporating knowledge about unreliability within a model, issues in the linking together of data from different sources such as education and health, and implications for the ways in which research studies, especially longitudinal ones, are designed. Illustrations of data analyses will be given.