Generating Probabilistic Knowledge Structures from Programme for International Student Assessment (PISA) Data
In this repo, I experiment with methods from Knowledge Space Theory and investigate how these methods predict when applied to a large-scale real-world educational assessment data, such as PISA.
Probabilistic knowledge structures provide a framework for modelling a person’s knowledge or skills based on their responses to a finite set of items. Instead of reducing performance to a single overall score, they make it possible to describe more detailed patterns of ability.
The basic idea is simple: a learning domain is represented by a set of tasks or questions, each of which a student can either solve or not solve. By looking at these response patterns, we can identify what students are likely to know and where they may struggle.
Knowledge structures are useful because they can represent relationships between items. For example, solving a difficult task may require understanding a simpler concept first. These relationships help describe different knowledge profiles or stages of learning.
Probabilistic knowledge structures add a layer of realism. In real data, students do not always respond consistently: they may guess correctly (lucky guess) or make a careless mistake (careless error). Probabilistic models account for this uncertainty, making them well suited for real response data such as PISA.
Further information and examples related to probabilistic knowledge structures and interactive Shiny applications can be found here: