Background A long-standing premise in test design is that contextualizing test items makes them concrete, less demanding, and more conducive to determining whether students can apply or transfer their knowledge. Purpose We assert that despite decades of study and experience, much remains to be learned about how to construct effective and fair test items with contexts. Too little is known about how item contexts can be appropriately constructed and used, and even less about the relationship between context characteristics and student performance. The exploratory study presented in this paper seeks to contribute to knowledge about test design and construction by focusing on this gap. Research Design We address two key questions: (a) What are the characteristics of contexts used in the PISA science items? and (b) What are the relationships between different context characteristics and student performance? We propose a profiling approach to capture information about six context dimensions: type of context, context role, complexity, resources, level of abstraction, and connectivity. To test the approach empirically we sampled a total of 52 science items from PISA 2006 and 2009. We describe the context characteristics of the items at two levels (named layers): general (testlet context) and specific (item context). Conclusion We provide empirical evidence about the relationships of these characteristics with student performance as measured by the international percentage of correct responses. We found that the dimension of context resources (e.g., pictures, drawings, photographs) for general contexts and level of abstractness for specific contexts are associated with student performance.
