Pressure to design more informative assessments that advance understandings about what knowledge respondents’ actually possess and to estimate proficiency profiles for multiple cognitive dimensions is increasing in educational assessment (Erickan, 2006; Leighton & Gierl, 2007). Diagnostic cognitive models (DCMs) provide one approach to the measurement of multiple dimensions of knowledge. These models rely on categorical latent attributes to represent respondents’ mastery or non-mastery of underlying skills. Given that large amounts of resources are invested in the development of large-scale assessments, such as the Progress in International Reading Literacy Study (PIRLS), some researchers’ endorse the application of DCMs to large-scale assessments (Chen & de la Torre, 2014).

The purpose of this study is to apply a DCM model to PIRLS data to examine the subcomponents of reading literacy measured by PIRLS and to compare students’ proficiency profiles across Canada and the U.S. An item-attribute matrix, referred to as a Q-matrix, will be constructed based on the specified purposes and processes defined by PIRLS. Items are written to target specific reading subcomponents, but the extent to which they adequately target meaningful subcomponents is unknown. A general DCM model will be applied to freely estimate hypothesized structural and measurement correlations between attributes. This process is similar to a CFA approach in that the saturated model is used as a tool to assess statistical evidence to evaluate theory-based hypotheses. Data-model fit will be compared to examine item-attribute alignment across both groups. Attribute classification rates and marginal probabilities for attributes will be compared to examine similarities and differences across groups. Evidence from this study will provide insight about the subcomponents of reading literacy as defined by PIRLS and patterns of proficiency across two countries.