Analysis of the 2021 PIRLS Data for Türkiye Using Cognitive Diagnostic Modeling

This study examines the Turkish PIRLS 2021 data through the lens of cognitive diagnostic modeling (CDM). Türkiye last participated in the PIRLS exam in 2001, prior to 2021. Türkiye’s participation in this international assessment after a twenty-year hiatus is of great significance. The analysis was restricted to responses to one publicly released PIRLS 2021 literary passage (“The Empty Pot”) administered to 667 students in the Turkish sample. Applying CDM to these assessment data makes it possible to identify students’ specific reading-related strengths and limitations and to generate more fine-grained feedback. Additionally, CDM can provide insights into the approach needed to better develop students’ reading skills during the educational process. In this study, the text “The Empty Pot” was selected for analysis, and data from 667 students in Türkiye who completed this text were included in the analysis. First, with input from subject matter experts, the PIRLS attributes “Interpreting and integrating ideas/information” was divided into “Interpretation” and “Using personal knowledge and experiences.” The final Q-matrix includes the attributes “Making inferences,” “Interpretation,” “Using personal knowledge and experiences,” and “Examination and Evaluation.” Additionally, a Q-matrix validation step was applied to correct possible classification errors. The analyses involved estimating and comparing six CDMs: G-DINA, DINA, DINO, A-CDM, LLM, and R-RUM. The fit evidence indicated that G-DINA provided the most adequate representation of the observed response data. The results indicated that 39 students (5.85%) were classified as experts (1111) in all attributes. Another notable result is that 69 students who took the exam were not proficient in any attributes (0000), indicating that under the selected model, nearly 10% of the students was assigned to the non-mastery profile. Results also showed that making inferences showed the highest estimated mastery probability in this analysis among the four, using personal knowledge and experience was identified as the most difficult. Because the study used a retrofitted CDM design based on a single released passage and a validated Q-matrix with some remaining interpretive limitations, the findings should be interpreted as passage-specific diagnostic evidence rather than nationally generalizable estimates of reading proficiency.