The Effect of Modeling Missing Data With IRTree Approach on Parameter Estimates Under Different Simulation Conditions

Area of Science:

• Psychometrics
• Statistical modeling
• Educational measurement

Background:

• Missing data is a common challenge in psychometric and educational research.
• Existing methods like Expectation-Maximization (EM) and Multiple Imputation (MI) have limitations in handling missing data.
• The Item Response Tree (IRTree) approach presents a novel framework for modeling complex data structures, including missing responses.

Purpose of the Study:

• To evaluate the performance of the IRTree approach in modeling missing data.
• To compare IRTree's accuracy against traditional methods like EM and MI.
• To investigate the impact of different missing data mechanisms, test characteristics, and sample properties on method performance.

Main Methods:

• Utilized both simulation studies and empirical data for comprehensive evaluation.
• Assessed performance across varying missing data mechanisms (MCAR, MAR, MNAR).
• Calculated bias and Root Mean Square Error (RMSE) for ability estimates using the Expected a Posteriori (EAP) method.

Main Results:

• IRTree demonstrated superior accuracy in ability estimation, evidenced by lower RMSE compared to EM and MI.
• IRTree's performance was robust under Missing Completely At Random (MCAR) and Missing At Random (MAR) conditions.
• Optimal performance for IRTree was observed with longer tests, lower proportions of missing data, and moderate levels of omissions.

Conclusions:

• The IRTree approach is a promising alternative for handling missing data in item response theory models.
• IRTree offers more precise ability estimates than conventional EM and MI methods.
• IRTree has significant potential for application in low-stakes testing environments and for uncovering insights into missing data patterns.