The impact of model misfit on partial credit model parameter estimates

Area of Science:

• Psychometrics
• Educational Measurement
• Statistical Modeling

Background:

• The partial credit model (PCM) is widely used for analyzing polytomous item responses.
• PCM's lack of a discrimination parameter can lead to model-data misfit.
• Understanding the impact of this misfit on parameter estimation is crucial.

Purpose of the Study:

• To investigate the consequences of model misfit in the PCM.
• To assess the bias and efficiency of person and item parameter estimates under varying degrees of misfit.
• To evaluate the robustness of the PCM to data deviations.

Main Methods:

• A simulation study was designed to generate data based on the generalized partial credit model (GPCM).
• Different levels of unsystematic and systematic misfit were introduced into the generated data.
• Bias and efficiency of parameter estimates were evaluated using the PCM on simulated datasets.

Main Results:

• Small amounts of unsystematic misfit showed minimal impact on parameter estimate bias and efficiency.
• Large levels of unsystematic misfit led to considerable bias and loss of efficiency.
• Moderate levels of systematic misfit also resulted in substantial bias and reduced efficiency.

Conclusions:

• The PCM is sensitive to model-data misfit, particularly systematic deviations.
• Researchers should carefully assess model fit when using PCM for parameter estimation.
• Significant misfit necessitates caution in interpreting PCM-derived person and item parameters.