What You Don't Know Can Hurt You: Missing Data and Partial Credit Model Estimates

Area of Science:

• Psychometrics
• Longitudinal Data Analysis
• Psychological Measurement

Background:

• Missing data is a common issue in longitudinal studies.
• The Partial Credit Model (PCM) is sensitive to data quality.
• Understanding the impact of missing data on PCM is crucial for accurate measurement.

Purpose of the Study:

• To investigate the effect of missing completely at random (MCAR) item responses on PCM parameter estimates.
• To quantify the impact of increasing percentages of missing data on measurement quality.

Main Methods:

• Utilized longitudinal data from 307 adults in the Notre Dame Study of Health and Well-Being.
• Introduced simulated missing data at 20%, 50%, and 70% levels to existing data.
• Applied the Partial Credit Model (PCM) to assess parameter estimates.

Main Results:

• Item location and person trait estimates diverged from original values as missing data increased.
• Standard errors for these estimates also increased with higher levels of missing data.
• Demonstrated a negative impact of MCAR data on PCM parameter accuracy.

Conclusions:

• Missing completely at random (MCAR) data significantly degrades the quality and precision of Partial Credit Model (PCM) estimates.
• Researchers should be cautious when interpreting PCM results from datasets with substantial missingness.
• The study highlights the importance of addressing missing data in psychometric analyses.