The One-Parameter Logistic Model Can Be True With Zero Probability for a Unidimensional Measuring Instrument: How One...

Area of Science:

• Psychometrics
• Educational Measurement
• Behavioral Research

Background:

• The one-parameter logistic (1PL) model and Rasch model are foundational in psychometrics for analyzing unidimensional scales.
• Assessing the probability of these models holding true for dichotomous items is crucial for scale validity.

Purpose of the Study:

• To investigate the probability of the 1PL or Rasch model being true for unidimensional scales with dichotomous items.
• To explore the consequences of removing items that do not fit these models.

Main Methods:

• Analysis of the probability of 1PL/Rasch model adherence for unidimensional scales.
• Simulation studies using large datasets to examine item elimination effects.

Main Results:

• The probability of the 1PL or Rasch model being correct can be zero even for unidimensional scales.
• Removing items not fitting these models can result in seriously misleading ability estimates.
• Item elimination leads to increased standard errors and prediction errors for the latent trait.

Conclusions:

• Researchers must be cautious when removing items based solely on 1PL/Rasch model fit.
• Misleading ability estimates can have significant implications for educational and behavioral research findings.
• Alternative approaches to scale refinement may be necessary to ensure accurate measurement.