Related Experiment Videos
How many IRT parameters does it take to model psychopathology items?
Steven P Reise1, Niels G Waller
1Department of Psychology, University of California, Los Angeles 90095, USA. reise@psych.ucla.edu
Psychological Methods
|August 20, 2003
Summary
The three-parameter logistic model (3PLM) slightly improved model fit over the two-parameter logistic model (2PLM) for personality scales. However, scoring direction significantly impacted item parameters, suggesting content ambiguity in personality assessments.
Area of Science:
- Psychometrics
- Psychological Measurement
- Personality Assessment
Background:
- Item response theory (IRT) models are crucial for analyzing psychological test data.
- The Minnesota Multiphasic Personality Inventory--Adolescent (MMPI-A) is a widely used personality assessment tool.
- Evaluating the fit of different IRT models is essential for accurate interpretation.
Purpose of the Study:
- To compare the fit of the 2-parameter logistic model (2PLM) and the 3-parameter logistic model (3PLM).
- To investigate the impact of scoring direction on item parameters within personality scales.
- To identify potential sources of model misspecification in IRT analyses of personality data.
Main Methods:
- Analysis of 15 unidimensional factor scales from the MMPI-A item pool.
- Application of log-likelihood chi-square deviance tests to assess model fit.
- Examination of residual statistics and lower asymptote parameters (c).
Main Results:
- The 3PLM demonstrated a marginally improved fit compared to the 2PLM, though the difference was negligible based on residual statistics.
- A substantial proportion of items (10-30%) exhibited significant lower asymptote parameters (c ≥ 0.10).
- These large lower asymptote parameters were observed irrespective of whether scales were scored in the pathology or nonpathology direction.
Conclusions:
- While the 3PLM offers a slight improvement, the practical difference in fit over the 2PLM may be minimal for these MMPI-A scales.
- Item-content ambiguity, potentially due to item-level multidimensionality, is a likely cause for substantial lower asymptote parameters.
- The direction of scale scoring critically influences item response theory analyses, highlighting the importance of careful model specification and interpretation.