The InterModel Vigorish as a Lens for Understanding (and Quantifying) the Value of Item Response Models for...

Area of Science:

• Psychometrics
• Statistical Modeling

Background:

• Statistical models, including those in item response theory, require indices to assess their appropriateness for specific data.
• Existing indices may have limitations in comparing non-nested models or generalizing across contexts.

Purpose of the Study:

• To introduce and evaluate the InterModel Vigorish (IMV) as a novel index for quantifying the accuracy of dichotomous item response models.
• To demonstrate the IMV's utility in comparing model predictive performance across diverse simulated and empirical datasets.

Main Methods:

• The InterModel Vigorish (IMV) was developed to quantify accuracy based on prediction improvement between two sets of predictions.
• Comparative analyses were conducted using simulated data to contrast IMV with common indices like AIC and RMSEA.
• Empirical validation involved applying the IMV to 89 dichotomous item response datasets.

Main Results:

• The IMV demonstrated desirable features, including portability and generalizability, making it suitable for comparing non-nested models.
• Qualitative differences in behavior were observed between the IMV and other common indices.
• Empirical applications confirmed the IMV's practical utility and substantiated claims regarding its performance.

Conclusions:

• The InterModel Vigorish (IMV) is a valuable new indicator in psychometrics for assessing statistical model accuracy.
• Its ability to facilitate easy comparison of predictions across various contexts enhances its utility for researchers.
• The IMV provides a robust measure for evaluating and comparing item response models.