DIF Detection With Zero-Inflation Under the Factor Mixture Modeling Framework.

Area of Science:

• Psychometrics
• Statistical modeling

Background:

• Zero-inflated data, common in educational and psychological assessments, pose challenges for detecting differential item functioning (DIF).
• Excessive zeros can mask true DIF effects, leading to underdetection and potentially biased results.

Purpose of the Study:

• To propose and evaluate a novel DIF detection procedure for zero-inflated data with unobserved heterogeneity.
• To compare the performance of the proposed Factor Mixture Modeling (FMM) with MIMIC approach against the traditional Likelihood Ratio (LR) DIF test.

Main Methods:

• Utilized Factor Mixture Modeling (FMM) with MIMIC (multiple-indicator multiple-cause) to estimate latent classes and address unobserved heterogeneity.
• Conducted a Monte Carlo simulation study to compare the proposed FMM procedure with the LR DIF test.

Main Results:

• The FMM approach demonstrated superior detection power compared to the LR DIF test for zero-inflated data.
• Results highlighted the critical importance of accounting for latent heterogeneity in the analysis of such data.
• Empirical data analysis confirmed FMM's ability to identify additional DIF items missed by the LR test.

Conclusions:

• Factor Mixture Modeling (FMM) offers a more powerful and accurate method for DIF detection in zero-inflated datasets with unobserved subgroups.
• Accounting for latent heterogeneity is essential for robust DIF detection and accurate psychometric analysis of zero-inflated data.