My specific lines of research include statistical methods to conduct efficient measurement bias evaluation, longitudinal modeling under measurement change and person-specific time metrics, and applications in the context of education and replication studies.
Efficient measurement bias evaluation
When measurement items elicit different responses from individuals who have different background characteristics, irrespective of their levels of underlying construct levels, such items display differential item functioning (DIF) or measurement bias. My research extends a statistical learning approach called Bayesian regularization to detect significant DIF effects through parameter selection rules. This approach can simultaneously evaluate all items for measurement bias across multiple background variables that are categorical or continuous, without prior assumptions regarding which items are free of DIF that are typically required by traditional methods. My next step is to further understand the penalty choice and scoring performance of the penalized model.
Chen, S. M., Bauer, D. J., Belzak, W. M., & Brandt, H. (2022). Advantages of spike and slab priors for detecting differential item functioning relative to other Bayesian regularizing priors and frequentist lasso. Structural Equation Modeling: A Multidisciplinary Journal, 29(1), 122–139. https://doi.org/10.1080/10705511.2021.1948335
Brandt, H., Chen, S. M., & Bauer, D. J. (2023). Bayesian penalty methods for evaluating measurement invariance in moderated nonlinear factor analysis. Psychological Methods. https://doi.org/10.1037/met0000552
2. Longitudinal modeling with changing measurement
To model construct growth with scale items that have changing measurement properties over time, my research develops a longitudinal moderated factor analysis model. This model allows measurement and growth parameters to vary as a function of person background covariates. This approach can accommodate time variables with many unique values without requiring as many latent dimensions; it can also represent measurement change (i.e., DIF) over time across multiple background variables. Bayesian regularization methods can be applied to automatically detect DIF effects across all grouping variables. We have evaluated this approach for its DIF detection and growth recovery performance in simulation. The next step is to examine its effectiveness in extrapolating person-specific growth trajectories.
Chen, S. M. & Bauer, D. J. (In press). Modeling construct change over time amidst potential changes in construct measurement: a longitudinal moderated factor analysis approach. Psychological Methods. https://dx.doi.org/10.1037/met0000685
Preprint and Supplemental Materials: https://osf.io/preprints/psyarxiv/uwt5a
3. Applications in educational testing and psychology replications
Finally, I am broadly interested in applying the regularized measurement modeling and evaluation methods to the empirical contexts of educational assessment and replication studies in psychology. My postdoc projects involve understanding the influence of measurement bias and factor variance differences on performance disparities among demographic groups. I am also working on evaluating the implications of measurement bias in replication study results.