Research

The ALPHA Lab will conduct research at the intersection of psychometrics, machine learning, and statistical mediation. These areas share a common goal: to strengthen the methodological foundations of psychological and health research. Each represents a major pillar of our work, and graduate students in the lab are encouraged to develop projects in any of these domains—or at their intersections.

Integrating Psychometrics and Machine Learning

A green square microchip with pins connects to a large orange network node, and branches to three smaller orange nodes on a light tan background.
This illustration was created by Oscar Gonzalez using ChatGPT

Our lab explores how machine learning algorithms can expand and refine the psychometric toolbox for assessment. Machine learning, a data-driven branch of artificial intelligence, identifies patterns in data and improves through experience. Many of the questions that psychometricians have long addressed through inferential statistics, such as scale development, item evaluation, and prediction, can now be revisited through the lens of modern algorithms.

We aim not only to compare traditional and machine-learning frameworks but also to integrate them, developing new tools that combine the rigor of psychometrics with the flexibility of machine learning.

Sample articles:

Measurement in Statistical Mediation

Mediation model diagram. Orange box X points to teal box M (path a) and orange box Y (path c'). Box M points to box Y (path b).
This illustration was created by Oscar Gonzalez using ChatGPT

Statistical mediation analysis helps identify mechanisms that explain how one variable influences another. Yet, despite its central role in psychological and health research, measurement issues in mediation models remain understudied.

Our work focuses on improving mediation research through the mediation-by-design paradigm—a framework in which conceptual theories of change guide intervention design and evaluation. We emphasize that identifying mechanisms in a causal process requires precise and interpretable measurement of the constructs involved. Through methodological development and simulation, we study how measurement quality shapes mediation conclusions and our theoretical understanding of change processes.

Sample articles:

Validity and Screening Accuracy

Brown clipboard with a bronze clip holding a beige paper displaying three teal checkmarks and corresponding brown list lines.
This illustration was created by Oscar Gonzalez using ChatGPT

A core part of our research involves evaluating the validity of psychological assessments, especially those used for screening and classification. We examine the accuracy and consistency of screening measures and how these metrics are influenced by differential item functioning.
Our interests also include the jingle and jangle fallacies—situations where different measures capture the same construct (jangle) or where similarly named measures capture different constructs (jingle). By clarifying construct overlap and refining validity evidence, we seek to improve the interpretability and fairness of psychological assessment.

Sample articles: