Xingyao (Doria) Xiao, Ph.D.

Postdoctoral Scholar at Stanford GSE researching statistical methodology for longitudinal and adaptive data.

My research operates at the intersection of statistical methodology and educational technology, specifically developing robust psychometric frameworks for high-stakes digital assessments. I specialize in Bayesian longitudinal modeling and the integration of machine learning into measurement.

Current Research Themes

1. AI-Integrated Measurement & Hybrid Scoring

I investigate how machine learning (ML) scoring can scale constructed-response assessments when integrated with human judgment.

  • My recent work shows that systematic rater bias is a greater threat to reliability than random machine noise.
  • I have identified a “bias compensation” mechanism: pairing ML models with human raters of opposing severity profiles can cancel out directional errors and restore accuracy.
  • This research supports using the Many-Facet Rasch Model (MFRM) and anchoring strategies to stabilize scales in sparse scoring designs.

2. Bayesian Longitudinal & Growth Mixture Modeling (GMM)

A core focus is the stability and evaluation of latent variable models.

  • I develop Bayesian diagnostics to identify local identifiability issues and convergence failures in growth mixture models.
  • My work utilizes marginal likelihood to compare model selection strategies for longitudinal behavioral change.

3. Global Development (LEVANTE Project)

At Stanford, I work on the LEVANTE project to create internationalized measures of learning for children aged 5–12.

  • I focus on measurement invariance and statistical methods for modeling developmental change across cultural contexts.
  • We aim to validate core cognitive tasks that remain robust across languages and research sites.

4. Adaptive Measurement & Digital Innovation

  • CAT Fairness: Researching calibration error mitigation and robust stopping rules in computer-adaptive testing.

Methods & Technical Toolbelt

I lean heavily on open-science and “GitHub-first” workflows to solve measurement challenges.

  • Models: Multidimensional IRT (mIRT), Many-Facet Rasch Models (MFRM), Bayesian GMM, and Differential Item Functioning (DIF).
  • AI/NLP: LLM evaluation (GPT-4), NLP-based scoring frameworks, and automated prompt engineering.
  • Statistical Software: Expert proficiency in R (Stan, ggplot2, Shiny), Python, SQL, and Mplus.