Quantitative Psychologist · Psychometrician

Hau-Hung Yang

Measuring the mind, rigorously — finite-sample theory for adaptive testing.

I build finite-sample guarantees for computerized adaptive testing, bridging Item Response Theory and Online Convex Optimization — with causal work on Taiwanese administrative data.

Recent

I received my Ph.D. in Psychology from National Taiwan University in June 2026 and am a postdoctoral scholar at the Institute of Statistical Science, Academia Sinica. My work sits where mathematical psychology, psychometrics, and econometrics meet — turning careful theory into methods that hold up in finite samples and on real data.

Research programs

Two lines of work, one toolkit

Latent-variable models, sequential inference, and large-scale estimation — applied to how we measure the mind, and to what people do in administrative records.

Psychometrics and mathematical psychology

  • Finite-sample theory for adaptive measurement: computerized adaptive testing analysed as online convex optimization, giving no-regret item selection, anytime-valid confidence sequences, and confidence-set UCB for online calibration (dissertation, 2026; IMPS 2026).
  • Statistical properties of psychometric models: identifiability of polychoric models under elliptical latent distributions (Psychometrika, 2025), stochastic approximation for adaptive testing, and the generalized Robbins–Monro process (Journal of Mathematical Psychology, 2024).
  • Open-source tools: the anytimeCAT R package for anytime-valid confidence sequences (in development) and an interactive Central Limit Theorem course.

Applied economics with administrative data

  • Labor-market sorting: linked employer–employee income-tax records and an AKM two-way fixed-effects decomposition of wage inequality (Taiwan Economic Review, 2026).
  • College-admission preferences: national application records and a random-utility model that separates institutional prestige from field of study (Economic Inquiry, accepted).
  • Pets and fertility: lottery windfalls and event studies on linked pet-registration and birth records show that children and pets are complements, not substitutes (SSRN working paper).

Covered by The New York Times (Dec. 2025)

The dissertation

Precision & robustness in adaptive testing

01 · The problem

Adaptive testing, finitely

Computerized adaptive testing assigns items on the fly to estimate ability with as few items as possible. Decades of asymptotic theory exist — but little governs how these algorithms behave in finite samples.

02 · Precision

No-regret & anytime-valid

Classical information-maximization is a no-regret algorithm under the Online Convex Optimization framework, and martingale concentration yields confidence intervals that stay valid at every step of the test.

03 · Robustness

Exploration under uncertainty

Borrowing Upper Confidence Bound methods from the multi-armed bandit literature, item selection stays stable against early-stage estimation error and uncertain item parameters during online calibration.

Research themes

Four lines of work

Select a theme to see the work behind it.

Foundations and estimation theory for latent-trait models.

  • Standard errors for the ability estimate in IRT models Plugin empirical-variance estimator · asymptotic validity
  • Identifiability of polychoric models with latent elliptical distributions Psychometrika, 90(2), 2025
  • Parametric bootstrap of standard errors in the 2-PL model PsyArXiv preprint
  • Existence & uniqueness of the MLE under log-concave assumptions Manuscript · PsyArXiv
  • When the test information curve misleads: exact bias, SE & RMSE curves Behaviormetric Society, 2025
  • Evaluation of PaGamO with a 2-PL IRT model 100K+ students · Bayesian state-space extension
Full project experience →

Finite-sample guarantees and stochastic approximation for CAT.

  • Precision & robustness in adaptive testing (dissertation) Online Convex Optimization × CAT · IMPS 2026
  • The generalized Robbins–Monro process for threshold estimation Journal of Mathematical Psychology, 2024
  • Stochastic approximation for computerized adaptive testing Behaviormetrika, 51(1), 2024
  • Consistency of Bayesian adaptive testing under the Rasch model Under revision · arXiv:2412.07170
  • Consistency & relative efficiency of a generalized Robbins–Monro process MathPsych, 2023
  • Threshold estimation with response time & confidence Taiwanese Psychological Association, 2021
Full project experience →

Decision processes, response times, and preference measurement.

  • Emotion as an arbitrator in decision-making under risk (fMRI) Utility-parameter estimation
  • Bisection-based vs. adaptive methods for loss-aversion elicitation Submitted
  • Joint Thurstonian IRT for Likert & comparative-judgment data Preference measurement
  • Retail investors during a financial bubble (fMRI) Manuscript in preparation
  • Coalition without trust: herd behavior in a bubble game fMRI hyperscanning · k-means response patterns
  • Cognitive structures underlying financial behavior Exploratory factor analysis · industry collaboration
Full project experience →

Causal work on Taiwanese administrative data.

  • Labor market sorting in Taiwan (AKM framework) Taiwan Economic Review, 54(2), 2026
  • Institution or major? College-admission preferences Economic Inquiry · accepted
  • Estimating preferences for college programs SETA 2024 · Population Association of Taiwan, 2024
  • Family, Health, and the Market Intra-household tax decisions · income-tax records
  • Cats, dogs, and babies: pets and fertility SSRN working paper · covered by The New York Times, Dec. 2025
  • Wage inequality in Taiwan over time Linked government administrative data
Full project experience →

Selected publications

Recent journal articles

  1. Yang, H.-H., & Hsu, Y.-F. (2024). The generalized Robbins–Monro process and its application to psychophysical experiments for threshold estimation. Journal of Mathematical Psychology, 120--121, 102855. [doi]
  2. Yang, H.-H., & Hsu, Y.-F. (2024). A note on the application of stochastic approximation to computerized adaptive testing. Behaviormetrika, 51(1), 259-276. [doi]
  3. Cheng, C., Yang, H.-H., & Hsu, Y.-F. (2025). Identifiability of polychoric models with latent elliptical distributions. Psychometrika, 90(2), 757–778. [doi]
  4. Chen, K.-M., Chen, Y.-C., Hung, C.-C., & Yang, H.-H. (2026). Institution or major? Understanding student preferences in college admissions. Economic Inquiry (Accepted, forthcoming).
  5. Chen, K.-M., Tsai, L.-T., & Yang, H.-H. (2026). Labor Market Sorting in Taiwan. Taiwan Economic Review, 54(2), 267–289. [doi]
All publications, working papers & talks →
Hau-Hung Yang

About

A quantitative psychologist

I am Hau-Hung (Hardy) Yang — hardy1yang on GitHub and across this site. My research is in mathematical psychology and psychometrics, with an emphasis on finite-sample theoretical guarantees for computerized adaptive testing. My dissertation, supervised by Prof. Yung-Fong Hsu, brings the Online Convex Optimization framework to Item Response Theory.

Alongside psychometrics, I collaborate with economists at NTU on projects using Taiwanese administrative data — labor-market sorting, college-admission preferences, and intra-household decision-making — and my earlier training was in cognitive neuroscience.

楊昊紘（Hau-Hung "Hardy" Yang），國立臺灣大學心理學博士（2026，量化方法），現任中央研究院統計科學研究所博士後研究學者。研究核心是心理計量與數理心理學——為電腦化適性測驗建立有限樣本的統計保證（線上凸最佳化、anytime-valid 信賴序列）；同時與經濟學者合作，以財稅、大學入學與寵物登記等行政資料進行實證研究。教學方面曾擔任臺大心理系統計課程助教八學期，並主講台灣經濟學會 AI 教學應用工作坊。

Contact

Let’s talk.

Open to research collaborations, seminars, and questions about the work.

hauhungyang@as.edu.tw

Institute of Statistical Science, Academia Sinica
128 Academia Road, Section 2, Nankang, Taipei 115, Taiwan