Bayesian Updates 🧠
No priors here — just pure learning from experience.
These are my working notes, trial-and-error logs, and favorite tricks in R, Stan, and beyond.
(Stay tuned)
Bayesian Updates 🧠
No priors here — just pure learning from experience.
These are my working notes, trial-and-error logs, and favorite tricks in R, Stan, and beyond.
(Stay tuned)
[2025-11-30]_Study
I am dedicated to addressing interpretability issues in longitudinal non-linear data. While most existing methods for longitudinal non-linear relationships are data-driven approaches that capture linear states, they lack interpretability for non-linear states. Therefore, I have been exploring suitable methods within the field of psychological research. During this process, I came across Hidden Markov Models (HMMs). This model is similar to a latent state transition observation model used in latent class analysis. While not my ideal solution for nonlinearity, I believe that it has potential applications in psychology. Consequently, I intend to expand my research to examine the application of this model in psychological contexts.
[2025-11-22]_Recommended readings
Why the Cross-Lagged Panel Model Is Almost Never the Right Choice(Lucas, 2025)
Building on this foundation in dynamic modeling, my research is now extending toward addressing major methodological limitations in the field. In particular, recent influential critiques (e.g., Lucas, 2025) have raised substantial concerns about the validity of the widely used cross-lagged panel model (CLPM), highlighting its inability to distinguish stable trait-like sources of individual differences from genuine within-person temporal causal effects.
This critical assessment has defined my key emerging research interest in exploring and applying superior alternatives that better isolate pure within-person change:
1) random intercept cross-lagged panel model (RI-CLPM)
2) STARTS model
3) Dynamic panel models
[2025-08-20]_Recommended readings
Comparing Likert and visual analogue scales in ecological momentary assessment
The study compared a seven-point Likert scale and a visual analogue scale (VAS) for measuring emotional states in a 14-day EMA design. Bayesian multilevel analyses showed that VAS produced higher within-person means, stronger temporal dependencies, and greater correlations with external psychopathology measures. At the same time, differences in variance, response stability, or participant burden were minimal. These findings suggest that VAS may be preferable in EMA studies targeting affective states relevant to psychopathology.
[2025-07-17]
In recent months, I have been studying natural sciences and applied statistics in medicine to integrate these perspectives into my research on psychometrics.
Cubic splines: 1) www.nature.com/articles/s41409-019-0679-x
2) bookdown.org/ssjackson300/Machine-Learning-Lecture-Notes/splines.html
Doubly robust (DR) methods pmc.ncbi.nlm.nih.gov/articles/PMC3070495/
Multi-state Models
Generalized Estimating Equations