01Prediction
Behavioral Prediction
Gradient-boosted tabular models of human outcomes — with calibration checks, honest baselines, and validation discipline treated as the deliverable, not the garnish.
Quantitative Behavioral Science
Modeling human behavior with statistics, code, and measurement discipline.
I am a social psychologist and research statistician. My work builds predictive models of human behavior — gradient-boosted learners, transformer-based language models, and the psychometric instruments that keep both honest — deployed inside institutional data systems at operational scale.
xgboost · deberta-v3 · sbert · bertopic · irt · pyspark · r → python
I.
Research programs — three lines of work
001 / 004
Theory in, prediction out, measurement all the way through.
Psychological theory sets the question; statistics and machine learning carry the load; nothing ships without a validation story.
02Measurement
Measurement & Psychometrics
Instruments before models: item response theory, reliability, construct validity — the quantitative-psychology layer that sits under every credible prediction task.
03Language
Language as Data
Transformer pipelines — DeBERTa-v3 classifiers, sentence encoders, BERTopic — turning free text into structured, analyzable signals for organizational research.
II.
Method — one pipeline, four disciplines
002 / 004
The same four moves, every project.
Frame
theory → estimand
Translate a behavioral question into a decision-relevant estimand before any data is touched.
Measure
irt · reliability · validity
Build or audit the instruments; a model on a broken measure is a precise error.
Model
xgboost · deberta-v3
Tabular and language models, selected against baselines that have earned their numbers.
Deploy
pyspark · r → python
Ship into institutional pipelines; monitor drift the way you would monitor an instrument.
III.
Selected work — talks, papers, code
003 / 004
Posted as it clears review.
Presentations & Papers
Military Operations Research Society (MORS) — Symposium contributions
Predictive behavioral pipelines and measurement work presented to the defense-analytic community across multiple symposia.
category · conference
SSRN distribution versions — conference decks & working papers ↗
Distribution-cleared versions of presentations and papers, posted as they clear review.
author page · papers.ssrn.com
Curriculum vitae
Full publication and service record.
pdf forthcoming
Open Code
Selected public repositories: modeling tooling, text-analysis pipelines, and legacy-framework translations.
github.com/limynet
R → Python translations
RShiny-era analytical frameworks rebuilt on Python infrastructure (Streamlit, Polars, PySpark) for long-term maintainability.
category · infrastructure
Measurement utilities
Small instruments: scoring, calibration plots, item-analysis helpers used across projects.
packaging forthcoming
IV.
The ontology — a knowledge organization system
004 / 004
This domain is also a knowledge system.
Behind this page sits a personal KOS — private by default: a wiki-shaped ontology database of concepts, methods, artifacts, and the relations between them. Working notes and notebooks sync behind access control; public excerpts are opt-in. The graph also feeds retrieval for my own research agents.
- modelskos-lite · d1 graph
- relationsbroader · narrower · related
- captureonenote / keep → ingest
- accessprivate by default · opt-in excerpts
- statusin construction