Lance Gui
Community@lancegui · Tucson
PhD Econ Student at UA
Agent Skills by Lance Gui
Showing 18 vetted skills indexed across 1 GitHub repositories.
lanceguianalysis-state-management
Maintains durable YAML-based analysis state in docs/analysis/ for resumable long-running empirical projects.
lanceguianalysis-craft
Guides minimal, legible, and surgical editing of analysis code in R, Julia, and Python.
lanceguiexecuting-analysis-plans
Executes approved empirical analysis plans with validated steps and parallel robustness checks.
lanceguidata-preparation
Plans and audits dataset cleaning, joining, deduplication, and reconciliation with a logged decisions ledger.
lanceguidata-contracts
Validates data transforms, joins, and aggregations with contracts and invariants in R, Julia, Python, and Stata.
lanceguicausal-identification
Validates causal identification strategies and diagnostics before estimating treatment effects.
lanceguiusing-causal-powers
Routes data analysis, econometrics, and causal-inference tasks to the appropriate discipline skill before execution.
lanceguianalysis-checkpoints
Enforces human-in-the-loop checkpoints before consequential analytical decisions during data analysis execution.
lanceguipre-analysis-plan
Locks hypotheses, estimands, and specifications before outcome data is seen.
lanceguistructural-estimation
Estimates structural economic model primitives and simulates counterfactual equilibria in R, Julia, or Python.
lanceguiwrong-number-debugging
Diagnose wrong computed results by bisecting data pipelines to locate the failing stage.
lanceguiquestion-framing
Defines the estimand, population, unit, and decision before any data analysis begins.
lanceguianalysis-review
Reviews data analyses for silent failures like leakage, fanned-out joins, and fished specifications.
lanceguipredictive-modeling
Builds leakage-free predictive models with deployment-matched evaluation in R, Python, or Julia.
lanceguiresult-verification
Verifies analysis results by reconciling totals, reproducing from clean sessions, and tying figures to prose.
lanceguidescriptive-evidence
Produces validated descriptive statistics, trends, and stylized facts from empirical datasets.
lanceguifigure-craft
Renders verified analysis results as presentation-ready figures in R, Python, or Julia.
lanceguiproject-organization
Organizes empirical research repositories into paper-centric folder structures with standardized naming and git checkpoints.
Frequently Asked Questions About Lance Gui
FAQPage SchemaWhat tasks can I accomplish with Lance Gui's analysis skills?▼
You can frame research questions, build and validate datasets, run causal designs (DiD, IV, RDD, synthetic control), estimate structural models like BLP demand and dynamic discrete choice, build predictive models, debug wrong numbers, verify results, and produce publication-grade figures in R, Julia, Python, or Stata.
Who are these skills designed for?▼
They target empirical economists, PhD researchers, and data analysts producing academic papers, policy evaluations, or replication packages. Anyone estimating causal effects, structural models, or predictions from data — and needing auditable, reproducible results — benefits from the enforced checkpoints and validation discipline.
How do the skills enforce analysis quality in practice?▼
Skills fire at specific phases: question-framing fixes definitions before execution, pre-analysis-plan locks hypotheses before outcomes are seen, data-contracts asserts join cardinality and invariants, analysis-checkpoints forces human approval for consequential decisions, and result-verification reproduces numbers from a clean session.
What languages and prerequisites do these skills require?▼
Skills operate in R, Julia, Python, and Stata, with figure-craft supporting ggplot2, matplotlib, and Makie. No other dependencies are specified. The using-causal-powers skill acts as the entry router, dispatching to the appropriate discipline skill based on the analytical goal.
How do the skills handle causal versus predictive work differently?▼
Routing is by goal, not algorithm. Causal claims require stated identification strategies and design diagnostics via causal-identification, even when ML methods are used. Prediction deliverables use predictive-modeling with leakage checks and deployment-mirroring splits, keeping the two evidentiary standards separate.