econ-r

econ-r

Use this skill to conduct applied econometrics and statistics with the econ-r-mcp MCP server. The central rule is simple: do not invent empirical output. Coefficients, standard errors, p-values, diagnostics, sample counts, plots, model comparisons, and tables must come from executed R code and saved artifacts.

Default Workflow

1. Clarify the research question only when the prompt lacks the outcome, treatment/exposure, unit, time period, or decision context needed to proceed.
2. Define the estimand before choosing a model: association, ATE/ATT/LATE/local RD effect, forecast target, or descriptive statistic.
3. Use inspect_dataset before modeling. Identify variables, missingness, types, candidate unit/time columns, duplicates, and sample restrictions.
4. Identify the outcome, unit, time, treatment/exposure, instrument/running variable, controls, and clustering level.
5. Propose a research design in plain language, including why the identifying variation is credible or why it is only associational.
6. Run diagnostics before final interpretation: panel diagnostics for panel/DiD, first stage for IV, density/bandwidth checks for RD, stationarity checks for time series, balance/overlap checks for matching or weighting.
7. Estimate a baseline model with an exact formula and package command.
8. Run robustness checks that test the design, not just coefficient shopping.
9. Generate tables/plots with create_modelsummary, model-specific table outputs, and/or render_quarto.
10. Interpret results with clear separation between causal claims and associations.

Tool Use

• Use health_check at the start of a new environment or when an R package/tool failure is plausible.
• Use init_project for a new empirical workspace.
• Use install_r_packages only for packages in the allowlist; do not suggest arbitrary package installation without updating config explicitly.
• Use inspect_dataset before estimating models.
• Use run_panel_diagnostics before panel fixed effects or DiD.
• Use run_regression_fixest for OLS, fixed effects, clustered SEs, GLM-style fixest models, and baseline panel regressions.
• Use run_iv when the design has endogenous regressors and credible instruments.
• Use run_did for TWFE event studies or Callaway-Sant'Anna DiD.
• Use run_rdrobust for regression discontinuity.
• Use run_time_series_diagnostics before time-series modeling or forecasting.
• Use create_modelsummary for publication-style model tables from saved model objects or result JSON.
• Use render_quarto for reproducible reports.
• Use lookup_econometrics_guidance before econometric method choices. Prefer deterministic method tags such as iv, did, rd, panel_fe, time_series, or forecasting.
• Use lookup_econometrics_guidance with method="aer_textbook" when grounding an answer in the locally installed AER package's Applied Econometrics with R companion materials. Treat returned AER material as textbook/package guidance, not empirical output.
• Use list_artifacts and read_artifact to inspect logs, result JSON, tables, and plots.
• Use reproduce_run before relying on an earlier result when reproducibility matters.