| name | statistical-modeling |
| description | Use when estimating a statistical or econometric model, running a regression, specifying an identification strategy, testing a hypothesis, or fitting any model to empirical data. Guides the process (assumptions, estimation, reporting, diagnostics) without forcing a fixed method list. |
Statistical Modeling
Overview
This skill defines the process for statistical modeling in empirical research. It is method-agnostic and field-agnostic — the appropriate method for a research question comes from the data and the estimand, not from a fixed list. Reference files organized by method family (cross-section, panel, causal inference, time series) offer starting points, but they are not exhaustive and not a boundary. Bayesian methods, machine learning, spatial models, and any other approach can be used as long as the modeling process below is followed.
When to Use
- Estimating any regression or model
- Designing an identification strategy for a causal question
- Testing a hypothesis
- Fitting a time-series model
- "Run a DiD for me"
- "Fit a VAR"
- "Is this the right model for the data?"
- Diagnosing a model after estimation
Mandatory Steps
Every estimation follows the six-phase modeling process below. Each phase is mandatory — skipping one corrupts inference downstream.
- Define the estimand
- Verify assumptions match the data
- Choose the method
- Estimate
- Diagnose
- Report
Each phase is expanded below.
1. Define the Estimand
What parameter do you want to learn? Is it causal or descriptive? What population does it apply to? Without a clear estimand, method choice is premature. Write the estimand down in one sentence before touching the data.
2. Verify Assumptions Match the Data
Every method has assumptions — exogeneity, stationarity, parallel trends, common support, no interference, proportional hazards, and so on. Check the ones that apply to your method. Document which assumptions are plausible given the context and which are weak. Weak assumptions force either a different method or a robustness discussion.
