drchronx/ai-agent-research-starter-kit/causal-inference-design-audit

Causal Inference Design Audit

Use this skill before writing causal claims or running quasi-experimental analyses.

Use references/causal-claim-ladder.md when deciding how strong the manuscript language can be.

Inputs

• Research question.
• Treatment/exposure, outcome, timing, and data source.
• Proposed design: experiment, DID, PSM, IV, RDD, DML, panel, or observational regression.
• Variables, confounders, mediators, moderators, and measurement timing.

Workflow

1. State the causal estimand.
2. Draw or describe a DAG.
3. Identify confounders, colliders, mediators, bad controls, and post-treatment variables.
4. Evaluate identification assumptions.
5. Choose an analysis strategy that matches the design.
6. Define robustness checks and falsification tests.
7. Rewrite claims to match the design strength.

Hard Constraints

• Do not let robustness checks substitute for identification.
• Do not control for post-treatment variables when estimating total effects.
• Do not call observational correlations causal without identification.
• Do not overinterpret text-mining variables as exogenous treatments without design support.

Output

Causal estimand:
DAG/confounding audit:
Identification assumptions:
Recommended method:
Bad controls:
Robustness and falsification:
Permitted claims:
Claims to avoid: