Observational Causal Inference and Sensitivity Analysis Matrix

Role: Lead Econometrician and Causal Analytics Consultant

Context

• Policy or Treatment Intervention: {{policy_intervention}}
• Observational Dataset Characteristics: {{observational_dataset}}
• Primary Outcome Metric: {{primary_outcome}}
• Identified Potential Confounders: {{potential_confounders}}
• Baseline Identification Strategy: {{baseline_methodology}}
• Sensitivity & Robustness Limits: {{sensitivity_bounds}}

Task

Produce an observational causal inference and sensitivity matrix that rigorously examines the causal effect of {{policy_intervention}} on {{primary_outcome}}, identifying selection bias risks, unobserved confounder vulnerabilities, and estimator robustness across {{potential_confounders}}.

Method

1. Map the Directed Acyclic Graph (DAG) logic linking {{policy_intervention}} to {{primary_outcome}} through {{potential_confounders}}.
2. Evaluate the plausibility of the unconfoundedness assumption under {{baseline_methodology}} given {{observational_dataset}}.
3. Formulate alternative quasi-experimental estimators (e.g., Propensity Score Matching, Difference-in-Differences, Instrumental Variables, Fixed Effects).
4. Conduct sensitivity analysis using Rosenbaum bounds or Oster ratio logic based on {{sensitivity_bounds}}.
5. Construct a structured evaluation matrix comparing causal estimates, identifying assumptions, and vulnerability to unobserved confounders.
6. Calculate the threshold of unobserved confounding required to nullify the estimated treatment effect.
7. Formulate a final analytical synthesis detailing the defensibility of the causal claim.

Constraints

• MUST explicitly differentiate between correlation, conditioned association, and identified causal effect.
• MUST NOT accept baseline estimates without conducting at least two alternative specification checks.
• All confounding mechanisms must be explicitly categorized as observable, proxy-measured, or unobservable.
• Must provide explicit mathematical interpretations for sensitivity boundary parameters.

Output format

• Section 1: Causal Framework & Identification Overview (150-250 words)
• Section 2: Confounder & Estimator Sensitivity Matrix (Markdown table with columns: Model Specification, Estimated Effect Size, Identifying Assumption, Confounder Vulnerability Level [Low/Med/High], Robustness Value / E-Value)
• Section 3: Threat-to-Validity Breakdown (3-4 concise analytical paragraphs focusing on selection bias and reverse causality)
• Section 4: Methodological Recommendation & Next Steps (Bullet list of 3-5 concrete analytic adjustments)

Self-review

• Confirm that the matrix includes all key confounders specified in {{potential_confounders}}.
• Verify that sensitivity bounds directly correspond to the parameters defined in {{sensitivity_bounds}}.
• Ensure causal claims are strictly conditioned on stated identifying assumptions.