causal-evidence-analyzer

name	causal-evidence-analyzer
description	Analyze causal relationships by searching for and evaluating scientific evidence. Use when the user asks whether X causes Y, inquires about causal mechanisms, or wants to distinguish causation from correlation. Triggers include questions like "does X cause Y", "is there a causal relationship between", "what's the evidence that X leads to Y", or "causation vs correlation" discussions.

Causal Evidence Analyzer

Evaluate causal claims by searching for scientific evidence and organizing findings by strength of causal inference.

Evidence Hierarchy

Rank and label all evidence using these levels (strongest to weakest):

Level	Type	Causal Inference Strength
1	Meta-analysis of RCTs	Strongest—synthesizes multiple experiments
2	Randomized Controlled Trial (RCT)	Strong—random assignment isolates cause
3	Natural/Quasi-experiment	Moderate—exploits exogenous variation
4	Regression analysis (observational)	Weak—susceptible to confounding
5	Expert opinion/Descriptive correlation	Insufficient for causal claims

Workflow

1. Parse the Causal Question

Identify:

• Exposure/cause (X): The proposed causal factor
• Outcome (Y): The effect being investigated
• Population: Who/what is affected (if specified)
• Mechanism: Proposed pathway (if specified)

2. Search for Evidence

Search for academic/scientific sources using queries like:

• "X causes Y" meta-analysis
• "X" "Y" randomized controlled trial
• "X" "Y" causal effect
• "X" "Y" systematic review

Prioritize: peer-reviewed journals, Cochrane reviews, NBER/working papers from reputable institutions, government health agencies (CDC, WHO, NIH).

3. Evaluate and Classify Each Source

For each piece of evidence:

1. Identify study design → assign evidence level (1-5)
2. Note sample size and effect magnitude
3. Identify potential confounders acknowledged or unaddressed
4. Note replication status (single study vs. replicated finding)

4. Synthesize Findings

Write a prose narrative that:

1. Leads with the strongest evidence (Level 1-2 if available)
2. Explicitly labels each evidence level inline (e.g., "A Level 1 meta-analysis of 23 RCTs found...")
3. Identifies key confounders that complicate causal inference
4. Notes alternative explanations (reverse causation, common cause, selection bias)
5. Distinguishes causation from association clearly

5. State Conclusion with Appropriate Hedging

Use calibrated language based on evidence strength:

Evidence Available	Appropriate Conclusion Language
Consistent Level 1-2 evidence	"Strong evidence supports a causal relationship"
Level 2-3 with some inconsistency	"Moderate evidence suggests X may cause Y, though..."
Only Level 3-4 evidence	"Observational evidence shows association; causation not established"
Only Level 5 or conflicting evidence	"Current evidence insufficient to determine causality"

Causal Inference Concepts to Apply

Confounders: Variables that influence both X and Y, creating spurious association. Always ask: "What third factor could explain this relationship?"

Reverse causation: Y might cause X instead. Example: Depression associated with unemployment—but does unemployment cause depression, or does depression cause job loss?

Selection bias: Non-random sampling can create apparent relationships. Example: Hospital patients sicker than general population.

Dose-response: Stronger evidence if more X → more Y (or less Y if protective).

Temporality: Cause must precede effect. Cross-sectional studies cannot establish this.

Biological/theoretical plausibility: Is there a credible mechanism? Absence weakens causal claims.

Output Format

Structure response as prose paragraphs (not bullet lists), including:

1. Opening: Restate the causal question and preview the conclusion
2. Evidence review: Discuss findings organized from strongest to weakest evidence, with explicit level labels
3. Confounders and limitations: Key threats to causal inference
4. Conclusion: Calibrated statement on whether causation is established, with appropriate hedging

Always cite sources with enough detail to locate them (authors, year, journal if available).