📊 Data Science & Analytics · Exploratory Analysis

Avoid correlation traps in stakeholder analyses

Stress-test relationships with confounder thinking and simple falsification checks.

intermediate~25 minData ScientistsAnalystsResearch Scientists

Steps

1Ask what could drive both variables; list top three confounders
2Stratify or control for obvious confounders and see if effect survives
3Run falsification tests: effect should vanish under placebo splits
4Frame findings as associations unless design supports causality
5Pre-register hypotheses for decisions that matter
6Communicate uncertainty ranges, never naked point estimates

Common Pitfalls

▲Simpson's paradox flipping conclusions after segmentation
▲Post-hoc story fitting presented as discovery

Commands

Install with skills CLI

$ npx skills add aniruddhaadak80/skills --skill exploratory-analysis-correlation-causation

Install globally

$ npx skills add aniruddhaadak80/skills --skill exploratory-analysis-correlation-causation -g

Tags

#statistics#causal-inference#data-science#exploratory-analysis

Install this skill

$ npx skills add aniruddhaadak80/skills --skill exploratory-analysis-correlation-causation -a opencode

SKILL.md

# Avoid correlation traps in stakeholder analyses

> Stress-test relationships with confounder thinking and simple falsification checks.

## When to Use This Skill

Stress-test relationships with confounder thinking and simple falsification checks.

## Steps

1. Ask what could drive both variables; list top three confounders
2. Stratify or control for obvious confounders and see if effect survives
3. Run falsification tests: effect should vanish under placebo splits
4. Frame findings as associations unless design supports causality
5. Pre-register hypotheses for decisions that matter
6. Communicate uncertainty ranges, never naked point estimates

## Common Pitfalls

- Simpson's paradox flipping conclusions after segmentation
- Post-hoc story fitting presented as discovery

View source on GitHub ↗

Profile distributions, missingness, leakage risks, and unit sanity before trusting any result.

📊 Data Science & Analytics·~30m

Open →

Build cohort and funnel analyses that hold upintermediate

Segment users by acquisition period and trace retention/funnels without sampling traps.

📊 Data Science & Analytics·~35m