메뉴
Run correlation analysis on connected data, identify strong relationships, and suggest regression models. Use when exploring relationships between variables.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Analyze an A/B test experiment and provide a clear ship/iterate/no-ship recommendation. Use when evaluating experiment results.
Discover natural groupings in data using clustering algorithms with evaluation and visualization. Use when segmenting data or finding patterns.
Run an end-to-end data analysis pipeline — connect, profile, analyze, and report. Use when performing a complete analysis workflow.
Connect to a data source, profile its schema and quality, and recommend appropriate analyses. Use when starting work with a new dataset.
Analyze a time series and generate forecasts with confidence intervals. Use when predicting future values from historical data.
Explore a graph or network dataset — structure, statistics, paths, and visualization. Use when working with DOT, GML, GraphML, or Mermaid files.
| name | analyze-correlations |
| description | Run correlation analysis on connected data, identify strong relationships, and suggest regression models. Use when exploring relationships between variables. |
| allowed-tools | mcp__localdata__describe_database mcp__localdata__execute_query mcp__localdata__analyze_hypothesis_test mcp__localdata__analyze_effect_sizes |
| argument-hint | <database-name> |
Discover and quantify relationships between variables in a connected dataset.
Get the schema. Call describe_database with the database name from $ARGUMENTS. Identify all numeric columns across tables. Note column names and which table each belongs to.
Extract numeric data. Call execute_query to select all numeric columns from the primary table. If data spans multiple tables, join on shared keys. Limit to 50,000 rows if the dataset is large.
Test pairwise correlations. For each meaningful pair of numeric columns, call analyze_hypothesis_test with test type "correlation". Use Pearson for normally distributed data, Spearman for skewed or ordinal data. Record the correlation coefficient and p-value for each pair.
Filter significant results. Retain pairs where p-value is below 0.05. Sort by absolute correlation strength. Classify relationships:
Measure effect sizes. For the top 5 strongest correlations, call analyze_effect_sizes to quantify practical significance. Compare statistical significance (p-value) against practical significance (effect size). Flag cases where a correlation is statistically significant but practically negligible.
Check for confounders. Look for pairs of strong correlations that share a common variable. Note potential confounding relationships where A correlates with B and A correlates with C.
Present results. Provide a ranked table of correlations with columns: variable pair, correlation coefficient, p-value, effect size, and interpretation. Group by strength category.
Recommend next steps. For the strongest relationships:
/localdata-mcp:regression with the best predictor-outcome pairs