Correlation Matrix Studio
Calculate correlations between numeric columns and inspect variable pairs.
Different variables. Shared patterns.
Row filters & observation export
Rules combine with AND. Numeric bounds are inclusive and exclude missing or invalid numbers. Variables remain eligible based on the imported dataset; a slice may have too few pairs or no variation.
Load observations to add filters.
Drop CSV file here
A place for connections to emerge.
Load observations to explore how your variables move together.
See what moves
together.
Build a correlation heatmap from numeric observations. Compare linear and rank relationships while keeping paired counts and data quality in view.
How to explore relationships
- Load observations. Open a CSV with named columns and one record per observation, or try the sample dataset.
- Choose your variables. Select eligible numeric columns and optionally add row rules. Rules combine with AND; Clear rules restores all imported observations. Select all restores the variable selection.
- Choose your perspective. Compare Pearson and Spearman. Click a cell or press Enter to inspect its scatter plot, coefficient and paired/excluded counts. A constant or sparse slice can produce an unavailable result.
- Keep the context. Download the coefficient matrix, detailed pair results, image or filtered observations. Keep the original dataset and method context. Correlation alone does not establish causation.
Should I choose Pearson or Spearman?
Pearson measures linear relationships between numeric values. Spearman measures monotonic relationships using ranks, with average ranks for ties. Both range from −1 to +1. The pair inspector plots original numeric values for either method, so its Spearman scatter is not a rank plot.
Which columns and rows are analysed?
Import raw CSV observations with named columns. At least 80% of nonblank values in the first 500 rows must parse as numbers for a column to qualify. Eligibility stays based on the imported dataset. Row rules combine exact case-sensitive text matches and inclusive numeric bounds with AND; invalid or missing numbers do not match numeric bounds. Clearing rules restores the source population.
How do missing values and the pair inspector work?
Each pair uses its finite paired observations. Click a matrix cell or activate it with Enter or Space to see the coefficient, paired/excluded counts and scatter plot. Too few pairs or no variation makes the coefficient unavailable, not zero. Plots above 1,000 valid pairs show a disclosed, evenly spaced source-order sample; the coefficient still uses every valid pair.
Does a strong correlation prove causation or significance?
No. Coefficients alone do not establish causation, statistical significance or practical importance. This version has no p-values, hypothesis tests or clustering. Inspect observations, paired counts and the sampling context before drawing conclusions.
Which results can I save, and is my data uploaded?
Matrix CSV preserves full available numeric precision. Pair-results CSV includes the method, unique variable pairs, coefficients and paired/excluded counts; it omits self-pairs and mirrors. Unavailable coefficients are blank and marked unavailable. Filtered-observation CSV preserves all original columns and matching source values. Image labels are rounded to two decimals. Analysis runs in the browser using libraries served from this site, and work is not automatically saved; browser memory limits apply.