What this answers

This calculator answers "how much would the fitted regression line change if this one point were removed?" combining a point's leverage (how unusual its predictor value is) and its residual (how far off the fit lands for it) into a single influence number.

How it is calculated

Cook's distance for each point scales its squared standardized residual by a factor involving its leverage relative to the rest of the data, so that a point needs both an unusual predictor value and a meaningfully large residual to score a high influence value; having only one of the two produces a much smaller Cook's distance than having both. A commonly cited (though not universal) rule of thumb flags values above 4 divided by n as worth a closer look.

Worked example

For four ordinary points following a clean linear pattern plus a fifth point at x equal to 20 that breaks the pattern, that fifth point combines both high leverage (an extreme x value) and a large residual (its y value does not fit the established trend), giving it by far the largest Cook's distance among all five points, well above the conventional threshold.

Assumption audit

Calculated from your data: each point's residual, leverage, and resulting Cook's distance, plus the conventional 4-over-n threshold for your sample size.

Evidence to review: whether a flagged point's outsized influence comes from a genuinely extreme but real observation or from a data-entry error; the number alone cannot distinguish the two.

You must verify: the real-world validity of a flagged point before deciding what to do about it; a high Cook's distance is evidence to investigate, never automatic justification for removing a data point.

What this result does not mean

A high Cook's distance means a point disproportionately shapes the fitted line, not that the point is necessarily wrong; some of the most scientifically important observations are exactly the influential, unusual ones. The 4-over-n threshold is a disclosed, commonly used heuristic, not a statistically derived cutoff.

Limitations

This calculator supports a single predictor for its current input shape; like Residual Diagnostics, the underlying math generalizes to multiple predictors but this page's data-entry format is built around one x and one y column.