The Scatter Plot, Explained

A scatter plot places pairs of measurements as points on two axes — a suspected cause on the horizontal, the effect on the vertical — to show whether the two move together. It is the fastest honest answer to the question every improvement team eventually asks: when X changes, does Y change with it?

How it’s used

The pattern of points does the talking. A tight upward drift says the variables rise together; a downward drift says one falls as the other climbs; a curve says the relationship changes character across the range; a formless cloud says the suspected cause is probably innocent. In Six Sigma work the scatter plot is the checkpoint between believing a cause and funding a fix — theories are cheap, and the plot is where they first meet data. With enough points, the plot also hints at strength: the tighter the band, the more of the effect the suspected cause explains.

Its one great trap is as famous as the tool: correlation is not causation. Two variables can move together because a third variable moves both, or by coincidence in a small sample — and small samples are especially generous with false patterns. The plot earns suspicion; only deliberate testing earns conviction.

A worked example: cure temperature and failed bonds

An assembly plant suspects its adhesive failures trace back to cure oven temperature. For three weeks, each batch’s oven temperature is logged alongside its bond failure count, and the pairs go onto a scatter plot. The picture is immediate: failures pile up at the low end of the temperature range and thin out above it. That pattern justifies the next step, not the fix itself. The team runs controlled batches at set temperatures, confirms the effect holds when temperature is moved deliberately, and only then tightens the oven controls. The plot found the suspect; the experiment convicted it. One detail keeps the team honest: failures never quite reach zero even at ideal temperatures — a reminder that temperature is a cause, not the whole story.

• Plot a few dozen pairs at minimum — three points make an anecdote, not evidence.
• Treat any pattern as association until you have changed X on purpose and watched Y respond.
• Hunt for lurking variables; a third factor can move both axes at once.
• If the points form clusters, stratify — mark each point by machine, shift, or lot before interpreting.
• Collect across the full operating range; a narrow slice of X can flatten a real relationship into apparent noise.

A scatter plot is where a theory meets its first honest audience.

Scatter analysis begins at Green Belt ($299, 35 hours), where reading the pattern is the skill. Black Belt training ($499, 60 hours, with a 150-question proctored exam) formalizes it — correlation coefficients, regression, and designed experiments that turn “these move together” into “this causes that, and by this much.”