What Does Correlation Mean in Six Sigma?

Correlation measures the strength and direction of the linear relationship between two continuous variables, summarized in a coefficient that runs from −1 to +1. At +1 the points march up a perfect straight line; at −1, down one; near zero, the variables show no linear connection at all. In Six Sigma, correlation is a screening tool: a fast, cheap way to ask whether a suspected cause even moves with the effect before spending real effort proving anything.

How it’s used

During Analyze, a team holding a list of candidate causes plots each one against the output and computes correlation coefficients. Strong correlations promote suspects for deeper study; weak ones let the team retire theories cheaply. The plotting matters as much as the number. The coefficient captures only straight-line relationships, so a strong curved pattern — output rising and then falling as temperature climbs, say — can score near zero while being the most important relationship in the dataset. The rule among practitioners is absolute: look at the scatter plot before trusting the number.

The second discipline is refusing to leap from correlation to cause. Two variables can move together because one drives the other, because a third variable drives both, or by simple coincidence. Ice cream sales correlate with drowning incidents; summer causes both. Correlation nominates suspects. Verification — controlled comparisons, designed experiments — is what convicts them.

A worked example

Suppose a commercial bakery is fighting cracked crusts. The team suspects oven humidity and logs both humidity and crack counts across dozens of production runs. The scatter plot shows a clear downward drift — higher humidity, fewer cracks — and the coefficient comes out strongly negative. That is not proof; humidity varies with ambient weather, which also affects dough temperature. But it is enough to justify the next step: a controlled trial holding everything else steady while humidity is set deliberately. The correlation earned the experiment rather than replacing it. The example is generic; the sequence — plot, screen, verify — is the method itself.

• Plot first, compute second — the coefficient is blind to curves, clusters and outliers.
• A single extreme point can manufacture a strong correlation or bury a real one.
• Correlation never proves causation; always ask what third variable could drive both.
• A narrow operating range weakens correlation — a variable held nearly constant cannot reveal its effect.
• Near zero means no linear relationship, not no relationship.

Correlation buys you a suspect, never a conviction.

Scatter diagrams — correlation’s visual form — are introduced with the basic quality tools in our Yellow Belt program ($129, 14 hours). The coefficient itself, and the regression modeling that builds on it, arrive at Green Belt, where Analyze-phase screening is taught as a discipline rather than a reflex.