What Does Stratification Mean in Six Sigma?

Stratification is the practice of dividing data into meaningful layers — by machine, shift, operator, supplier, region, or product line — before drawing conclusions from it. Aggregated data averages away its own evidence; stratified data lets each source testify separately. Among the seven basic quality tools it is the odd one out: not a chart, but a habit of thought that makes every other chart sharper.

How it works

Stratification begins at collection, not at analysis. Every observation gets tagged with its origins — which line produced it, which shift, which template, which channel — because layers that were never recorded can never be separated later. The analysis itself is simply the same chart drawn several times: a Pareto per shift, a histogram per machine, a scatter plot with each supplier’s points marked differently. The differences between the panels are the finding, and when one panel looks unlike the rest, the cause hunt has an address. The tagging costs seconds at collection time and is nearly impossible to reconstruct afterward — a lesson most teams learn exactly once.

The technique guards against two standing illusions. An aggregate can look stable while one stratum burns and another compensates. And an aggregate can appear to improve while no stratum improved at all — the mix simply shifted toward the better source. Both illusions dissolve the moment the layers are viewed separately.

A worked example: the improvement that wasn’t

A claims office celebrates a falling error rate. A skeptical analyst stratifies the data by intake channel and finds two flat lines: paper claims, error-prone as ever, and portal claims, clean as ever. Neither channel improved. The overall rate fell only because customers drifted toward the portal, changing the blend. The distinction is not academic — the office was about to credit a training program for the improvement and renew it. Stratification redirected the money toward the actual lever: moving the remaining paper volume, and its errors, onto the portal. Same data, same office — a completely different decision.

• Decide the strata before collecting data, and tag every observation at the source.
• Draw the same chart per stratum and read the differences between the panels.
• Audit every improvement for mix shift: did the layers improve, or just the blend?
• Slice no thinner than the data can support — ten strata of six points apiece prove nothing.
• When one stratum stands out, hunt for a cause, not a culprit.

Aggregated data keeps secrets; stratified data testifies.

Stratification runs through the entire curriculum. Our Yellow Belt ($129, 14 hours) introduces it alongside the basic tools, and the Green Belt ($299, 35 hours) makes it second nature — in Measure you plan the data tags before collecting, and in Analyze you slice before you conclude. It may be the cheapest analytical habit in the whole method.