Problem Stratification is the practice of separating problem data into meaningful groups so patterns become visible.

Instead of asking only:

How many defects did we have?

the team asks:

Where, when, and under what conditions are the defects concentrated?

Stratification often narrows a root cause investigation before complex analysis is needed.

Start with a clearly defined problem

The problem measure should be consistent before the data is split.

Problem Statement helps define the expected condition, actual condition, gap, scope, timing, and impact.

If the definition changes between groups, the comparison becomes unreliable.

Choose meaningful stratification factors

Common factors include:

  • product;
  • machine;
  • cavity;
  • shift;
  • operator;
  • supplier;
  • material lot;
  • time of day;
  • defect type;
  • customer.

The categories should reflect how the process could realistically vary.

Do not create dozens of arbitrary groups simply because the data system allows it.

Use Pareto analysis inside the problem

Pareto Analysis can show which defect types, products, or assets account for most of the loss.

For example:

  • 60% of defects occur on Product Family B;
  • within Product Family B, 75% occur on Machine 4;
  • within Machine 4, most occur on second shift.

Each level narrows the investigation.

Compare similar good and bad conditions

Stratification becomes stronger when the team can compare conditions where the problem is present with similar conditions where it is absent.

Is/Is Not Analysis provides a structured comparison of what the problem is and is not.

This can reveal differences that deserve causal testing.

Look for change points

If a problem suddenly appears in one stratum, ask what changed.

Change Point Analysis helps compare the last known good condition with the new abnormal condition.

A supplier, setup, software revision, tool, or staffing change may become visible only after the data is separated.

Be careful with small samples

A group with very few observations may appear unusually good or bad by chance.

The team should consider:

  • sample size;
  • exposure;
  • production volume.

Ten defects on ten thousand units is different from ten defects on fifty units.

Use rates or normalized measures where appropriate.

Use stratification to guide investigation

A stratification result is not automatically a root cause.

If defects are concentrated on one machine, the machine may be:

  • creating the problem;
  • running a different product mix;
  • receiving different material;
  • measured differently.

Cause Verification should test whether the suspected condition actually changes the problem.

Keep the analysis visual

Simple tools can be powerful:

  • Pareto charts;
  • stacked bars;
  • run charts;
  • tables;
  • heat maps.

The visual should make concentration visible without creating unnecessary statistical complexity.

Common mistakes

Stratifying before the problem is defined, using categories with very small exposure, comparing counts when production volumes differ significantly, creating too many categories, treating concentration as proof of cause, and ignoring meaningful factors because the data is difficult to obtain are common mistakes.

Practical sequence

  1. define the problem measure.
  2. identify realistic sources of variation.
  3. separate the data by meaningful factors.
  4. normalize for exposure where needed.
  5. rank important concentrations.
  6. compare good and bad strata.
  7. look for change points.
  8. develop causal hypotheses.
  9. verify suspected causes.
  10. repeat stratification if the problem remains broad.

The practical lesson

Problem Stratification makes a large problem smaller.

Before asking why the entire process is failing, find where the failure is actually concentrated.

This topic also connects with Problem Decomposition. Use that method when the improvement requires the related operating or management discipline.