Skip to content
Quality Tools | Part 7 of 8

Cause and Effect Diagram and Scatter Diagram: From Suspicion to Evidence

How to collect possible causes in a cause and effect diagram, check each one with data and test a lever with a scatter diagram and a trial.

Counting and sorting have located the problem on powder coating line 2: thin coating sits on the bottom hanging row, with 408 parts in four weeks or 5.1 percent of the parts on that row. Located is not explained. Two tools take the team from the pattern to the cause. The cause and effect diagram collects what could be behind it. The scatter diagram tests whether a suspected lever really moves the result.

The cause and effect diagram

The branches: people, machine, material, method, measurement, environment.

  1. Write the effect as precisely as the data allows: what, where, how much. “Coating problems” is too vague. “Thin coating on the bottom row” works.
  2. Collect causes with the people who do the work. One cause per note, no judging.
  3. Sort the notes onto the branches. Ask for each cause why it happens and add the answer as a smaller branch.
  4. Mark the three to five causes the team finds most likely.
  5. Decide for each marked cause how it is checked, and by whom.

Every cause gets a check

The diagram shows what people think, not what is true. On line 2 the team listed nine possible causes and checked each:

Possible cause Check Result
Lower guns at the same powder output as the upper guns Trial with settings Confirmed
Stroke of the gun mover ends above the lower edge of the parts Measured at the booth Confirmed
Hooks covered with cured powder, poor grounding Resistance of 40 hooks Partly, 6 of 40
Deep housings always hang on the bottom row Hanging plan Confirmed
No setting sheet per part family Asked both shifts Confirmed
Powder lot with finer grain Sorted by lot Rejected
Different settings by shift Sorted by shift Rejected
Gauge reads low on the lower third Reference foil Rejected
Humidity in the hall Scatter diagram Rejected

Four of nine suspected causes were rejected with data that already existed on the check sheet. That is a typical ratio. The rejected causes stay in the record, so that the same discussion does not start again in three months.

The scatter diagram

A scatter diagram plots pairs of two measured variables as points. If the cloud forms a band, the variables move together. The correlation coefficient r expresses how closely: near 1 or -1 for a narrow band, near 0 for a round cloud.

The team ran a trial on the bottom row. The powder output of the lower guns was set to ten values from 80 to 170 grams per minute, and two deep housings were measured at each setting. The 20 points form a narrow rising band:

  • The line through the points is thickness = 20 + 0.4 x powder output.
  • The correlation coefficient is 0.95.
  • Single parts differ from the line by up to 5 µm.

Use the line

The line answers practical questions. The current setting of 120 grams per minute gives 20 + 0.4 x 120 = 68 µm, which matches the mean of 67.8 µm that the histogram had shown for the bottom row. To reach 84 µm, the level of the top row, the setting must be (84 - 20) / 0.4 = 160 grams per minute. Every 10 grams per minute more adds 4 µm on average.

The setting was changed to 160. The mean of the bottom row rose to 84.1 µm, as predicted, and the parts with thin coating fell from 720 to 168 in four weeks.

Correlation is not cause

Two variables can move together by chance or because a third drives both. The difference lies in how the data was produced. A scatter diagram from historical records is a hint. It is good for choosing what to test. A trial in which one variable is set on purpose while everything else stays the same is evidence. The team on line 2 did the second.

Stay inside the range you tested. Methods that drill from a verified cause down to its root are covered in the Problem Solving module.

Newsletter

One Lean idea.
Every two weeks.