Each Tool Can Be Drawn, But Data Can't Be Linked? —— Five-Step Method for Interconnected Analysis of the Seven QC Tools
The quality meeting at a hardware factory was quite lively. The quality engineer posted a Pareto chart, saying, "Dimensional deviations account for 70% of the defects." The process engineer followed with a histogram, stating, "The distribution looks normal; the problem must be with the equipment." The equipment engineer then presented a scatter plot, claiming, "Temperature and dimensions are basically unrelated." All three charts were drawn meticulously, but their conclusions contradicted each other, and the meeting ended with "Let's investigate further."
The issue isn't in the drawing skills but in the fact that the three charts are based on different sets of data. The Pareto chart uses the inspection summary from last month, the histogram uses a sample of fifty pieces from this week, and the scatter plot uses parameters recorded by the equipment itself—different time windows, different sampling methods, and different measurers. Combining these three charts is like averaging the scores from three different exams.
The true power of the seven tools lies not in how standard a single chart is drawn but in whether they can form a chain of reasoning.
1. Why the Seven Tools Always "Draw Their Own"
The first reason is the different criteria. The Pareto chart uses the inspection summary from last month, the histogram uses a sample of fifty pieces from this week, and the scatter plot uses parameters recorded by the equipment itself—different time windows, different sampling methods, and different measurers. Combining these three charts is like averaging the scores from three different exams.
The second reason is different people and different outputs. Each chart has a different person in charge, and each person concludes their own section: "Dimensional deviations are the main issue," "The equipment is relatively stable," "Temperature has little impact." The conclusions remain on the charts, and no one passes them to the next tool for verification, so no one discovers that the previous conclusion is actually unfounded.
Remember the main thread: check sheets define the facts, histograms and stratification identify "which group is different," scatter plots determine "whether it is the cause," Pareto charts and fishbone diagrams decide "which one to tackle first," and control charts answer "whether the process is stable after the change." If any link in this chain is broken, the final conclusion will be unsupported.
2. The Starting Point of Interconnected Analysis: Define a Set of "Common Data" First
Before drawing any chart, write down five things on a single page:
1. Time window and scope. Use a complete batch or a shift as a unit, ensuring it covers the period when the issue occurred.
2. Sampling method. Specify how to sample, how many to sample, and who will measure. The entire process should use only one method. Changing the method or the measurer halfway through will distort the distribution chart.
3. Measurement system. Clearly state the gauge number and calibration status. Before different people measure, ensure the repeatability and reproducibility are acceptable; otherwise, all subsequent charts will amplify measurement errors.
4. Stratification fields. Prepare at least two, such as machine/ cavity, shift, and raw material batch number. Without stratification fields, stratification is impossible, and bimodal distributions cannot be investigated.
5. Record attribution. Each chart should be traceable to the original records from the production line. The person who draws the chart should verify the data, and "data provided by others" is not acceptable.
This page is written on the check sheet. The check sheet is not just a small tool used casually; it is the entry point for the seven tools. If the fields are wrong, the subsequent six charts will also be incorrect.
3. Five-Step Interconnected Path
Step 1: Use check sheets to solidify facts. First, fix the fields, units, and recording times, and continuously record data for a time window. The output of this step is not a chart but a clean, verifiable data table.
Step 2: Use histograms to observe the shape. A unimodal, centered, and roughly symmetrical distribution indicates that the data comes from the same population. If a bimodal, isolated, flat-topped, or significantly skewed distribution appears, it usually suggests that the data is mixed from different sources—this is the signal to proceed with stratification.
Step 3: Use stratification to break it down. Stratify by machine, cavity, shift, and raw material batch number, comparing the mean and dispersion of each layer to identify the layer with the greatest difference. The goal of this step is to narrow the scope from "the entire line" to "a specific machine or shift."
Step 4: Use scatter plots to verify relationships. For the few candidate factors remaining after stratification, plot the factor values on the x-axis and the characteristic values on the y-axis. Observe the trend, dispersion, and any outliers. Note: correlation does not equal causation. Multiple variables can change within the same time frame, and the scatter plot provides a list of suspects, not a verdict.
Step 5: Use Pareto charts to prioritize and control charts to close the loop. Rank improvement priorities based on defect percentages or loss amounts, focusing on a few key items first. After implementing the measures, return to the control chart to confirm whether the process is under control and whether the capability has improved, using the Cpk comparison before and after the improvement as evidence of a closed loop.
If no relationship is identified in Step 4, insert a fishbone diagram: list all candidate factors (people, machines, materials, methods, environment, measurement) using the fishbone diagram, and then return to the scatter plot to verify each one. More tools are not always better; the key is to fill in the missing links.
4. A Simple Example
A complaint about the outer diameter of stamped parts being out of tolerance was addressed using the five steps:
- Check sheet: Continuous three shifts, sample 120 pieces, record outer diameter, machine, cavity, shift, and mold temperature.
- Histogram: Bimodal distribution, indicating the problem is not "overall too large or too small" but mixed from two sources.
- Stratification: Stratify by cavity, the mean value of Cavity 1 is 0.03mm lower than the others.
- Scatter plot: Focus on Cavity 1, the outer diameter shows a clear positive correlation with mold temperature (45-62°C), with a correlation coefficient of about 0.72.
- Pareto chart and control chart: List "mold temperature fluctuation in Cavity 1" as the top priority, install a mold temperature control band and tighten the control band range. Two weeks later, the control chart shows no abnormalities, and the Cpk increases from 0.86 to 1.35.
The entire chain used only a few hours of on-site data, but each step narrowed the scope: 120 pieces → two sources → one cavity → one factor → one measure. There is no redundant work between the tools, and each chart contributes to reducing the scope for the next chart.
5. Four Common Pitfalls
Pitfall 1: Charts for charts, decisions for decisions. The charts are completed and included as attachments in the monthly report, but they do not become an improvement list or assign responsibilities. To judge whether a chart is useful, see if it generates an action item with a time and responsible person.
Pitfall 2: Criteria drift. The histogram uses data from the entire day, the Pareto chart only uses data from nonconforming products, and the stratification only selects the day shift. Changing the data source renders all subsequent reasoning meaningless. The criteria should be fixed in the first step, and any change in criteria must be redrawn and noted.
Pitfall 3: Confusing correlation with causation. When the mold temperature increases, the dimensions indeed change, but the raw material batch number also changed during the same period. Concluding that installing a temperature control system is necessary might be a waste of money. After the scatter plot, it is best to perform another stratification or small-scale verification.
Pitfall 4: Piling up charts without focusing. Drawing all seven charts may seem rigorous, but it does not provide a clear answer. The sequence of tools is the sequence of focus. After stratification, the candidate factors usually narrow down to two or three. Focusing efforts on these factors is more effective than drawing all seven charts.
6. One-Sentence Summary
The seven tools are not seven separate instruments but a chain of reasoning that starts from facts, gradually narrows down, and finally closes the loop with data. If they can't be linked, even the most standard charts are just seven pretty pictures.
First, define common data, then narrow down step by step according to "check sheet → histogram → stratification → scatter plot → Pareto chart and control chart." Only by linking the tools can the scope be reduced from the entire line to a single factor.
Knowledge code: 5.2.4
Version: v20260923
Author: QTank QTank is dedicated to providing systematic knowledge, methodologies, and practical tools for quality management professionals, helping companies continuously improve their quality capabilities.