Why Do Problems Persist Despite Drawing Pareto Charts and Filling Fishbone Diagrams? —— Ten Common Misuses of the Seven QC Tools and Their Corrections

1. Introduction: A Workshop That Used All Tools but Solved Nothing

In the SMT assembly workshop of an electronic component company, the welding defect rate climbed from 1.2% to 2.8% within half a year, leading to a continuous stream of customer complaints. The quality department organized a QC team, determined to "solve the problem with scientific tools."

The team was very diligent: they first designed a record form to collect data for a month, used a Pareto chart to identify the top three defects, drew a fishbone diagram to analyze the causes, and used a histogram to examine the size distribution. They even tried to plot a few sets of variable relationships with a scatter diagram. The report was over twenty pages long, with beautiful charts that pleased both the leadership and the customers. However, three months later, the defect rate remained unchanged at 2.7%, and even increased slightly.

What went wrong? Later, they consulted a veteran quality professional with twenty years of experience. He asked only three questions: "Is your Pareto chart stratified by product model?" "Is 'insufficient personnel training' a cause or a countermeasure in your fishbone diagram?" "How many samples did you collect for the histogram?" The team was at a loss: the Pareto chart was drawn with all data mixed together; the fishbone diagram was filled with "enhance training" and "improve awareness"; the histogram was based on only 18 samples.

The answer was clear: the tools were used correctly in form but incorrectly in method. This is not an isolated case. In many manufacturing enterprises, the "formal use" of the seven QC tools is far more common than imagined—charts become more and more beautiful, but problems become more and more confusing. This article will break down the eleven most common misuse scenarios in workshops and provide actionable correction methods.

2. Understand the Division of Labor Among the Seven Tools to Avoid Misuse

The seven QC tools—record form, stratification, Pareto chart, fishbone diagram, scatter diagram, histogram, and control chart—are often treated as "seven independent weapons," which is the biggest cognitive bias. In reality, they are seven links in a data processing assembly line, each answering a specific question:

The root cause of misusing tools is almost always skipping steps and picking the wrong link: drawing a Pareto chart without stratification; using a fishbone diagram to "conclude" without verification; using a histogram as a control chart and treating specification limits as control limits. Remember one thing: the tools are not wrong, but the people using them are—think clearly about the question you want to answer before deciding which tool to use.

3. Ten Common Misuse Scenarios and Their Corrections

Scenario One: Pareto Chart Without Stratification, Directly Summarizing All Factory Data

A stamping factory combined all defects from three workshops into one Pareto chart, with "burrs" ranking first. Consequently, the entire factory launched a vigorous campaign to tackle burrs. However, Workshop A indeed had many burrs, but the main issue in Workshop B was "dimensional tolerance," and in Workshop C, it was "scratches"—a single summary chart mixed three entirely different root causes together.

Correction Method: Stratify first, then sort. At least stratify by workshop, product family, or shift, and draw a separate Pareto chart for each stratum. If a summary is necessary, first confirm that the problem structures in each stratum are consistent (the top three issues are the same across strata). The value of a Pareto chart lies in "focusing," and the prerequisite for focusing is data homogeneity.

Scenario Two: "Other" Category in Pareto Chart Takes Up Too Much, Rendering the Top Items Meaningless

Some Pareto charts lump dozens of minor defects into the "other" category, resulting in "other" taking up 35% and ranking second. In this case, the "80/20" rule is completely ineffective—focusing on the top three items only explains less than half of the problems.

Correction Method: Keep the "other" category below 10%, preferably no more than 5%. If "other" is too large, it indicates that the stratification dimension is incorrect. Re-stratify from a different angle until the main categories can explain most of the problems.

Scenario Three: Fishbone Diagram Lists "Countermeasures" as "Causes"

"Insufficient personnel training," "aging equipment," "enhanced inspections," "improved quality awareness"—these terms fill the fishbone diagrams in workshops. Upon closer inspection, is "insufficient training" a cause? It is merely an explanation of why employees make mistakes, or even an unverified guess. "Enhanced inspections" are fundamentally countermeasures, and listing them as causes is like prescribing medicine before making a diagnosis.

Correction Method: Each cause must answer "why it leads to this defect." After writing, ask yourself: if I eliminate this cause, will the problem really disappear? If you cannot answer this question, either dig deeper or delete it. The fishbone diagram should only carry "cause hypotheses," and countermeasures should be left for subsequent improvement stages.

Scenario Four: Fishbone Diagram Filled with All Six Major Categories, Becoming a "Hodgepodge"

All six main bones (people, machines, materials, methods, environment, measurement) are filled, with seven to eight sub-causes under each, resulting in a dense chart with fifty to sixty "causes." It looks comprehensive, but in reality, none of the causes have been verified, and the chart offers no actionable guidance.

Correction Method: Focus on the 2-3 most concentrated main bones (e.g., data indicates "machines" and "methods" are the most suspicious), and dig deep into each. Explore 3-5 layers under each main bone (ask more "why" questions) rather than spreading out horizontally. A deep and narrow fishbone diagram is more effective than a wide and shallow one.

Scenario Five: Record Form Designed as a "Checklist," Recording No Facts

The most common error in record forms looks like this: a column of defect types followed by a series of empty spaces where operators mark a "正" (tick) when they find a defect. By the end of the month, only the "quantity" is recorded, with no information on "which equipment," "which time period," "which batch," or "what phenomenon." When further analysis is needed, the data is useless.

Correction Method: Record forms should capture at least four elements—time, location (equipment/station), object (batch/product), and phenomenon (specific defect description). Instead of designing them as statistical tables, design them as "chronological logs" to record facts, and leave the statistics for the analysis stage. Spending ten more minutes in the data collection stage can save ten hours in the analysis stage.

Scenario Six: Drawing Conclusions from a Histogram with Insufficient Sample Size

Using 18 or 20 samples to draw a histogram, the graph appears jagged and erratic, yet someone reads "bimodal distribution" or "unstable process" from it. When the sample size is insufficient, the shape of the histogram is purely random noise, and any interpretation is arbitrary.

Correction Method: The sample size for a histogram should be at least 50, and preferably over 100. The number of bins should be determined by an empirical formula (approximately the square root of the sample size). Too few bins can obscure distribution characteristics, while too many can create severe jaggedness. When the sample size is insufficient, collect more data before interpreting.

Scenario Seven: Judging Conformance Based on the Mean Being Within Tolerance in a Histogram

In a machining workshop, 80 shaft diameters were measured, with an average of 20.01mm and a tolerance of 20±0.05mm. The average is very close to the centerline, and the workshop supervisor says "no problem." However, the histogram shows a wide and skewed distribution, with the tail already extending beyond the lower tolerance limit—batch defects are just a matter of time.

Correction Method: When examining a histogram, consider three things: the distribution shape (normal, bimodal, skewed), the comparison of distribution width to tolerance width, and whether the distribution center aligns with the tolerance center. If the distribution tail touches the specification limit, even if the average is perfect, it must be addressed. The average is like a "health report," while the distribution is like a "CT scan."

Scenario Eight: Mistaking Correlation for Causation in a Scatter Diagram

A scatter diagram shows a "clear positive correlation" between workshop temperature and welding defect rate, leading the team to conclude that "temperature causes defects" and spend a large sum on upgrading the temperature control system. However, the defect rate did not decrease—the actual cause was a change in solder paste batches during that period, and temperature just happened to change in sync.

Correction Method: A scatter diagram can only prove that "two variables change together," not that "one causes the other." After observing a correlation, ask three questions: Is there a third variable affecting both? Does the logic make sense? Have you verified the causality with controlled variables? Confirming causality ultimately requires experiments (such as DOE), not just a scatter diagram.

Scenario Nine: Incorrect Stratification Dimension or Overly Fine Stratification

Someone stratified the data by "day of the week" and found the highest defect rate on Wednesdays, excitedly concluding that "Wednesdays have a problem." In reality, Wednesdays just had the highest production schedule and output—normalizing by production volume showed no abnormal defect rate on Wednesdays. Others stratify too finely, leaving only three to five samples in each cell, which is statistically meaningless.

Correction Method: Choose stratification dimensions based on process logic (people, machines, materials, methods, environment, measurement), not arbitrarily "testing" various irrelevant dimensions. If the sample size in each stratum is insufficient after stratification, combine adjacent strata or extend the collection period to ensure each stratum has enough data to support judgments.

Scenario Ten: Using Specification Limits as Control Limits in a Control Chart

Many companies set the upper and lower limits of control charts directly to the tolerance limits, believing that "points within the tolerance are normal." This is the biggest misunderstanding of control charts: control limits reflect the historical variation of the process and have nothing to do with tolerances. Using specification limits as control limits means that the process may already be out of control, but it won't show on the chart. By the time points exceed the limits, defects have already occurred in batches.

Correction Method: Control limits must be calculated using process data (mean ± 3 standard deviations), and the process stability must be confirmed (no out-of-control points) before calculating the control limits. Control charts should be continuously updated, not just drawn once and hung on the wall. Specification limits answer "whether it is acceptable," while control limits answer "whether the process is stable"—two different things, don't mix them up.

Scenario Eleven: Using Each Tool Independently, Never Linking Them Together

Another more hidden misuse is when the record form collects data and is then archived, the Pareto chart sorts the data and is then finished, the fishbone diagram is drawn and hung in the meeting room, and the seven tools operate independently. Analysis is cut off at the exit of each link—data is data, charts are charts, and improvements are improvements, with no transmission in between, meaning none of the seven tools are used effectively.

Correction Method: Use the tools in a chain: "collect → stratify → focus → hypothesize → verify → monitor." Data from the record form should be stratified first, and the stratified homogeneous data should be used to draw a Pareto chart to focus on key issues. The focused items should be expanded into hypotheses using a fishbone diagram, and these hypotheses should be verified using a scatter diagram or small batch trials. Once effective countermeasures are identified and solidified, use a control chart to monitor the sustained effect. The output of one link is the input for the next, ensuring the chain remains unbroken and the tools truly work.

4. Self-Check List to Avoid Pitfalls: Making the Seven QC Tools Truly Effective

The hallmark of using tools correctly is not the beauty of the charts but the robustness of the conclusions. After each analysis, use the following checklist to self-assess:

1. Is the data homogeneous? (Has it been stratified by key dimensions?)
2. Can the Pareto chart's focus items explain about 80% of the problems? (Is the "other" category too large?)
3. Are all items in the fishbone diagram "causes," not "countermeasures" or "complaints"?
4. Have all causes been verified with data or on-site observations, or are they just assumptions?
5. Is the sample size for the histogram sufficient? Is the conclusion based on the distribution shape or just the mean?
6. Has the correlation in the scatter diagram excluded a third variable? Has causality been verified?
7. Does the control chart use control limits or specification limits? Is the process truly stable?
8. Are the seven tools used in a chain: "collect → stratify → focus → hypothesize → verify," or are they used independently?

If any of these questions cannot be answered, it means the analysis is not thorough enough. Don't rush to write an improvement report. Returning to the SMT workshop at the beginning: after reworking with this method, the team first stratified the data by product model, identifying that defects were concentrated in a particular power module. The fishbone diagram focused on "equipment" and "methods," delving deep into the two main bones to pinpoint the suspect points: the reflow soldering oven temperature curve setting and the solder paste printing parameters. Using batch data for verification, they ultimately confirmed that the oven temperature curve was not updated after a model change. After improvements, the defect rate for this product dropped from 2.8% to 0.9% over three months. The tools are still the same seven, but the difference lies in: thinking clearly about the question you want to answer and using them correctly.

5. One-Sentence Summary

The value of the seven QC tools lies not in the charts but in the chain: stratify first, then focus, and finally verify, with each step being robust and verifiable.

Think clearly about the question you want to answer before choosing a tool.

Knowledge code: 5.2.4

Version: v20260810

Author: Quality Think Tank Quality Think Tank is dedicated to providing systematic professional knowledge, methodologies, and practical tools to quality management practitioners, helping enterprises continuously improve their quality capabilities.