In-Depth Interpretation of the Seven QC Tools · Histogram

Introduction

Relying solely on the average can mislead you twice.

Two teams have an average defect rate of 3%, but one team's rate is consistently between 2.5% and 3.5%, while the other team's rate fluctuates dramatically from 0.5% to 8%—which one can you accept?

The average only reveals one aspect of the data. A histogram, however, is the mirror that reveals the true distribution of the data.

Chapter 1: The Essence of Histograms

1.1 What is a Histogram?

A histogram is a statistical tool that displays the distribution of data using bars. It divides the range of data values into several equal-width intervals (bins) and then counts the frequency of data points in each interval, representing the frequency by the height of the bars.

Core Logic:
  Continuous data → Group by intervals → Count frequency in each group → Draw bars

  By observing the shape of the bars, you can determine:
  Where is the center of the data? (Mean/Median)
  How dispersed is the data? (Range/Standard Deviation)
  What is the distribution shape? (Normal? Skewed? Bimodal?)
  Is the data out of specification limits? (Is it nonconforming?)

1.2 Why the Average is Not Enough

Assume two sets of data, both with an average of 50:

  Data A: 49, 50, 51, 49, 50, 51, 49, 50, 51
    → Average = 50, minimal variation

  Data B: 10, 30, 50, 70, 90, 40, 60, 20, 80
    → Average = 50, but with significant variation

The average values are identical, but the quality levels are vastly different!

The average is just one characteristic of the data. A histogram can tell you more:

• Where is the center of the data? (Mean/Median)
• How dispersed is the data? (Range/Standard Deviation)
• What is the distribution shape? (Normal? Skewed? Bimodal?)
• Is the data out of specification limits? (Is it nonconforming?)

1.3 Three Major Functions of Histograms

1.4 Histogram vs. Bar Chart

Many people confuse histograms with bar charts, but they are different:

Chapter 2: Drawing and Interpreting Histograms

2.1 Seven-Step Method for Drawing a Histogram

Step 1: Collect Data
  Collect at least 50 data points, preferably more than 100

Step 2: Determine Data Range
  Maximum value - Minimum value = Range (R)

Step 3: Determine Number of Bins
  Empirical formula: Number of bins k = √n or k = 1 + 3.3lg(n)
  Common reference: 50 data points → 7 bins, 100 → 10 bins, 200 → 12 bins

Step 4: Determine Bin Width
  Bin width = Range / Number of bins (round to a convenient number for calculation)

Step 5: Determine Bin Boundaries
  Upper and lower limits for each bin, without overlap

Step 6: Count Frequencies
  Count the number of data points in each bin

Step 7: Draw the Histogram
  X-axis = Numerical intervals, Y-axis = Frequency
  Draw the bars and label the specification limits (if applicable)

2.2 Typical Shapes of Histograms

① Normal Distribution (Bell-shaped)
Frequency ^
 7 |          █          \n
 6 |          █          \n
 5 |       █  █  █       \n
 4 |       █  █  █       \n
 3 |    █  █  █  █  █    \n
 2 | █  █  █  █  █  █  █ \n
 1 | █  █  █  █  █  █  █ \n
   +—————————————————————→ Numerical intervals
  Process is normal, most common

② Skewed Distribution (Left-skewed/Right-skewed)

  Right-skewed (Skewed to the right)
Frequency ^
 6 | █                   \n
 5 | █  █                \n
 4 | █  █  █             \n
 3 | █  █  █  █          \n
 2 | █  █  █  █  █       \n
 1 | █  █  █  █  █  █  █ \n
   +—————————————————————→ Numerical intervals

  Left-skewed (Skewed to the left)
Frequency ^
 6 |                   █ \n
 5 |                █  █ \n
 4 |             █  █  █ \n
 3 |          █  █  █  █ \n
 2 |       █  █  █  █  █ \n
 1 | █  █  █  █  █  █  █ \n
   +—————————————————————→ Numerical intervals

③ Bimodal Distribution
Frequency ^
 6 |    █           █    \n
 5 |    █           █    \n
 4 |    █           █    \n
 3 |    █  █     █  █    \n
 2 | █  █  █  █  █  █  █ \n
 1 | █  █  █  █  █  █  █ \n
   +—————————————————————→ Numerical intervals
  Two different distributions mixed together
  Cause: Different equipment, different shifts, different materials

④ Flat-top Distribution
Frequency ^
 5 | █  █  █  █  █  █  █ \n
 4 | █  █  █  █  █  █  █ \n
 3 | █  █  █  █  █  █  █ \n
 2 | █  █  █  █  █  █  █ \n
 1 | █  █  █  █  █  █  █ \n
   +—————————————————————→ Numerical intervals
  Multiple sets of data with different means mixed together

⑤ Isolated Island Distribution
Frequency ^
 7 |          █                \n
 6 |       █  █  █             \n
 5 |       █  █  █             \n
 4 |    █  █  █  █  █        █ \n
 3 |    █  █  █  █  █        █ \n
 2 | █  █  █  █  █  █  █     █ \n
 1 | █  █  █  █  █  █  █  █  █ \n
   +———————————————————————————→ Numerical intervals
  A few data points deviate from the main distribution
  Cause: Brief anomalies, measurement errors

⑥ Sawtooth Distribution
Frequency ^
 6 |             █             \n
 5 |       █     █     █       \n
 4 | █     █     █     █     █ \n
 3 | █     █  █  █  █  █     █ \n
 2 | █  █  █  █  █  █  █  █  █ \n
 1 | █  █  █  █  █  █  █  █  █ \n
   +———————————————————————————→ Numerical intervals
  Improper bin width setting or data rounding issues

Chapter 3: Practical Cases of Histograms

Case 1: Manufacturing — Diameter Distribution of Shafts

Background: A machining workshop measures the diameter of shafts, with a specification requirement of φ50±0.1mm
100 data points were collected

Histogram Analysis:
  → Normal distribution, center at 50.02mm (slightly above the upper limit)
  → All data within the specification limits (49.90-50.10mm)
  → Distribution width is about 60% of the specification width

Conclusion:
  ✓ Sufficient process capability (CPK≈1.2)
  ✓ But the center is slightly above the upper limit, monitor tool wear trends

Recommendation:
  Set the target value to 50.00mm, allowing for tool wear

Case 2: Bimodal Distribution — Mixed Data from Different Equipment

Background: A plastic injection molding workshop collects product weight data
The histogram shows a clear bimodal distribution

Analysis:
  After stratifying by equipment:
  → Equipment A: Average weight 102g
  → Equipment B: Average weight 98g
  → Significant difference in average values between the two pieces of equipment

Root Cause:
  Different temperature settings on the two pieces of equipment

Action:
  Standardize the process parameters for both pieces of equipment
  After adjustment, the histogram returns to a normal distribution

Case 3: Skewed Distribution — Service Industry Waiting Time

Background: Data on customer waiting times at a bank counter

The histogram shows a clear right-skewed distribution (long tail on the right)

Analysis:
  → Most customers wait 3-8 minutes
  → A few customers wait more than 20 minutes
  → This is a typical "long-tail distribution" in the service industry

Action:
  ── Set a 15-minute over-warning
  ── Increase counters during peak hours
  ── Conduct a special analysis for customers with excessively long waiting times

Case 4: Isolated Island Distribution — Identifying Measurement Errors

Background: An electronics factory measures resistance values

The main distribution of the histogram is normal between 100-105Ω
But there is an isolated island at 95Ω (3 data points)

Investigation reveals:
  → These 3 data points come from a new inspector
  → They used an uncalibrated measuring tool

Action:
  ── Recalibrate the measuring tool
  ── Retrain the inspector
  ── Recall the measurement data for the batch and re-measure

Chapter 4: Common Misunderstandings of Histograms

Misunderstanding 1: Insufficient Sample Size

× Incorrect Approach:
  Only collect 20 data points and draw a histogram
  → The bars are uneven, and the true distribution is unclear

✓ Correct Approach:
  Collect at least 50 data points, preferably more than 100
  The more samples, the clearer the distribution shape

Misunderstanding 2: Improper Number of Bins

× Too Few Bins:
  Only divide into 3-4 bins, losing a lot of information
  Cannot see distribution details

× Too Many Bins:
  Divide into 20+ bins, too many bars, and a messy distribution
  Cannot see the overall trend

✓ Correct Approach:
  Determine the number of bins using the formula k = √n or empirical methods
  50 data points → 7 bins, 100 → 10 bins

Misunderstanding 3: Not Comparing with Specification Limits

× Incorrect Approach:
  Only draw the histogram, without the upper and lower specification limits
  → Cannot determine if the data is within specification

✓ Correct Approach:
  Label the upper and lower specification limits (USL/LSL) on the histogram
  Intuitively judge if the process capability is sufficient

Misunderstanding 4: Not Investigating Abnormal Shapes

× Incorrect Approach:
  See bimodal, skewed, or isolated island shapes and do nothing
  → Miss the opportunity to identify issues

✓ Correct Approach:
  Abnormal shapes are clues!
  Bimodal → Mixed data from different sources → Stratified analysis
  Isolated island → Special cause → Investigate root cause
  Skewed → Unstable process → Parameter adjustment

Misunderstanding 5: Only Looking at Shape, Not Data Volume

× Incorrect Approach:
  30 data points, draw a beautiful "normal distribution"
  → Assume the process is good

✓ Correct Approach:
  Small samples may accidentally show any shape
  The larger the sample size, the more reliable the conclusion
  Use a control chart first to determine if the process is in control

Chapter 5: Combining Histograms with Other Tools

5.1 Histogram + Control Chart

Best Partner:

  Control Chart → Determine if the process is in control (time dimension)
  Histogram → Determine the distribution shape and position (static dimension)

Combined Use:
  ① First, use a control chart to determine if the process is in control
  ② After the process is in control, use a histogram to analyze the distribution shape
  ③ Compare with specification limits to evaluate process capability
  ④ Use a control chart for continuous monitoring

5.2 Histogram + Stratification Method

Combined Use:

  Step 1: Draw an overall histogram to identify abnormal shapes (e.g., bimodal)
  Step 2: Use the stratification method to stratify by equipment/shift/material
  Step 3: Draw a histogram for each stratum
  Step 4: Identify the source of differences and implement targeted improvements

  → The histogram identifies "problems," and the stratification method finds "where"

5.3 Histogram + Normality Test

Two ways to determine if data follows a normal distribution:

  Qualitative: Check if the histogram is bell-shaped
  Quantitative: Perform a normality test (e.g., Anderson-Darling test)

Why is it important:
  → If the data does not follow a normal distribution
  → The formulas for calculating CPK and CP may not be applicable
  → Data transformation or non-parametric methods may be needed

5.4 Histogram + Specification Limits → Process Capability Index

Prerequisites for calculating the process capability index (CP/CPK):
  ① The process is in control (confirmed by control chart)
  ② The data follows a normal distribution (confirmed by histogram)

CP = (USL - LSL) / 6σ
CPK = min(CPL, CPU)
  where CPL = (μ - LSL) / 3σ
       CPU = (USL - μ) / 3σ

Judgment Criteria:
  CPK ≥ 1.33 → Sufficient process capability
  1.0 ≤ CPK < 1.33 → Adequate process capability, but requires attention
  CPK < 1.0 → Insufficient process capability, requires improvement

Chapter 6: Advanced Uses of Histograms

6.1 Histogram + Specification Limits (Process Capability Visualization)

Best practice histograms include:
  ── Bars: Data distribution
  ── Two vertical lines: USL (upper specification limit) and LSL (lower specification limit)
  ── One vertical line: Target value (Target)
  ── Labels: Mean, standard deviation, CPK value

  → One chart can determine if the process meets requirements

6.2 Comparison Histograms Before and After Improvement

Histogram Before Improvement:
  ── Distribution center deviates from the target
  ── Large distribution width
  ── Data points out of specification limits

Histogram After Improvement:
  ── Distribution center aligned with the target
  ── Smaller distribution width
  ── All data points within specification limits

  → Two charts side by side, intuitively showing the improvement effect

6.3 Grouped Histograms

Display histograms for different groups on the same chart using different colors
(Use semi-transparent fills to see overlapping areas)

Value:
  ── Intuitively compare distribution differences between groups
  ── Quickly identify which group needs improvement
  ── Avoid misjudgment of bimodal distributions

Chapter 7: Evaluation Standards for Histograms

Summary: The "Way" and "Method" of Histograms

Method (How to draw):
  ── Collect more than 100 data points
  ── Determine the number of bins and bin width
  ── Draw the bars and label the specification limits
  ── Observe the distribution shape

Way (Why to draw):
  ── Not to draw a "beautiful distribution chart"
  ── To "see the truth behind the average"
  ── To "identify issues in the data and find improvement directions in the distribution"

The greatest value of a histogram is not "describing data," but "exposing the lies of the average."

A manager who focuses only on the average may be leading a team that is producing in an extremely unstable process—while he believes everything is normal.

A histogram is the tool that helps him see the truth.

Knowledge code: 5.2.4 Document Version: v1.0
Generated Date: 2026-05-03
Author: Quality Think Tank

Issue 6: Histogram (Histogram)