Tools For Quality Improvement
Tools For Quality Improvement
Tools For Quality Improvement
QUALITY
IMPROVEMENT
The School of Quality
TOOLS FOR
QUALITY
IMPROVEMENT
TOOLS FOR QUALITY IMPROVEMENT
T
eams of people working together on projects in order to achieve pre-defined quality objectives or
eliminate the root cause(s) of a specific problem are an integral part of ST culture. Our emphasis is to
ensure that all these team members are sufficiently equipped to undertake their projects. This
requires basic understanding of a methodology which provides a framework to guide them through
their project and working knowledge of the more common data collection, analysis and interpretation tools.
This self-training guide is a collection of such tools. You will find that regardless of the methodology or process
your team uses to carry out a project, the tools discussed here will allow you to analyze and interpret data in
almost all cases.
The material presented in this self-training guide is by no means comprehensive. It is intended to enhance and
complement your current knowledge of the tools so that you and your team can use them to effectively tackle
your projects.
If you have any comments, suggestions or feedback concerning this self-training guide, please feel free to contact
Murali at [email protected]
INDEX
1. Flow diagram 4
2. Brainstorming 9
3. Cause-effect diagram 12
4. Data collection 16
5. Graphs and charts 19
6. Pareto analysis 24
7. Scatter diagram 29
8. Histogram 34
9. Box plot 40
10. Stratification 46
Summary 51
Self-test 53
Recommended solutions 55
1. FLOW DIAGRAM
CONCEPTS
A flow diagram is a graphic representation of the sequence of steps that we perform to produce an output. The
output may be a physical product, a service, information, or a combination of the three. A flow diagram is used
at the beginning of a project to determine the project scope and boundaries; during the project to provide a
common understanding of the process under investigation; and towards the project end, when it could be used
to make comparisons between current (old) and improved (new) processes. The common symbols used in a flow
diagram are:
1. The activity symbol is a simple rectangle box, which gives a brief description (1-2 words) of an
an activity in the process.
2. The decision symbol is a diamond shaped box used to denote any point in the process where a
decision, such as yes/no, up/down or on/off, must be made, leading to different process
branches.
4
3. The terminal symbol is a rather flat rectangle with rounded corners, used to denote the
beginning and end of a process or sub-process, often using the words start, begin, or end.
4. Flow lines have an arrow at the end, indicating the process flow to the next activity or step. Besides
the usual downward flow, flow lines can indicate a process loop (going back to an activity) or a
process branch (going to a different set of activities or a sub-process), based on the decision made.
5. The document symbol has the shape of a printout or paper and is used when the activity
described is a printed form.
6. A database symbol is a cylinder and is used when the activity involves access to or retrieval from
a database.
7. The wait symbol is a trapezoid, providing a means to show where normal processing is halted or
suspended to be continued later on.
8. The connector or continuity symbol is a circle with a letter or number in it, used to denote a
break point in the process (most frequently used when the flow diagram reaches the edge of the
paper). An identical symbol is then used to show where the diagram continues.
1. FLOW DIAGRAM
EXAMPLE
The example below illustrates how flow diagrams can prove useful in both in the diagnostic and remedial phases
of a project. The left hand side of Figure 1 below shows the process that airlines used a few years ago to issue
tickets and boarding passes to passengers without ticket. An examination of the diagram shows that passengers
must stand in two lines, first to purchase the ticket and then at the gate to board. Giving the ticketing agents
access to the seating database is an obvious process improvement that is sure to make passengers happy by
eliminating the need to stand in one of the lines. This improvement is shown on the right side of the figure.
Figure 1. Flow diagram of airline process for issuing tickets and boarding passes
5
1. FLOW DIAGRAM
CONSTRUCTION
1. Define how the flow diagram will be used in your project.
2. Decide on desired outcome of the flow diagram: high level, matrix or detailed? Construct a high level
diagram before constructing a detailed diagram.
A high level flow diagram groups all related activities and decisions in blocks or sub-processes, so that the
final diagram is a collection of sub-processes linked together from start to finish.
A matrix (or cross-functional) flow diagram shows the process flow both sequentially and by function. The
diagram is divided into rows or columns, each representing a department, function or other sub-unit of the
system under study. Each symbol of the flow diagram is placed in the row or column of the unit responsible
for executing it.
6 4. Document each step in the sequence, starting from the first (or last) step. Draw the process accurately and
consistently from the top left to the bottom right edge of the page.
5. When a decision symbol is encountered, select one branch and continue the flow diagram of that branch.
6. When you run into an unfamiliar activity of a sub-process, make a note and continue the flow diagram.
7. Repeat steps 4, 5 and 6 until the last (or first) activity is reached.
8. Go back and flow diagram the other branches of the decision symbols.
9. Review the diagram for any missed decision points or special cases that might cause some work to follow a
different process.
10. Discuss how to fill in the unfamiliar steps or areas of the flow diagram, and verify its accuracy (observe the
process directly, interview experts, show the diagram to line workers, and so on).
AVOIDING PITFALLS
When analayzing a problem failure, be sure to document the actual process, and not how the process should
be.
Don’t forget to update the flow diagram as new information is uncovered or as the process is modified.
1. FLOW DIAGRAM
QUESTIONS
1. When should you use a flow diagram?
4. How do you proceed when you encounter a branch point when drawing the flow diagram of your process?
8
5. How can a flow diagram help you better understand a process?
6. What are the common pitfalls to avoid when using a flow diagram as a tool?
HOW TO BRAINSTORM
1. Phrase the statement properly. Focus on the issue but be broad enough for creativity.
2. Prepare.
Communicate the subject to participants ahead of time
Determine the ideal number of participants 9
Provide an appropriate medium to list all contributions
If the subject is project selection, collect data on the severity of problem and contributions to it. If the subject is
theories, go to the cause-effect diagram. If the subject is remedies, evaluate ideas in context of time, costs,
probability of success, and so on. If the subject is resistance or culture, develop strategies for the serious ones.
2. BRAINSTORMING
INTERPRETATION
Brainstorming is a proven way of generating many ideas on a topic and increasing the creativity of the
participants. The resulting lists of ideas usually contain more new and innovative ideas than lists obtained in
other ways.
The next step in analyzing the ideas of a brainstorming is to narrow down the list to the vital few, which the
team could comfortably focus on. This narrowing or prioritizing of ideas can take the form of voting, a Pareto
analysis or consensus.
AVOIDING PITFALLS
While brainstorming is a great way to generate lists of ideas, it should not be used as a substitute for data.
10 The most common barriers to the flow of ideas and creativity are: the inability of the group to suspend
judgment and analysis until the list of ideas has been completed and the dominance of one or a few
individuals in presenting ideas.
2. BRAINSTORMING
QUESTIONS
1. What is the first step in a proper brainstorming session?
2. Name at least three conceptual and three practical rules to observe during a brainstorming session.
3. How are the large number of ideas generated in a successful brainstorming session handled?
4. Is there a need to collect data to support ideas and theories generated from brainstorming session?
Why/why not?
11
5. How can a team avoid the dominance of one or a few individuals during a session?
The most important consideration in the construction of a cause-effect diagram is the clear understanding of the
cause-effect relationship. All possible sources of causes need to be considered. In addition to the four W’s
12 (What, When, Where, Who), there are two other lists that can help you remember to consider several classes of
possible causes for a problem. These lists are characterized as the five M’s (Man, Machine, Method, Material,
Measurement) in manufacturing and the five P’s (People, Process, Place, Provision, Patron) in services.
EXAMPLE
Figure 2.Cause-effect diagram for lost control of car
3. CAUSE-EFFECT DIAGRAM
In Figure 2, the phenomena to be explained is "lost control of car." Some of the possible factors contributing to
lost control are a flat tire, a slippery road, mechanical failure, and driver error. Each of these major categories of
causes may, in turn, have multiple causes. A flat tire may come from a nail, a rock, glass, or a blow-out from
material failure. The causal relationship can be traced back still more steps in the causal chain if necessary or
appropriate. For example, loss of control may arise from a mechanical failure. This failure may be a brake
failure, which, in turn, may come either from fluid loss or from worn brake pads, and so on.
CONSTRUCTION
1. Define the effect or symptom for which the causes must be identified.
2. Place the effect or symptom being explained at the right, enclosed in a box. Draw the central spine as a thick
line pointing to it.
7. Continue adding possible causes until each branch reaches a root cause.
When interpreting a cause-effect diagram, your primary task is to check for completeness. To do so, consider the
following points:
Be certain you have at least asked how each of the four W’s and each the five M’s or five P’s might apply to
the effect.
Generally, each main branch of the diagram will have at least three or four additional branches.
If some main branches have substantially fewer causes effected to them, or if the causes on them do not go
back as many steps in the causal chain, you may not have as full an understanding of that element of the
process.
14 Verify that the cause at the end of each causal chain is potentially a root cause by checking if you can trace a
logical causal relationship from that cause, through all its intermediate causes, to the final effect being
explained. That cause is, in principle, directly controllable. Therefore, if shown to be true, that cause could be
eliminated, and the effect would disappear or be reduced.
AVOIDING PITFALLS
Be careful not to confuse the orderly arrangement of theories found in a cause-effect (C-E) diagram with the
real data obtained through empirical testing.
Be sure to test each causal relation in the cause-effect diagram for logical consistency to ensure its usefulness.
Construct the diagram only once the symptoms have been analyzed as thoroughly as existing information will
permit, which will help avoid a large, complex diagram that can be difficult to use.
Do not limit the theories that are proposed and considered or you may unintentionally mask or ignore the
ultimate root cause(s).
3. CAUSE-EFFECT DIAGRAM
QUESTIONS
1. What is the primary purpose of a cause-effect diagram?
2. In a typical cause-effect diagram, what description is placed in the box on the extreme right?
3. Which common tool could a team use to generate inputs for their cause-effect diagram?
4. Name the two checking activities to do immediately after completing the diagram.
15
5. What needs to be done after analyzing a cause-effect diagram?
6. What is the danger of limiting the input of ideas when drawing the diagram?
Information generation begins and ends with questions. To generate information, we need to:
formulate precisely the question we are trying to answer
collect the data and facts relating to that question
analyze the data to determine the factual answer to the question
present the data in a way that clearly communicates the answer to the question.
In order to generate useful information, planning for good data collection proceeds along the following lines:
What question do we need to answer?
How will we recognize and communicate the answers to the question?
What data analysis tools (Pareto diagram, histogram, bar, graph, etc...) do we envision using, and how will we
communicate the results?
16 What type of data do we need in order to construct this tool and answer the question?
Where in the process can we get this data?
Who in the process can give us this data?
How can we collect this data from these people with minimum effort and chance of error?
What additional information do we need to capture for future analysis, and traceability?
2. Consider appropriate data-analysis tools and ensure the data needed for the analysis are being collected.
4. Select unbiased collectors; understand the environment in which the data is being collected.
5. Design data collection forms using the KISS motto: Keep It Short and Simple.
Reduce opportunities for errors.
Capture data for analysis, reference and traceability.
The form should be self explanatory.
Incorporate a professional look.
INTERPRETATION
Data should be used to answer the questions for which it was collected using the tools identified while planning
for the collection. The following are types of biases that may occur and which you should be aware of. Some
can be avoided with careful planning, while others require attention during analysis.
Exclusion bias: Some part of the scope of the process being investigated has been left out.
Interaction bias: The process of collecting the data itself may affect the process being studied.
Perception bias: The attitudes and beliefs of the data collectors can sometimes influence what they see and
how they record it. 17
Operational bias: Failure to follow the established procedures is the most common operational bias.
Non-response bias: Missing data can bias the result. The fact that they are missing is a clue that they are
different from the rest in some way.
Estimation bias: The formula and methods used to calculate statistics from the collected data may give certain
types of biases.
AVOIDING PITFALLS
Avoid using data collection when creativity is important: brainstorming theories, possible remedies, or cultural
issues. Before collecting new data, ensure that the answer cannot be found already using existing data.
4. DATA COLLECTION
QUESTIONS
1. Why does data collection constitute an integral part of the problem solving process?
2. What are five key considerations when planning to collect data for your project?
3. What are the major considerations when designing a data collection form?
4. A common bias to avoid in data collection is the non-response bias. What is this?
18
5. Discuss the merits and disadvantages of using existing data as opposed to collecting new data.
On line and bar graphs, the horizontal axis portrays the independent variable. The vertical axis portrays the
dependent variable. The independent is the one that is believed to cause or determine the dependent one. On
a single line graph, there is only one point for each independent variable value. There can be multiple points
with the same dependent variable value.
EXAMPLES 19
Bar graphs
1. Determine the range of the vertical axis and the size of each increment. Label the axis, including the unit of
measure.
2. Select an appropriate bar graph - simple, grouped or stacked.
3. Determine the number of bars. Draw and label the horizontal axis.
4. Determine the order of the bars; draw them on the graph.
5. Label and title the graph.
Pie charts 21
1. Determine the percentage of each category
2. Convert the percentage values into degrees of angle.
3. Draw a circle and carve out the segments of the pie chart.
4. Label the segments and title the chart.
INTERPRETATION
Keep in mind these guidelines when constructing graphs:
Graphic integrity: Construct your graph so that it represents your actual data. Use data labels to reinforce
graphic information and ensure that the true information is conveyed.
Consistent scale: Numeric scales must maintain regular intervals. Different graphs that might be compared to
each other should all be drawn to the same scale.
Ease of reading: If a graph is easy to read, it will be better understood and remembered. Labels, placed
closely to the object they identify, improve clarity.
Consistency of symbols: When two or more graphs are to be compared, use the same symbols to minimize
confusion in interpreting the graphs.
Simplicity: Be sure not to obscure information with unnecessary lines, colors, 3-D, shading and other
decoration. Text and decoration should add value or information.
5. GRAPHS AND CHARTS
AVOIDING PITFALLS
Be sure your graph does not distort reality by choosing the appropriate scale.
Use consistent scales from one chart to the next when they are to be compared.
22
5. GRAPHS AND CHARTS
QUESTIONS
1. When is a line graph best used? A bar graph? A pie chart?
2. In a line graph, what are the two principal axes called? Which one is placed horizontally and which one is
placed vertically?
3. What is the difference between a grouped bar and a stacked bar presentation?
23
4. What are the key concepts to bear in mind when preparing a graphic chart?
5. How does varying the scale on one axis affect the presentation of a line graph?
In the early 1950s, Dr. Joseph Juran noted a universal phenomenon that has come to be called the Pareto
principle. The Pareto principle states that in any group of actors contributing to a common effect, only a relative
few account for the bulk of the effect. Juran has also coined the terms "vital few" to refer to the small number of
contributions which account for the bulk of the effect, and "useful many" to refer to all the other factors which
account for a smaller proportion of the effect.
Pareto diagrams and tables are presentation techniques used to show the facts and separate the vital few from
the useful many. They are widely used to help quality improvement teams and quality councils make key
decisions at various points in the quality improvement or problem-solving sequence.
One key application of Pareto charts is to make comparisons between "before" and "after" charts to show the
effect or impact of remedial or improvement actions in the "vital few" areas identified.
24
Well-constructed Pareto diagrams and tables include three basic elements:
Contributors to the total effect, ranked by the magnitude of their contribution.
Magnitude of the contribution of each factor expressed numerically.
Cumulative-percent-of-total effect of the ranked contributors.
EXAMPLE
A quality improvement team was formed to improve the format of forms coming from field sales offices to the
home office. There were 18 items on the order form, designated A-R. The team developed a check sheet which
it used to collect the frequency of errors on the forms for one week. The results of the study, in the form of a
Pareto table is shown in Figure 8.
The team then went on to plot the Pareto diagram shown in Figure 9. The implications of the analysis is
profound. If the team can resolve the causes creating errors for the four vital few information items, they can
significantly improve the quality of the order forms coming from sales offices.
This is an important point: without the facts and without a Pareto analysis, the team would have been faced
with the much larger and more costly task of trying to find ways to prevent errors from occurring on all 18 items.
Thus, we can clearly see from a Pareto table or chart that significant improvement can be achieved with a
smaller, but more precisely focused effort.
6. PARETO ANALYSIS
Figure 8. Pareto table of errors on order forms
25
4. Draw and label the left vertical axis (label the axis from 0 to the grand total or just beyond).
6. Draw and label the right vertical axis (label the axis from 0 to 100%). Line up the 100% mark with the grand
26 total on the left axis and label the axis (for example, "Cumulative Percent of Total").
7. Draw bars to represent the magnitude of each contributor's effect. The height of each bar corresponds to the
magnitude of the contribution as measured on the left axis.
8. Draw a line graph to represent the cumulative percent of total. The plotted points correspond to the
cumulative percent as measured on the right axis. As a check, the cumulative percent of total for the first
ranked contributor should be even with the height of the first bar.
Determining the break point is not an exact science. In practice, a quality improvement team faced with
interpreting a Pareto diagram that does not show a clear break point usually takes the following approach:
Identify those few contributors which account for about 60% of the quality effect.
Call these the “vital few”, and begin the diagnostic journey.
When the diagnostic and remedial journeys are completed for these vital few, repeat the Pareto analysis.
The contributors that were in the awkward zone - not clearly part of the vital few or the useful many - may
now be among the vital few.
AVOIDING PITFALLS
The most common mistake with Pareto analysis is failure to use the technique. This results in unfocused, low-
payoff quality improvement efforts.
Pareto analysis is most powerful when we use objective data and facts, rather than opinions, to rank the order
of the categories.
Sometimes the data indicate no clear distinction between the categories. In this situation, all the bars on the
Pareto diagram are roughly the same height or it takes more than half of the categories to account for more
than 60% of the quality effect.
6. PARETO ANALYSIS
QUESTIONS
1. How is a Pareto chart different from a normal bar chart?
4. How can the Pareto tool be used to illustrate improvements in a targeted area?
28
5. After constructing a Pareto diagram, what steps can a team take if no obvious distinctions between the various
categories appear?
EXAMPLE
Managers in a large corporation were complaining about the lack of administrative support services received
from a central pool offering special on-call administrative services. The team chartered to investigate this started
with the hypothesis that the managers' expectations were being met but that they were unaware of it since the
time was broken up into a few hours one day and a few hours the next. 29
To investigate the validity of the complaint, the team surveyed the managers and asked them to specify the
number of hours a week of support they expected. The team then reviewed time records to see how much
support each manager actually received. The scattered diagram of this paired data is shown below:
2. Decide which variable will be plotted on the horizontal axis (if showing a cause-effect relationship, put the
suspected cause variable on the horizontal axis).
3. Draw and label the horizontal and vertical axes. For each label, make the lowest label slightly lower than the
minimum value and the highest label slightly higher than the maximum value.
4. Plot the paired data. Use concentric circles to indicate identical paired-data points.
30 INTERPRETATION
Analyzing scatter diagrams is a 4-step process.
1. Develop a plausible and relevant theory about the suspected relationship between 2 variables of interest.
2. Collect appropriate paired data and construct a scatter diagram.
3. Identify and classify the pattern of correlation.
4. Question your original theory and consider the other explanations for the observed pattern of correlation.
Strong negative correlation: The value of Y clearly decreases as the value of X increases.
Because the points are so conspicuously grouped around an imaginary line (the trend line) through the
middle of the mass of the points, we are very confident of the relationship between X and Y. This suggests
that the system under study is one where control of one variable results in control of the other.
Weak, positive correlation: The value of Y increases somewhat as the value of X increases.
Weak negative correlation: The value of Y decreases somewhat as the value of X increases.
Complex correlation: The value of Y appears to be related to X, but the relationship is not simple or smooth.
The relationship between X and Y requires examination.
No correlation: For any value of X, Y can have both large and small values. There does not appear to be any
particular relationship between X and Y. Look for other variables that influence Y.
7. SCATTER DIAGRAM
Figure 11a. Strong positive correlation Figure 11b. Strong negative correlation
31
Range of operation
Be equally wary of the value of gathering and analyzing data beyond the range of the operation of the process.
Effect of scale
Proper interpretation of a scatter diagram depends on careful attention to construction details. Generally, you
should draw the axes of scatter diagrams such that (1) the maximum and minimum values on each axis
correspond closely to the maximum and minimum values of each variable, and (2) the axes are roughly the
same length.
Numerical summary
A number of computer programs print numerical summaries, such as the correlation coefficient and equations of
32 regression lines, without providing a graphic picture of the data. Numerical summaries, by themselves, can be
very misleading.
Confounding factors
With a scatter diagram, we study one cause-effect pair among many. It is therefore possible that the correlation
we observe in our scatter diagram is due to a cause other than the one we are plotting. This other unseen cause
is sometimes called a confounding factor.
Absence of stratification
Stratification can make a correlation pattern appear, or it can completely change our interpretation of an
apparent correlation.
7. SCATTER DIAGRAM
QUESTIONS
1. What is the primary purpose of a scatter diagram?
2. The variables involved in a scatter diagram are sometimes called the (suspected) cause variable and the
(observed) effect variable? Which is to be plotted on the horizontal axis? Which is to be plotted on the
vertical axis?
3. Explain the following patterns: a strong positive relationship, a weak negative relationship, no relationship.
4. What is the danger of interpreting a diagram beyond the range of the data from which it was constructed?
33
5. Scatter diagrams are effective to the extent that they prove or disprove a theory or physical explanation.
Where does this theory originate?
Concept 4: Patterns of variation are easier to see when the data are summarized pictorially in a histogram.
EXAMPLE
34
This example illustrates how a histogram could be used to interpret data to verify the existence of a problem. A
manufacturer of electronic telecommunications equipment was receiving complaints from the field about low
volume sound on long distance connections. Bettina in France couldn't hear her mother in Germany.
A string of amplifiers manufactured by the company was used to boost the signal at various points along the way
in these long connections. The boosting ability of the amplifiers (or, the "gain") was therefore a prime suspect in
the case.
The design of the product has called for a gain of 10dB. This means that the output from each amplifier should
be about 10 times the input signal. This amplification makes up for the natural fading of the signal over long
distances. Recognizing that it is impossible to make every amplifier with a gain of exactly 10dB, the design
allowed the amplifiers to be accepted if the gain fell between 7.75dB and 12.25dB. Since there are literally
hundreds of amplifiers boosting the signal along the way, low gain amplifiers would be compensated by high
gain ones to give an acceptable volume.
The QI team investigating the problem arranged to measure the gain of 120 randomly picked amplifiers. The
measurements made are shown in Figure 12. The data in this form is not easy to analyze; there are 120
numbers to examine. More importantly, since all the measurements fell within the 7.75-12.25dB range, it is
tempting to conclude that the data was of little value.
The team decided to construct a histogram to obtain a different perspective (see Figure 13). It becomes
immediately obvious that while all the amplifiers fell within the range set, they were certainly not evenly
distributed around 10dB. Most of the amplifiers had a lower than nominal value of 10dB. If most of the
amplifiers in a long distance connection were boosting their signals a little bit less than expected, the net result
would be a low volume level. In other words, Bettina won't be able to hear her mother in Germany.
8. HISTOGRAM
Figure 12. Gain of 120 tested amplifiers
35
Figure 13. Histogram with results of amplifier gain testing
8. HISTOGRAM
CONSTRUCTION
1. From the table of raw data, determine the range, high and low values.
(Range = high value - low value).
Note: Consider using a box plot if the total number of data points is less than 40, unless this is a result of data
stratification.
4. Round the cell width to a convenient number: make the cell width 1, 2, or 5; 0.1, 0.2 or 0.5; 10, 20, 50 etc.
36 5. Construct the cells by listing the cell boundaries. The first cell should include the lowest value. Cell
boundaries should be one more significant digit than the data.
6. Tally the number of data points in each cell. Check that the total tally equals the number of data points.
7. Draw and label the horizontal axis. The range on this axis should include one cell width beyond the lowest
and highest value.
8. Draw and label the vertical axis. Label it from 0 to a multiple of 5 above the highest value.
9. Draw in the bars to represent the number of data points in each cell. The height of the bars should equal the
number of data points in that cell as measured on the vertical axis.
10. Title the chart (include total number of data points). Show nominal values and limits if applicable.
8. HISTOGRAM
INTERPRETATION
We know that the values in any set of data will vary and that variation will display some pattern. The goal of our
analysis of a histogram is to:
identify and classify the pattern of variation
develop a plausible and relevant explanation for the pattern.
The explanation must be based on the team's knowledge and observation of the specific situation, and it must
be confirmed through additional analysis.
The Double-Peaked Distribution is usually a combination of two bell-shaped distributions and suggests that
two distinct processes are at work.
The Plateau Distribution is likely to be the result of many different bell-shaped distributions with centers
spread evenly throughout the range of the data. 37
The Comb Distribution typically indicates data errors.
The Skewed Distribution typically occurs when a practical limit, or a specification limit, exists on one side and
is relatively close to the nominal value. Such skewed distributions are not inherently bad, but a team should
question the impact of the values in the long tail.
The Truncated Distribution is often a smooth, bell-shaped distribution with a part of the distribution removed,
or truncated, by some external force, such as screening, 100% inspection, or a review process. These
truncation efforts are an added cost and are, therefore, good candidates for removal.
The Isolated-Peak Distribution suggests that two distinct processes are at work. But the small size of the
second peak indicates an abnormality that doesn't happen often or regularly.
The Edge-Peaked Distribution occurs when the extended tails of the smooth distribution has been cut off and
lumped into a single category at the edge of the range of the data. This shape very frequently indicates
inaccurate recording of the data.
AVOIDING PITFALLS
Be careful not to use histograms in an inappropriate context.
On the other hand, be sure to recognize and use this tool in appropriate situations.
8. HISTOGRAM
Figure 14a. Bell shaped histogram Figure 14b. Double-peaked histogram
38
Figure 14g. Isolated peaked histogram Figure 14h. Edge peaked histogram
8. HISTOGRAM
QUESTIONS
1. How is a histogram different from an ordinary bar chart?
3. How many bars or categories would be appropriate in a data set of 75 values? 150 values? 225 values?
4. What could be a plausible physical explanation for a double peaked pattern in a histogram? For a skewed
pattern?
39
5. What is the recommended minimum number of values needed to constructed a histogram? Why?
6 What alternative tool is recommended if there is insufficient data values to construct the histogram?
The basic box plot is a five-number summary of a set of data. These five numbers are shown in Figure 15 and
are defined as follows:
Lowest value: the smallest value in the set of data. Figure 15. Box plot
First quartile: the point below which 25% of the data lie.
Median: the middle value in the set of data.
Third quartile: the point below which 75% of the data lie.
Highest value: the largest value in the set of data.
The box on a box plot encloses the central 50% of the observations. Its
40 beginning and end are the first and third quartiles, respectively.
The line dividing the box is the median, which divides the data into two equal
groups. Half of the observations lie below the median, and half lie above it.
The lowest and highest values are at the end of lines, sometimes called
whiskers, extending from the box (for this reason, the box plot is often referred
to as a box and whisker chart).
The key benefit of the tool is that the five simple summary statistics, shown in the form of a box plot, enable a
team to see some of the aspects of the pattern of variation inherent in the data even if the number of data
points is too small to produce a meaningful histogram.
EXAMPLE
The administration manager of a large office received complaints about the availability (or rather, the
unavailability) of copier machines. Investigations revealed that while users felt that the current two machines in
the building were adequate enough, the copiers were frequently out of service, awaiting the arrival of the repair
person. Further investigation revealed that failure rates of the copiers were within reasonable bounds. The
response time of the repair person was the major controllable factor in the availability of the copiers.
The copiers were serviced by two different firms. The QI team formed to investigate this problem decided to
conduct a simple study to determine the response of each to a repair call. The team defined "response time"
9. BOX PLOT
as the time between the call for repair and the arrival of the repair person. They then collected the data by
comparing various records and log books on a random sample of visits by each contractor. Due to the difficulty
of collecting the data, a small sample of 10 visits was collected.
The small samples negated the use of a histogram as an investigative tool in this case. So a box plot was used.
The data was reformatted into an ordered set as follows:
41
The box plot for each contractor was then constructed (see Figure 16c). The team could now draw the following
conclusions:
Figure 16c. Box plot for each contractor
The median response time of B was 190 minutes,
compared to 247 minutes for A.
3. Draw and label the horizontal axis; provide labels to identify each box plot.
4. Draw and label the vertical axis; label the axis and its unit of measure:
begin labeling the axis until you reach a multiple of 1, 2 or 5 that is the smallest value in any box plot.
continue labeling the axis until you reach a multiple of 1, 2 or 5 that is larger than the highest value in any
plot.
5. Draw the box plots. Include a key to define the various parts of the box plot.
6. Title the chart and show nominal values and limits if applicable.
42
Definition of symbols:
d = the depth; the number of observations to count from the beginning of the ordered data set
M = the Median or central value
n = the number of observations in the set of data
Q1 = the first quartile
Q3 = the third quartile
Formulae:
Depth of Median = d(M) = (n+1)/2
Depth of First Quartile = d(Q1) = (n+2)/4
Depth of Third Quartile = d(Q3) = (3n+2)/4
9. BOX PLOT
INTERPRETATION
Values in any set of data will vary, and the variation will display some pattern. Box plots are a good way to
compare patterns of variation among data sets.
Based on this analysis, the team can use its knowledge and further observation to develop theories about the
underlying process. In this respect, interpreting box plots is similar to interpreting histograms.
Because they have fewer data points, box plots limit our ability to classify the pattern of variation. The histogram is
a much better tool; if you have 40 or more data points, use histograms to analyze the pattern of variation. Figures
17a-h on the following page shows each common histogram pattern and the corresponding box plot shape. 43
AVOIDING PITFALLS
Be careful not to use box plots in an inappropriate context.
9. BOX PLOT
Figure 17a. Bell shaped box plot/histogram Figure 17b. Double-peaked box plot/histogram
Figure 17c. Plateau box plot/histogram Figure 17d. Comb box plot/histogram
44
Figure 17e. Skewed box plot/histogram Figure 17f. Truncated box plot/histogram
Figure 17g. Isolated peaked box plot/histogram Figure 17h. Edge peaked box plot/histogram
9. BOX PLOT
QUESTIONS
1. When is a box plot considered a good substitute for the histogram?
3. In a box plot comparing two processes, name at least three features which distinguish the better process.
4. Why is it better to use a histogram rather than a box plot in certain situations?
45
The characteristic used to separate the data is called a stratification variable. Each stratification variable will have
two or more values. A category is defined by specific combinations or ranges of variables and may be defined by
more than one variable.
EXAMPLE
46 A pharmaceutical company learned that one of its previously successful products was gaining a reputation in the
medical community as being ineffective. A review of the original clinical studies and current QA lab data offered
no insight into the degradation in the performance of the drug. From a brainstorming session, a theory was
developed that the active ingredient in the compound could be breaking down more quickly than expected
while the medication sat on the shelf. The team decided to investigate this shelf-life theory further.
Past QA lab data indicated that the production process consistently produced products with 149mg to 151mg of
active ingredient. The specification called for 148mg to 152mg; so the process was on target and fully capable of
producing the product in conformance to specification. Also, past clinical studies showed that a minimum of 120mg
of active ingredient was needed for the drug to be effective in use. Shelf-life studies conducted 10 years ago as part
of the government approval procedure had shown that the active ingredient broke down at a rate of 1mg/month. So,
after the maximum shelf life of 24 months, the drug should still have enough potency to be effective.
Samples of various lots of the drug were pulled from shelves and brought to the lab to determine the amount of
active ingredient in mg. The tests included samples that were older than 24 months that have been returned as
part of the company's routine procedure for handling expired products. The age of the product was determined
from the lot number stamped on each bottle.
The scatter diagram of the data is shown in Figure 18a. As expected, the graph shows a definite relationship
between shelf life and weight of active ingredient. Furthermore, the scatter diagram seems to indicate a non-
linear relationship - the breakdown seemed to proceed slowly at first, and faster later on. The key point, of
course, is that in many cases the weight of the active ingredient had fallen below the 120mg minimum after as
few as 13 months.
The next phase of the diagnostic journey was to determine the root cause of the poor shelf-life characteristic.
Several theories on how the active ingredient may be interacting with some other chemical in the final
10. STRATIFICATION
formulation were tested. By searching through records associated with the various lots of products in the sample,
the team finally solved the case by stratifying the scatter diagram it first developed.
Figure 18b shows the same paired data as before, but this time the data had been stratified to reveal something
hidden before. A filler ingredient in the formulation of the drug came from two suppliers. In the stratified scatter
diagram, lots made with the filler from supplier A are plotted as open dots, while lots made associated with
supplier B are plotted as black dots. Two distinct patterns can now be discerned. The lots produced with filler
from supplier B showed rapid degradation of the active ingredient, while lots associated with supplier A showed
a slower degradation. This combination of degradation rates and differences in the number of data points for A
and B accounts for the non-linear relationship that the team observed in the original scatter diagram. Chemical
analysis of the filler from A and B revealed the presence of a substance in B's formulation that reacted with the
active ingredient to accelerate its breakdown. The remedy involved a change in supplier B's formulation.
47
10. STRATIFICATION
HOW TO STRATIFY
1. Select the stratification variables. If new data are to be collected, include all potential stratification variables.
2. Establish categories that are to be used for each stratification variable. The categories may be either discrete
values or ranges of values.
3. Sort observations into the categories of one of the stratification variables. Each category will have a list of the
observations that belong to it.
4. Calculate the phenomenon being measured for each category. These calculations are usually either a count of
the number of observations in the category or a calculation of an average value for those observations.
6. Prepare and display the results for other stratification variables. Repeat steps 2 through 5. Do second stage
stratification as appropriate.
48
7. Plan for additional confirmation. Usually additional data, other techniques, and/or controlled experiments will
be used to confirm the initial stratification results.
INTERPRETATION
Looking at the results
If the stratification results are in bar-graph form, it is easy to look across the categories of a variable to see
whether one or a few of the categories stand out. Does one supplier have a particularly large defect rate? What
types of medication are most prone to errors?
What to do next?
After stratification, if the results give a clear indication of the likely source of the phenomenon being studied,
then the team will want to validate their initial results or will want to gain further details on the precise cause.
If the initial stratification does not yield useful results, there are two possible courses of action:
First, the team could conduct a two-stage stratification that is, stratify by a second variable within each
category of a first variable.
Another mistake is to jump to the conclusion that the anomalous category is the cause of the problem. The
anomalous category is probably where we should go looking for the cause, but the category itself is not
necessarily the cause. If the team is going to collect new data, they should make an effort to collect as much
identifying information as they think could possibly prove useful for stratification.
49
10. STRATIFICATION
QUESTIONS
1. Define the term "stratification" in the context of analyzing data. Which other statistical tool(s) can stratification
be used with?
2. Where is stratification most effectively used in a quality improvement or problem solving project?
3. How are stratified variables and their underlying meanings best illustrated?
4. What are the pitfalls to avoid when using stratification to analyze a set of data?
50
5. Stratification by itself will not lead a team to the root cause of the problem under investigation. What needs
to be done next?
In general, all projects must start with a clear and unambiguous mission statement - a clear description of the
problem to be addressed and the team's stated objectives when this has been achieved. Existing data from files,
logs and historical databases provides an unbiased and easily available source from which the team can establish
current performance measures before deciding on targets.
At the outset of the problem solving journey, flow diagrams of the system/process under investigation provide
members a common understanding of the "as is" picture. They also provide a means for the team to determine
the scope and boundaries of the project. From here, specific areas or sub-processes can be identified as useful
points to initiate investigations or data collection efforts. A Pareto of these constituent contributors can then
throw up the vital few issues to be resolved immediately.
Brainstorming and cause-effect diagrams are tools used to organize word data. Used consistently, they allow a
team to generate useful data in a systematic and focused way. Because several potential issues to be investigated
could be generated, a Pareto may once again be used to sieve the vital few from the useful many. 51
If actual data from the process under investigation is available, line graphs, bar and pie charts are the
appropriate tools to use to determine past and current status and trends.
A brainstorming session could again be conducted to generate theories on the suspected causes. New (or
existing) data could then be collected to test to prove or disprove these theories. Once again, the team could
rely on the flow diagram to identify appropriate data collection points. Alternately, a histogram or box plot could
be used to highlight deviations from the normal process. The deviations, interpreted correctly could also point to
the root cause of the problem.
Another useful tool for testing new data is the scatter diagram. A strong relationship testifies to the accuracy of
the theory put forth; a weak relationship points to an underlying source or problem as the root cause. Even a
"no relationship" interpretation assists the team. It says simply that the theory is wrong and can be dropped
without using further resources in that direction.
Following this iterative loop of theories testing, the team would have identified the root cause to the problem
(or, a few root causes in a process improvement project), and can now collect new data to confirm this (a new
scatter diagram or histogram/box plot should show a similar pattern). Having confirmed the root causes to be
resolved, corrective action could be planned and implemented. Planning tools (not discussed in this guide) like
tree diagrams, barriers and aids charts and implementation tables could be used here.
After implementing corrective actions, new data may be needed to determine how the new/modifed program is
performing. Again the various tools used above - line graphs/bar and pie charts, histograms/box plots, scatter
diagrams, Pareto charts and so on can be used to monitor improvement and/or progress. Comparisons between
before and after diagrams also provide a clear way to show the effectiveness of the remedial actions.
SUMMARY
At any point during the data interpretation/analysis, the team can use stratification as a tool to gain further
insights or depth to conclusions already reached with any of the other data analyzing tools, most commonly, bar
charts, scatter diagrams and histograms.
Finally, control charts (not discussed in this guide) and other SPC related tools should be used to monitor the
process and highlight deviations when they occur.
The tools, as you may have guessed, are just one part to an effective resolution of a problem/process
improvement. Work teams must also be equipped with the appropriate methodology to provide the team a
framework on their project journey from project identification and objective setting, through the diagnostic
journey, problem resolution, planning and implementation, to monitoring and maintaining achievements.
The last element of a successful problem/process improvement project addresses the human factor, or team
dynamics. To be effective in a team, members must be trained in and/or made aware of basic skills in team
activity. ST University and site training departments offer a variety of courses addressing this topic. See the ST
Learning Campus at learning.st.com for more information on available programs.
With all three elements of an effective team, your team can make a positive impact on the environment you
52 work in. Effective interaction between people, strictly adhering to a prescribed and proven methodology and
frequent, appropriate application of quality improvement tools is what we, as a company, are striving towards.
SELF-TEST
Each of the questions below indicate a scenario or situation that calls for the use of a particular tool or set of
tools. The tools are:
Indicate for each case which is the most appropriate tool. (It could be the same tool for several cases.)
1. A new team is meeting for the first time. Members, who work in the same area, are required to identify and
commence a project to reduce the number of rejects generated in their area. The team leader wants each
member to identify his/her area of responsibility where from experience, there is a (any) likelihood of a reject
occurring.
2. After a particularly successful brainstorming session, a group of people listed more than 40 likely causes of
time delay in processing a sales order document. It was immediately clear that not all of these were root
causes; several were actually the consequence of errors/mistakes created earlier. The team saw the need to
separate the data to reflect this distinction. 53
3. A cross-functional team was formed to resolve increasingly frequent customer complaints about poor response
to their queries. The members were representative of the various functions/departments involved. At the first
meeting, it was evident that each member was familiar with his/her department's role in the query response
process, but not necessarily with what happens before or after that stage of the process. The leader felt that
everyone must be familiar with the whole process before tackling the problem.
4. A team of workers was chartered to determine ways in which their dicing machine could process more wafers
and thereby increase throughput in their module. One member suggested increasing the speed of the saw for
faster dicing and therefore, cut more wafers in a given time. Another more senior operator objected, stating
that increasing saw speed would also mean increasing the MTBF (mean time to failure) of the fragile saw, and
therefore it could actually slow throughput if machine down time were increased due to the need to change
saws more frequently.
"Fine", said the leader, "let's put this theory to the test. We will dedicate one machine to dicing at an increased
speed and collect data on the number of saws used each day".
5. Members of a team investigating ways to reduce cycle time to process a credit note collected a sample of 100
past transactions to determine the average cycle time. It turned out that the average was 20 work days,
starting from the time a credit note request was received, to the time a credit note was generated. They also
determined that the process involved a total of four departments. The team felt that it would be useful to
know the spread of the 100 transactions over the 20 day cycle, as well as how this 20 day cycle was spread
over the four departments.
SELF-TEST
6. In the sale order problem stated earlier, the team finally arrived at the conclusion that of the 40 likely causes,
15 were distinct, potential root causes. Due to the limited manpower and time resources, it was not practical
to simultaneously investigate all of them. The team had to identify those few that, in their consensus, would
bring about the most optimum benefit.
7. In a study on consumption of paper, team members felt it would be useful to know the quantity of new paper
consumed throughout the office. However, when approaching the question of where and who is best to
collect the data, members were stumped.
8. A team was tasked to reduce the time required to perform a critical but complex task. This task, although
essential to the overall operation, was performed only when the item produced failed to meet a particular
criterion. Hence, the operation was infrequently performed and data was not easily available. The team
undertook some remedial action and wanted to determine the extent of improvement that was been
achieved.
9. As part of their SPC efforts, workers in charge of a particular process wanted to determine if the critical
measurements in their process conform to the defined control and specification limits. In the absence of the
relevant automated data capturing software, they collected data on a daily basis over a period of 30 working
54 days on these critical measures.
10. A problem solving team in the accounts office had the task to resolve a recent and increasingly frequent
number of customer complaints about being invoiced for the wrong items. In reviewing 60 erroneous invoices
generated from over the past two months, they were able to determine that in each invoice, the error
occurred in one of three categories: the day of week when the invoice were generated, the clerk who
generated the invoice, and the week of the month in which the invoice was generated. The team wanted to
determine which of these categories would be meaningful to investigate.
4. How do you proceed when you encounter a branch point when drawing a flow diagram of your process? 55
Continue with one branch until you reach a suitable terminal point, then return to branch point and continue
with the other branch. Repeat this until all branches are addressed. (To facilitate understanding of the process
flow, try to draw branch points as a 2-way, yes/no branch-off, with one branch continuing the process and the
other addressing other branching options).
5. How can you use an accurate flow diagram to better understand a process?
An accurate diagram will provide information on the process "as is". To gain a better understanding of the
process, users should "walk through" the diagram, step by step and ask appropriate questions addressing issues
of poor performance, redundancy and necessity.
6. What are the common pitfalls to avoid when using the flow diagram as a tool?
Failure to ensure a complete representation of the process will result in the team addressing the wrong issues.
Using a complex and detailed flow diagram during the early stages of the project may unnecessarily tie down
the team in non-critical areas of the process.
RECOMMENDED SOLUTIONS
2. BRAINSTORMING
1. What is the first step in a proper brainstorming session?
A clear and unambiguous description of the subject to be brainstormed helps keep participants focused.
4. Is there a need to collect data to support ideas/theories generated from brainstorming session? Why?
Ideas on their own constitute theories. By themselves, they may or may not explain the observed
discrepancy/problem. There is therefore a need to collect data to confirm or drop the suspicion.
56
5. How can a team avoid the dominance of one or a few individuals during a session?
Team members can avoid this problem by agreeing to strictly adhere to the conceptual and practical rules
during the brainstorming session.
3. CAUSE-E
EFFECT DIAGRAM
1. What is the primary purpose of a cause-effect diagram?
It is to organize ideas systematically to display causal links between them which lead to the observed effect.
2. In a typical C-E diagram, what description is placed in the box on the extreme right?
The effect or symptom for which the causes must be identified.
3. Which common tool could a team use to generate inputs for their CE diagram?
A brainstorming session generates useful inputs for the C-E diagram.
6. What is the danger of limiting the input of ideas when drawing up the diagram?
If the ideas put forth are not comprehensive, the real root cause(s) could be masked or ignored, thereby
leading the team away from an effective resolution of the problem.
RECOMMENDED SOLUTIONS
4. DATA COLLECTION
1 Why does data collection constitute an integral part of the problem solving process?
The proper analysis and interpretation of the appropriate data in a project will keep the team on-track and
minimize the danger of utilizing limited resources to mask or partially eliminate the root causes to the
problem.
2. What are 5 key considerations when planning to collect data for your project?
What questions need to be answered?
How to recognize and communicate these answers?
What appropriate tools to use to communicate these answers?
What type of data to collect?
Where in the process can the data be collected?
Who in the process can provide the data? who can collect?
How to minimize errors and/or maximize resources?
What additional data needed for future analysis or traceability?
3. What are the major considerations when designing a data collection form?
"KISS" principle; aim for self explanatory designs. 57
Reduce opportunities for error.
Ensure proper references and traceability.
Incorporate professionalism to solicit interest and cooperation.
4. A common bias to avoid in data collection is Non-response bias. Can you recall what this is?
Non-response bias is the phenomenon where a significant portion of the target group did not respond to the
data collection exercise; eg, the graveyard shift in a 24-hour plant decides not to participate in an opinion
survey scheduled immediately after their shift.
5. Discuss the merits and disadvantages of using existing data as opposed to collecting new data.)
Existing data are (usually) unbiased, easier to collect and compile and take much less time and effort. In
contrast, collecting new data requires proper designing of forms, identifying and training collectors. All this
will take up significant time and people resources. In addition, care will have to be taken to ensure non-biases
on the part of both data sources and collectors.
RECOMMENDED SOLUTIONS
5. GRAPHS AND CHARTS
1. When is a line graph best used? a bar graph? a pie chart?
Line graphs are best for showing current status and trends. Bar charts are used when a comparison between
categories is necessary and pie charts are frequently used to illustrate proportions contributing to the
phenomenon under study.
2. In a line graph, what are the 2 principal axes called? Which one is placed horizontally, which vertically?
The 2 axes are the independent axis (horizontal) and the dependent axis (vertical).
3. What is the difference between a grouped bar and a stacked bar presentation?
A grouped bar chart is an example of a bar chart stratified horizontally, showing the contributions of related
sub-categories contributing to a particular category. An example is Sales volume by geographical area within
country.
A stacked bar is an example of a vertical stratification. It is particularly useful to show how sub-contributions
add up to the whole for each contribution category.
4. What are the key concepts to bear in mind when preparing a graphic chart?
58 Graphic integrity, consistent scales and symbols, easy reading and simplicity.
5. How does varying the scale on one axis affect the presentation of a line graph?
Varying the scale results in a distorted graph. Excluding a significant portion of the proper range from which
the data is collected/measured may result in exclusion of data.
RECOMMENDED SOLUTIONS
6. PARETO ANALYSIS
1.How is a Pareto chart different from the normal bar chart?
The Pareto chart is a combination of a bar chart and a cumulative line graph. Additionally, the bars in a Pareto
chart are ordered or ranked from the tallest on the left to the least or shortest on the right.
4. How can the Pareto tool be used to illustrate improvements in a targeted area?
A comparison between "before" and "after" charts will show how effective the remedial or corrective actions
have been in reducing or eliminating the vital few contributors to the observed problem.
59
5. When, after constructing a chart, what steps can a team take if it appears that there are no obvious
distinctions between the various categories?
One reason why such a result has been obtained is that the categories of contributors have not been
appropriately defined. Try to think of other measures.
RECOMMENDED SOLUTIONS
7. SCATTER DIAGRAM
1. (What is the primary purpose of a scatter diagram?)
A scatter diagram is primarily used to test theories i.e., the strength (or weakness) of the relationship between
a suspected cause and the observed effect.
2. The variables involved in a scatter diagram are sometimes called the (suspected) cause variable and the
(observed) effect variable? Which is to be plotted on the horizontal axis? Which vertical?
The cause variable is plotted along the horizontal axis; the effect on the vertical axis.
5. Scatter diagrams are effective to the extent that they prove or disprove a theory or physical explanation.
Where does this theory originate from?
In the project context, brainstorming sessions and cause-effective diagrams are useful ways to generate such
theories. Another source could be from interpretation of the data from other tools - histograms/box plot
patterns for example, may indicate variations in the process.
RECOMMENDED SOLUTIONS
8. HISTOGRAM
1. How is a histogram different from the ordinary bar chart?
The primary objective of the Histogram is to show patterns or variations over a range of data. The horizontal
axis of the histogram graph reflects this range.
3. How many bars or categories would be appropriate in a data set of 75 values, 150 values, 225 values?
Respectively, 7, 8 and 9.
4. What could be a plausible physical explanation for a Double peaked pattern seen in a histogram, a skewed pattern?
The Double peaked pattern is an indication that 2 similar processes with differing means may be the
underlying cause to the observation.
A Skewed pattern is synonymous with the imposition of a practical limit on one side of the range of
measurements, close to the nominal or mean value.
5. What is the recommended minimum number of values needed to constructed a histogram? Why? 61
The recommended value is 40 data items. As stated, the objective of a Histogram is to provide a picture
pattern inherent in a set of data. Towards this end, the larger the sample, the more confidence we could have
that the peak, shape and spread of the histogram is representative of the process under investigation.
Although not necessarily precise, the number 40 is a good rule of thumb figure.
6. What alternative tool is recommended if there is insufficient data values to construct the histogram?
If collection of data samples is difficult or impractical, an alternative to using the Histogram is the Box Plot,
which allows for limited visibility of variation in data.
9. BOX PLOT
1. When is a box plot considered a good substitute for the histogram?
When there is insufficient data available and/or data collection is not practical.
3. In a box plot comparing 2 processes, name 3 features to distinguish the better process from the other.
Difference in the location of the median, difference in the amount of variation (length of each box), the
presence or absence of "outliners", symmetry or asymmetry in the data samples.
4. Why is it better to use a histogram rather than a box plot in certain situations?
Box plots provide a limited view of the pattern of variation in a data sample due to the fewer number of data
points.
RECOMMENDED SOLUTIONS
10. STRATIFICATION
1. Define the term "Stratification" in the context of analyzing data. With which other statistical tool(s) can
stratification be used?
Stratification is the breaking down of data samples into a second or even third level to gain further insights
into the information provided by the first analysis of the data. It is or could be used in conjunction with Bar
charts, Scatter diagrams, Histograms... in fact with any other tool if the data available could be broken down
into sub-groups or sub-categories.
2. Where is stratification most effectively used in a quality improvement or problem solving project?
Stratification is most often used to analyze the observed symptoms. It could also be utilized when testing
theories and when attempting to identify a root cause.
3. How are stratified variables and their underlying meanings best illustrated?
Stratification is a means to breakdown data into relevant or meaningful groups. Having done this, the next
step is to "pictorize" the data by means of one of the other tools, the choice of which would depend on the
nature of the data.
62 4. What are the pitfalls to avoid when using stratification to analyze a set of data?
Avoid interpreting too much from the resulting diagram, especially if the differences observed are not distinct.
Stratification should be seen as a tool to obtain additional insights to the problem. Do not assume that it will
immediately point to a root cause.
5. Stratification by itself will not lead a team to the root cause of the problem under investigation. What needs
to be done next?
If the results are positive, the next step following a stratification analysis is to collect additional data or details
to confirm the theory.
RECOMMENDED SOLUTIONS
SELF-T
TEST
1. A brainstorming session will allow all members present to contribute likely causes of rejects in their work
areas.
2. The Cause-Effect diagram could be drawn to show up the causal links between these likely or potential
causes, leading ultimately to the observed problem.
3. A high level or relationship map should be drawn so that all members may see/understand how all the
functions relate to one another. A detailed process flow diagram could then be provided by each member
responsible for a function.
4. Two theories have been proposed here. In both cases, a scatter diagram could be used to test them. The
suspected cause variable is the speed of the dicing machine, while the suspected effects are the number of
wafers diced (per unit time) and the number of saws consumed (per unit time).
5. Determining the spread of the 20-day cycle requires a Histogram - the team is contemplating the variation of
the process. The idea of breaking up the spread over the various contributing functions is an example of 63
stratification of the available data.
6. A Pareto of data collected on 15 distinct root causes will separate the vital, critical causes from the useful
many. This helps the team prioritize the causes in order to address the critical few first.
7. A detailed flow diagram of the operations involved in paper generation will show the team appropriate points
in the process to collect the data, as well as ascertaining responsibility for the collection.
8. As data is not readily available or collectable, a box plot should be considered. In this instance, the team
wanted to determine the benefits of its remedial effort. A comparison of a box plot before the remedial action
and one after should provide the needed analysis.
9. Line graph of the minimum, maximum and mean of the data collected each day superimposed on the control
and specification limits would indicate how the process is meeting the requirements.
10. This is another example of stratification, the variables involved being the three defined. Simple bar charts
drawn from the data for each of these categories will suffice to show any distortions to the acceptable norm.
®
Recycled and chlorine free paper
© Copyright 2003 STMicroelectronics. Printed in France.