14

14-1 Overview
14-2 Control Charts for Variation and Mean
14-3 Control Charts for Attributes
Statistical Process
Control
5014_TriolaE/S_CH14pp732-757 11/22/05 6:17 AM Page 732
C H A P T E R P R O B L E M
Is the production of aircraft altimeters
dangerous for those who fly?
The Altigauge Manufacturing Company produces aircraft
altimeters, which provide pilots with readings of their
heights above sea level. The accuracy of altimeters is im-
portant because pilots rely on them to maintain altitudes
with safe vertical clearance above mountains, towers, and
tall buildings, as well as vertical separation from other air-
craft. The accuracy of altimeters is especially important
when pilots fly approaches to landing while not being able
to see the ground. In the past, pilots and passengers have
been killed in crashes caused by wrong altimeter readings
that led pilots to believe that they were safely above the
ground when they were actually flying dangerously low.
Because aircraft altimeters are so critically important
to aviation safety, their accuracy is carefully controlled
by government regulations. According to Federal Avia-
tion Administration Regulation Part 43, Appendix E, an
altimeter must give a reading with an error of no more
than 20 ft when tested for an altitude of 1000 ft.
At the Altigauge Manufacturing Company, four al-
timeters are randomly selected from production on each
of 20 consecutive business days, and Table 14-1 lists the
errors (in feet) when they are tested in a pressure cham-
ber that simulates an altitude of 1000 ft. On Day 1, for
example, the actual altitude readings for the four se-
lected altimeters are 1002 ft, 992 ft, 1005 ft, and 1011 ft,
so the corresponding errors (in feet) are 2, 28, 5, and 11.
In this chapter we will evaluate this altimeter manu-
facturing process by analyzing the behavior of the er-
rors over time. We will see how methods of statistics
can be used to monitor a manufacturing process with
the goal of identifying and correcting any serious prob-
lems. In addition to helping companies stay in business,
methods of statistics can positively affect our safety in
very significant ways.
Table 14-1 Aircraft Altimeter Errors (in feet)
Day Error Mean Median Range Standard Deviation
1 2 28 5 11 2.50 3.5 19 7.94
2 25 2 6 8 2.75 4.0 13 5.74
3 6 7 21 28 1.00 2.5 15 6.98
4 25 5 25 6 0.25 0.0 11 6.08
5 9 3 22 22 2.00 0.5 11 5.23
6 16 210 21 28 20.75 24.5 26 11.81
7 13 28 27 2 0.00 22.5 21 9.76
8 25 24 2 8 0.25 21.0 13 6.02
9 7 13 22 213 1.25 2.5 26 11.32
10 15 7 19 1 10.50 11.0 18 8.06
11 12 12 10 9 10.75 11.0 3 1.50
12 11 9 11 20 12.75 11.0 11 4.92
13 18 15 23 28 21.00 20.5 13 5.72
14 6 32 4 10 13.00 8.0 28 12.91
15 16 213 29 19 3.25 3.5 32 16.58
16 8 17 0 13 9.50 10.5 17 7.33
17 13 3 6 13 8.75 9.5 10 5.06
18 38 25 25 5 8.25 0.0 43 20.39
19 18 12 25 26 12.25 15.0 31 13.28
20 227 23 7 36 9.75 15.0 63 27.22
5014_TriolaE/S_CH14pp732-757 11/22/05 6:17 AM Page 733
734 Chapter 14 Statistical Process Control
14-1 Overview
In Chapter 2 we noted that when describing, exploring, or comparing data sets,
it is usually important to consider center, variation, distribution, outliers, and
changing characteristics over time. The main objective of this chapter is to ad-
dress the last item: changing characteristics of data over time. When investigat-
ing characteristics such as center and variation, it is important to know whether
we are dealing with a stable population or one that is changing with the passage
of time.
There is currently a strong trend toward trying to improve the quality of
American goods and services, and the methods presented in this chapter are being
used by growing numbers of businesses. Evidence of the increasing importance of
quality is found in its greater role in advertising and the growing number of books
and articles that focus on the issue of quality. In many cases, job applicants (you?)
have a definite advantage when they can tell employers that they have studied
statistics and methods of quality control. This chapter will present some of the ba-
sic tools commonly used to monitor quality.
Minitab, Excel, and other software packages include programs for automati-
cally generating charts of the type discussed in this chapter, and we will include
several examples of such displays. Control charts are good examples of wonderful
graphical devices that allow us to see and understand some property of data that
would be more difficult or impossible to understand without graphs. The world
needs more people who can construct and interpret important graphs, such as the
control charts described in this chapter.
Control Charts for Variation
14-2 and Mean
Key Concept The main objective of this section is to construct run charts, R
charts, and charts so that we can monitor important features of data over time.
We will use such charts to determine whether some process is statistically stable
(or within statistical control).
The following definition formally describes the type of data that will be con-
sidered in this chapter.
x
Definition
Process data are data arranged according to some time sequence. They are
measurements of a characteristic of goods or services that result from some
combination of equipment, people, materials, methods, and conditions.
For example, Table 14-1 includes process data consisting of the measured error
(in feet) in altimeter readings over 20 consecutive days of production. Each day,
four altimeters were randomly selected and tested. Because the data in Table 14-1
5014_TriolaE/S_CH14pp732-757 11/22/05 6:17 AM Page 734
14-2 Control Charts for Variation and Mean 735
are arranged according to the time at which they were selected, they are process
data. It is very important to recognize this point:
Important characteristics of process data can change over time.
In making altimeters, a manufacturer might use competent and well-trained per-
sonnel along with good machines that are correctly calibrated, but if the personnel
are replaced or the machines wear with use, the altimeters might begin to become
defective. Companies have gone bankrupt because they unknowingly allowed
manufacturing processes to deteriorate without constant monitoring.
Run Charts
There are various methods that can be used to monitor a process to ensure that the
important desired characteristics dont changeanalysis of a run chart is one
such method.
Definition
Arun chart is a sequential plot of individual data values over time. One axis
(usually the vertical axis) is used for the data values, and the other axis (usu-
ally the horizontal axis) is used for the time sequence.
Figure 14-1 Run Chart of Individual Altimeter
Errors in Table 14-1
EXAMPLE Manufacturing Aircraft Altimeters Treating
the 80 altimeter errors in Table 14-1 as a string of consecutive mea-
surements, construct a run chart by using a vertical axis for the errors
and a horizontal axis to identify the order of the sample data.
SOLUTION Figure 14-1 is the Minitab-generated run chart for the data in
Table 14-1. The vertical scale is designed to be suitable for altimeter errors
Minitab
continued
5014_TriolaE/S_CH14pp732-757 11/22/05 6:17 AM Page 735
736 Chapter 14 Statistical Process Control
Interpreting Run Charts Only when a process is statistically stable can its
data be treated as if they came from a population with a constant mean, standard
deviation, distribution, and other characteristics.
ranging from 227 ft to 38 ft, which are the minimum and maximum values in
Table 14-1. The horizontal scale is designed to include the 80 values arranged
in sequence. The first point represents the first value of 2 ft, the second point
represents the second value of 28 ft, and so on.
In Figure 14-1, the horizontal scale identifies the sample number, so the
number 20 indicates the 20th sample item. The vertical scale represents the
altimeter error (in feet). Now examine Figure 14-1 and try to identify any
patterns that jump out begging for attention. Figure 14-1 does reveal this
problem: As time progresses from left to right, the heights of the points
appear to show a pattern of increasing variation. See how the points at the
left fluctuate considerably less than the points farther to the right. The Fed-
eral Aviation Administration regulations require errors less than 20 ft (or
between 20 ft and 220 ft), so the altimeters represented by the points at the
left are okay, whereas several of the points farther to the right correspond to
altimeters not meeting the required specifications. It appears that the
manufacturing process started out well, but deteriorated as time passed. If
left alone, this manufacturing process will cause the company to go out of
business.
Definition
Aprocess is statistically stable (or within statistical control) if it has only
natural variation, with no patterns, cycles, or unusual points.
Figure 14-2 consists of run charts illustrating typical patterns showing ways in
which the process of filling 12-oz cola cans may not be statistically stable.
Figure 14-2(a): There is an obvious upward trend that corresponds to
values that are increasing over time. If the filling process were to follow
this type of pattern, the cans would be filled with more and more cola
until they began to overflow, eventually leaving the employees swim-
ming in cola.
Figure 14-2(b): There is an obvious downward trend that corresponds to
steadily decreasing values. The cans would be filled with less and less cola
until they were extremely underfilled. Such a process would require a com-
plete reworking of the cans in order to get them full enough for distribution
to consumers.
Figure 14-2(c): There is an upward shift. Arun chart such as this one might
result from an adjustment to the filling process, making all subsequent val-
ues higher.
The Flynn Effect:
Upward Trend in
IQ Scores
Arun chart or control chart of
IQ scores would reveal that
they exhibit an upward trend,
because IQ scores have been
steadily increasing since they
began to be used about 70 years
ago. The trend is worldwide,
and it is the same for different
types of IQ tests, even those
that rely heavily on abstract and
nonverbal reasoning with mini-
mal cultural influence. This up-
ward trend has been named the
Flynn effect, because political
scientist James R. Flynn dis-
covered the trend in his studies
of U.S. military recruits. The
amount of the increase is quite
substantial: Based on a current
mean IQ score of 100, it is esti-
mated that the mean IQ in 1920
would be about 77. The typical
student of today is therefore
brilliant when compared to his
or her great-grandparents. So
far, there is no generally ac-
cepted explanation for the
Flynn effect.
5014_TriolaE/S_CH14pp732-757 11/22/05 6:17 AM Page 736
14-2 Control Charts for Variation and Mean 737
Figure 14-2(d): There is a downward shiftthe first few values are rela-
tively stable, and then something happened so that the last several values
are relatively stable, but at a much lower level.
Figure 14-2(e): The process is stable except for one exceptionally high
value. The cause of that unusual value should be investigated. Perhaps the
cans became temporarily stuck and one particular can was filled twice in-
stead of once.
Figure 14-2(f): There is an exceptionally low value.
Figure 14-2
Processes That Are Not
Statistically Stable
Minitab
(a) (b)
(c) (d )
(e) (f )
(g) (h)
5014_TriolaE/S_CH14pp732-757 11/22/05 6:17 AM Page 737
738 Chapter 14 Statistical Process Control
Definitions
Random variation is due to chance; it is the type of variation inherent in any
process that is not capable of producing every good or service exactly the
same way every time.
Assignable variation results from causes that can be identified (such factors
as defective machinery, untrained employees, and so on).
Later in the chapter we will consider ways to distinguish between assignable vari-
ation and random variation.
The run chart is one tool for monitoring the stability of a process. We will
now consider control charts, which are also extremely useful for that same
purpose.
Control Chart for Monitoring Variation: The R Chart
In the article The State of Statistical Process Control as We Proceed into the
21st Century (Stoumbos, Reynolds, Ryan, and Woodall, Journal of the Ameri-
can Statistical Association, Vol. 95, No. 451), the authors state that control
charts are among the most important and widely used tools in statistics. Their
applications have now moved far beyond manufacturing into engineering, en-
vironmental science, biology, genetics, epidemiology, medicine, finance, and
even law enforcement and athletics. We begin with the definition of a control
chart.
Figure 14-2(g): There is a cyclical pattern (or repeating cycle). This pat-
tern is clearly nonrandom and therefore reveals a statistically unstable pro-
cess. Perhaps periodic overadjustments are being made to the machinery,
with the effect that some desired value is continually being chased but
never quite captured.
Figure 14-2(h): The variation is increasing over time. This is a common
problem in quality control. The net effect is that products vary more and
more until almost all of them are worthless. For example, some cola cans
will be overflowing with wasted cola, and some will be underfilled and un-
suitable for distribution to consumers.
A common goal of many different methods of quality control is this:
reduce variation in the product or service. For example, Ford became con-
cerned with variation when it found that its transmissions required signifi-
cantly more warranty repairs than the same type of transmissions made by
Mazda in Japan. A study showed that the Mazda transmissions had substan-
tially less variation in the gearboxes; that is, crucial gearbox measurements
varied much less in the Mazda transmissions. Although the Ford transmissions
were built within the allowable limits, the Mazda transmissions were more reli-
able because of their lower variation. Variation in a process can result from two
types of causes.
x
5014_TriolaE/S_CH14pp732-757 11/22/05 6:17 AM Page 738
14-2 Control Charts for Variation and Mean 739
We will assume that the population standard deviation s is not known as we
consider only two of several different types of control charts: (1) R charts (or
range charts) used to monitor variation and (2) charts used to monitor means.
When using control charts to monitor a process, it is common to consider R charts
and charts together, because a statistically unstable process may be the result of
increasing variation or changing means or both.
An R chart (or range chart) is a plot of the sample ranges instead of individ-
ual sample values, and it is used to monitor the variation in a process. (It might
make more sense to use standard deviations, but range charts are used more often
in practice. This is a carryover from times when calculators and computers were
not available. See Exercise 17 for a control chart based on standard deviations.) In
addition to plotting the range values, we include a centerline located at , which
denotes the mean of all sample ranges, as well as another line for the lower control
limit and a third line for the upper control limit. Following is a summary of nota-
tion for the components of the R chart.
R
x
x
Definition
Acontrol chart of a process characteristic (such as mean or variation) con-
sists of values plotted sequentially over time, and it includes a centerline as
well as a lower control limit (LCL) and an upper control limit (UCL). The
centerline represents a central value of the characteristic measurements,
whereas the control limits are boundaries used to separate and identify any
points considered to be unusual.
Notation
Given: Process data consisting of a sequence of samples all of the same size n,
and the distribution of the process data is essentially normal.
n 5size of each sample, or subgroup
5mean of the sample ranges (that is, the sum of the sample ranges divided by
the number of samples)
R
Monitoring Process Variation: Control Chart for R
Points plotted: Sample ranges
Centerline:
Upper control limit (UCL): D
4
(where D
4
is found in Table 14-2)
Lower control limit (LCL): D
3
(where D
3
is found in Table 14-2) R
R
R
The values of D
4
and D
3
were computed by quality-control experts, and they are
intended to simplify calculations. The upper and lower control limits of D
4
and
D
3
are values that are roughly equivalent to 99.7% confidence interval limits. It
is therefore highly unlikely that values from a statistically stable process would
fall beyond those limits. If a value does fall beyond the control limits, its very
likely that the process is not statistically stable.
R
R
Costly Assignable
Variation
The Mars Climate Orbiter was
launched by NASAand sent
to Mars, but it was destroyed
when it flew too close to its
destination planet. The loss
was estimated at $125 million.
The cause of the crash was
found to be confusion be-
tween the use of units used for
calculations. Acceleration data
were provided in the English
units of pounds of force, but
the Jet Propulsion Laboratory
assumed that those units were
in metric newtons instead of
pounds. The thrusters of the
spacecraft subsequently pro-
vided wrong amounts of force
in adjusting the position of the
spacecraft. The errors caused
by the discrepancy were fairly
small at first, but the cumula-
tive error over months of the
spacecrafts journey proved to
be fatal to its success.
In 1962, the rocket carry-
ing the Mariner 1 satellite
was destroyed by ground con-
trollers when it went off
course due to a missing minus
sign in a computer program.
5014_TriolaE/S_CH14pp732-757 11/22/05 6:17 AM Page 739
740 Chapter 14 Statistical Process Control
Table 14-2 Control Chart Constants
n: Number of s R
Observations
in Subgroup A
2
A
3
B
3
B
4
D
3
D
4
2 1.880 2.659 0.000 3.267 0.000 3.267
3 1.023 1.954 0.000 2.568 0.000 2.574
4 0.729 1.628 0.000 2.266 0.000 2.282
5 0.577 1.427 0.000 2.089 0.000 2.114
6 0.483 1.287 0.030 1.970 0.000 2.004
7 0.419 1.182 0.118 1.882 0.076 1.924
8 0.373 1.099 0.185 1.815 0.136 1.864
9 0.337 1.032 0.239 1.761 0.184 1.816
10 0.308 0.975 0.284 1.716 0.223 1.777
11 0.285 0.927 0.321 1.679 0.256 1.744
12 0.266 0.886 0.354 1.646 0.283 1.717
13 0.249 0.850 0.382 1.618 0.307 1.693
14 0.235 0.817 0.406 1.594 0.328 1.672
15 0.223 0.789 0.428 1.572 0.347 1.653
16 0.212 0.763 0.448 1.552 0.363 1.637
17 0.203 0.739 0.466 1.534 0.378 1.622
18 0.194 0.718 0.482 1.518 0.391 1.608
19 0.187 0.698 0.497 1.503 0.403 1.597
20 0.180 0.680 0.510 1.490 0.415 1.585
21 0.173 0.663 0.523 1.477 0.425 1.575
22 0.167 0.647 0.534 1.466 0.434 1.566
23 0.162 0.633 0.545 1.455 0.443 1.557
24 0.157 0.619 0.555 1.445 0.451 1.548
25 0.153 0.606 0.565 1.435 0.459 1.541
Source: Adapted from ASTM Manual on the Presentation of Data and Control Chart Analysis,
1976 ASTM, pp. 134136. Reprinted with permission of American Society for Testing and
Materials.
x
Dont Tamper!
Nashua Corp. had trouble with
its paper-coating machine and
considered spending a million
dollars to replace it. The ma-
chine was working well with a
stable process, but samples
were taken every so often and,
based on the results, adjust-
ments were made. These over-
adjustments, called tampering,
caused shifts away from the
distribution that had been
good. The effect was an in-
crease in defects. When statis-
tician and quality expert W.
Edwards Deming studied the
process, he recommended that
no adjustments be made unless
warranted by a signal that the
process had shifted or had be-
come unstable. The company
was better off with no adjust-
ments than with the tampering
that took place.
EXAMPLE Manufacturing Aircraft Altimeters Refer to
the altimeter errors in Table 14-1. Using the samples of size n 5 4
collected each day of manufacturing, construct a control chart for R.
SOLUTION We begin by finding the value of , the mean of the sample ranges.
The centerline for our R chart is therefore located at 521.2. To find the upper
and lower control limits, we must first find the values of D
3
and D
4
. Referring to
R
R 5
19 1 13 1
c
1 63
20
5 21.2
R
5014_TriolaE/S_CH14pp732-757 11/22/05 6:17 AM Page 740
14-2 Control Charts for Variation and Mean 741
Interpreting Control Charts
When interpreting control charts, the following point is extremely important:
Upper and lower control limits of a control chart are based on the
actual behavior of the process, not the desired behavior. Upper and
lower control limits are totally unrelated to any process specifications
that may have been decreed by the manufacturer.
When investigating the quality of some process, there are typically two key ques-
tions that need to be addressed:
1. Based on the current behavior of the process, can we conclude that the pro-
cess is within statistical control?
2. Do the process goods or services meet design specifications?
The methods of this chapter are intended to address the first question, but not the
second. That is, we are focusing on the behavior of the process with the objective
of determining whether the process is within statistical control. Whether the pro-
cess results in goods or services that meet some stated specifications is another
issue not addressed by the methods of this chapter. For example, the preceding
Minitab R chart includes upper and lower control limits of 48.36 and 0, which
result from the sample values listed in Table 14-1. Government regulations
Table 14-2 for n 5 4, we get D
3
5 0.000 and D
4
5 2.282, so the control limits
are as follows:
Upper control limit: D
4
5(2.282)(21.2) 548.4
Lower control limit: D
3
5(0.000)(21.2) 50.0
Using a centerline value of 5 21.2 and control limits of 48.4 and 0.0, we
now proceed to plot the sample ranges. The result is shown in the Minitab
display.
R
R
R
Minitab
5014_TriolaE/S_CH14pp732-757 11/22/05 6:17 AM Page 741
742 Chapter 14 Statistical Process Control
require that altimeters have errors between 220 ft and 20 ft, but those desired (or
required) specifications are not included in the control chart for R.
Also, we should clearly understand the specific criteria for determining
whether a process is in statistical control (that is, whether it is statistically stable).
So far, we have noted that a process is not statistically stable if its pattern resembles
any of the patterns shown in Figure 14-2. This criterion is included with some oth-
ers in the following list.
Criteria for Determining When a Process Is Not Statistically Stable
(Out of Statistical Control)
1. There is a pattern, trend, or cycle that is obviously not random (such as those
depicted in Figure 14-2).
2. There is a point lying outside of the region between the upper and lower con-
trol limits. (That is, there is a point above the upper control limit or below the
lower control limit.)
3. Run of 8 Rule: There are eight consecutive points all above or all below the
centerline. (With a statistically stable process, there is a 0.5 probability that a
point will be above or below the centerline, so it is very unlikely that eight
consecutive points will all be above the centerline or all below it.)
We will use only the three out-of-control criteria listed above, but some businesses
use additional criteria such as these:
There are six consecutive points all increasing or all decreasing.
There are 14 consecutive points all alternating between up and down (such
as up, down, up, down, and so on).
Two out of three consecutive points are beyond control limits that are 2
standard deviations away from the centerline.
Four out of five consecutive points are beyond control limits that are 1 stan-
dard deviation away from the centerline.
EXAMPLE Statistical Process Control Examine the R chart
shown in the Minitab display for the preceding example and deter-
mine whether the process variation is within statistical control.
SOLUTION We can interpret control charts for R by applying the three out-
of-control criteria just listed. Applying the three criteria to the Minitab display
of the R chart, we conclude that variation in this process is out of statistical
control. There are not eight consecutive points all above or all below the cen-
terline, so the third condition is not violated, but the first two conditions are
violated.
1. There is a pattern, trend, or cycle that is obviously not random: Going from
left to right, there is a pattern of upward trend, as in Figure 14-2(a).
2. There is a point (the rightmost point) that lies above the upper control limit.
5014_TriolaE/S_CH14pp732-757 11/22/05 6:17 AM Page 742
14-2 Control Charts for Variation and Mean 743
Control Chart for Monitoring Means: The Chart
An chart is a plot of the sample means, and it is used to monitor the center in a
process. In addition to plotting the sample means, we include a centerline located
at , which denotes the mean of all sample means (equal to the mean of all sample
values combined), as well as another line for the lower control limit and a third
line for the upper control limit. Using the approach common in business and in-
dustry, the centerline and control limits are based on ranges instead of standard
deviations. See Exercise 18 for an chart based on standard deviations. x
x
x
x
Monitoring Process Mean: Control Chart for
Points plotted: Sample means
Centerline: 5mean of all sample means
Upper control limit (UCL): 1A
2
(where A
2
is found in Table 14-2)
Lower control limit (LCL): 2A
2
(where A
2
is found in Table 14-2) R x
R x
x
x
continued
Bribery Detected
with Control Charts
Control charts were used to
help convict a person who
bribed Florida jai alai players
to lose. (See Using Control
Charts to Corroborate
Bribery in Jai Alai, by
Charnes and Gitlow, The
American Statistician, Vol.
49, No. 4.) An auditor for one
jai alai facility noticed that
abnormally large sums of
money were wagered for cer-
tain types of bets, and some
contestants didnt win as
much as expected when those
bets were made. R charts and
charts were used in court as
evidence of highly unusual
patterns of betting. Examina-
tion of the control charts
clearly shows points well be-
yond the upper control limit,
indicating that the process of
betting was out of statistical
control. The statistician was
able to identify a date at
which assignable variation
appeared to stop, and prose-
cutors knew that it was the
date of the suspects arrest.
x
INTERPRETATION We conclude that the variation (not necessarily the
mean) of the process is out of statistical control. Because the variation appears
to be increasing with time, immediate corrective action must be taken to fix the
variation among the altimeter errors.
EXAMPLE Manufacturing Aircraft Altimeters Refer to
the altimeter errors in Table 14-1. Using samples of size n 5 4 col-
lected each working day, construct a control chart for . Based on the
control chart for only, determine whether the process mean is within statisti-
cal control.
SOLUTION Before plotting the 20 points corresponding to the 20 values of ,
we must first find the value for the centerline and the values for the control
limits. We get
Referring to Table 14-2, we find that for n 5 4, A
2
5 0.729. Knowing the val-
ues of , A
2
, and , we can now evaluate the control limits.
Upper control limit: 1A
2
56.45 1(0.729)(21.2) 521.9
Lower control limit: 2A
2
56.45 2(0.729)(21.2) 529.0
INTERPRETATION The resulting control chart for will be as shown in
the accompanying Excel display. Examination of the control chart shows
that the process mean is out of statistical control because at least one of the
three out-of-control criteria is not satisfied. Specifically, the third criterion
x
R x
R x
R x
R
5
19 1 13 1
. . .
1 63
20
5 21.2
x 5
2.50 1 2.75 1
. . .
1 9.75
20
5 6.45
x
x
x
5014_TriolaE/S_CH14pp732-757 11/22/05 6:17 AM Page 743
744 Chapter 14 Statistical Process Control
is not satisfied because there are eight (or more) consecutive points all be-
low the centerline. Also, there does appear to be a pattern of an upward
trend. Again, immediate corrective action is required to fix the production
process.
Using Technology
STATDISK See the STATDISK Student
Laboratory Manual and Workbook that is a
supplement to this book.
MINITAB Run Chart: To construct a
run chart, such as the one shown in Figure
14-1, begin by entering all of the sample data
in column C1. Select the option Stat, then
Quality Tools, then Run Chart. In the indi-
cated boxes, enter C1 for the single column
variable, enter 1 for the subgroup size, and
then click on OK.
R Chart: First enter the individual sample
values sequentially in column C1. Next, se-
lect the options Stat, Control Charts, Vari-
ables Charts for Subgroups, and R. Enter
C1 in the data entry box, enter the sample
size in the box for the subgroup size, and
click on R Options, then estimate. Select
Rbar. (Selection of the R bar estimate
causes the variation of the population dis-
tribution to be estimated with the sample
ranges instead of the sample standard devi-
ations, which is the default.) Click OK
twice.
Chart: First enter the individual sample
values sequentially in column C1. Next, se-
lect the options Stat, Control Charts, Vari-
ables Charts for Subgroups, and Xbar.
Enter C1 in the data entry box, enter the size
of each of the samples in the subgroup
sizes box. Click on Xbar Options, then se-
lect estimate and choose the option of
Rbar. Click OK twice.
EXCEL To use the Data Desk XL add-
in, click on DDXL and select Process Con-
trol. Proceed to select the type of chart you
want. (You must first enter the data in col-
umn Awith sample identifying codes entered
in column B. For the data of Table
14-1, for example, enter a 1 in column B ad-
jacent to each value from day 1, enter a 2 for
each value from day 2, and so on.)
To use Excels built-in graphics features
instead of Data Desk XL, see the following:
Run Chart: Enter all of the sample data in
column A. On the main menu bar, click on
the Chart Wizard icon, which looks like a
bar graph. For the chart type, select Line.
For the chart subtype, select the first graph
x
Excel
continued
5014_TriolaE/S_CH14pp732-757 11/22/05 6:17 AM Page 744
14-2 Control Charts for Variation and Mean 745
14-2 BASIC SKILLS AND CONCEPTS
Statistical Literacy and Critical Thinking
1. Process Data What are process data?
2. Statistical Control What does it mean for a process to be out of statistical control?
3. Control Charts What is a control chart? What is an R chart? What is an chart? What
is the difference between an R chart and an chart?
4. Variation What is the difference between random variation and assignable variation?
Interpreting Run Charts. In Exercises 58, examine the run chart from a process of fill-
ing 12-oz cans of cola and do the following: (a) Determine whether the process is within
statistical control; (b) if the process is not within statistical control, identify reasons why
it is not; (c) apart from being within statistical control, does the process appear to be be-
having as it should?
5. 6.
x
x
in the second row, then click Next. Continue
to click Next, then Finish. The graph can be
edited to include labels, delete grid lines,
and so on.
R Chart: Step 1: Enter the sample data in
rows and columns corresponding to the data
set. For example, enter the data in Table 14-1
in four columns (A, B, C, D) and 20 rows as
shown in the table.
Step 2: Next, create a column of the
range values using the following procedure.
Position the cursor in the first empty cell to
the right of the block of sample data, then en-
ter this expression in the formula box: 5
MAX(A1:D1) 2 MIN(A1:D1), where the
range A1:D1 should be modified to de-
scribe the first row of your data set. After
pressing the Enter key, the range for the
first row should appear. Use the mouse to
click and drag the lower right corner of this
cell, so that the whole column fills up with
the ranges for the different rows.
Step 3: Next, produce a graph by follow-
ing the same procedure described for the
run charts, but be sure to refer to the column
of ranges when entering the input range.
You can insert the required centerline and
upper and lower control limits by editing
the graph. Click on the line on the bottom of
the screen, then click and drag to position the
line correctly.
Chart: Step 1: Enter the sample data in
rows and columns corresponding to the data
set. For example, enter the data in Table 14-1
in four columns (A, B, C, D) and 20 rows as
shown in the table.
Step 2: Next, create a column of the sam-
ple means using the following procedure.
Position the cursor in the first empty cell to
the right of the block of sample data, then
enter this expression in the formula box: 5
AVERAGE(A1:D1), where the range A1:D1
should be modified to describe the first row
of your data set. After pressing the Enter
key, the mean for the first row should ap-
pear. Use the mouse to click and drag the
lower right corner of this cell, so that the
whole column fills up with the means for
the different rows.
Step 3: Next, produce a graph by follow-
ing the same procedure described for the run
chart, but be sure to refer to the column of
means when entering the input range. You
can insert the required centerline and upper
and lower control limits by editing the graph.
Click on the line on the bottom of the screen,
then click and drag to position the line cor-
rectly. Its not easy.
x
5014_TriolaE/S_CH14pp732-757 11/25/05 8:37 AM Page 745
746 Chapter 14 Statistical Process Control
7. 8.
Constructing Control Charts for Aluminum Cans. Exercises 9 and 10 are based on the
axial loads (in pounds) of aluminum cans that are 0.0109 in. thick, as listed in Data Set
15 in Appendix B. An axial load of a can is the maximum weight supported by its side,
and it is important to have an axial load high enough so that the can isnt crushed when
the top lid is pressed into place. The data are from a real manufacturing process, and
they were provided by a student who used an earlier edition of this book.
9. R Chart In each day of production, seven aluminum cans with thickness 0.0109 in.
were randomly selected and the axial loads were measured. The ranges for the differ-
ent days are listed below, but they can also be found from the values given in Data Set
15 in Appendix B. Construct an R chart and determine whether the process variation
is within statistical control. If it is not, identify which of the three out-of-control crite-
ria lead to rejection of statistically stable variation.
78 77 31 50 33 38 84 21 38 77 26 78 78
17 83 66 72 79 61 74 64 51 26 41 31
10. Chart In each day of production, seven aluminum cans with thickness 0.0109 in.
were randomly selected and the axial loads were measured. The means for the differ-
ent days are listed below, but they can also be found from the values given in Data Set
15 in Appendix B. Construct an chart and determine whether the process mean is
within statistical control. If it is not, identify which of the three out-of-control criteria
lead to rejection of statistically stable variation.
252.7 247.9 270.3 267.0 281.6 269.9 257.7 272.9 273.7 259.1 275.6 262.4 256.0
277.6 264.3 260.1 254.7 278.1 259.7 269.4 266.6 270.9 281.0 271.4 277.3
Monitoring the Minting of Quarters. In Exercises 1113, use the following informa-
tion: The U.S. Mint has a goal of making quarters with a weight of 5.670 g, but any
weight between 5.443 g and 5.897 g is considered acceptable. A new minting machine
is placed into service and the weights are recorded for a quarter randomly selected
every 12 min for 20 consecutive hours. The results are listed in the accompanying
table.
11. Minting Quarters: Run Chart Construct a run chart for the 100 values. Does there ap-
pear to be a pattern suggesting that the process is not within statistical control? What
are the practical implications of the run chart?
12. Minting Quarters: R Chart Construct an R chart and determine whether the process
variation is within statistical control. If it is not, identify which of the three out-of-
control criteria lead to rejection of statistically stable variation.
13. Minting Quarters: Chart Construct an chart and determine whether the process
mean is within statistical control. If it is not, identify which of the three out-of-control
criteria lead to rejection of a statistically stable mean. Does this process need correc-
tive action?
x x
x
x
5014_TriolaE/S_CH14pp732-757 11/22/05 6:17 AM Page 746
14-2 Control Charts for Variation and Mean 747
Weights (in grams) of Minted Quarters
Hour Weight (g) s Range
1 5.639 5.636 5.679 5.637 5.691 5.6564 0.0265 0.055
2 5.655 5.641 5.626 5.668 5.679 5.6538 0.0211 0.053
3 5.682 5.704 5.725 5.661 5.721 5.6986 0.0270 0.064
4 5.675 5.648 5.622 5.669 5.585 5.6398 0.0370 0.090
5 5.690 5.636 5.715 5.694 5.709 5.6888 0.0313 0.079
6 5.641 5.571 5.600 5.665 5.676 5.6306 0.0443 0.105
7 5.503 5.601 5.706 5.624 5.620 5.6108 0.0725 0.203
8 5.669 5.589 5.606 5.685 5.556 5.6210 0.0545 0.129
9 5.668 5.749 5.762 5.778 5.672 5.7258 0.0520 0.110
10 5.693 5.690 5.666 5.563 5.668 5.6560 0.0534 0.130
11 5.449 5.464 5.732 5.619 5.673 5.5874 0.1261 0.283
12 5.763 5.704 5.656 5.778 5.703 5.7208 0.0496 0.122
13 5.679 5.810 5.608 5.635 5.577 5.6618 0.0909 0.233
14 5.389 5.916 5.985 5.580 5.935 5.7610 0.2625 0.596
15 5.747 6.188 5.615 5.622 5.510 5.7364 0.2661 0.678
16 5.768 5.153 5.528 5.700 6.131 5.6560 0.3569 0.978
17 5.688 5.481 6.058 5.940 5.059 5.6452 0.3968 0.999
18 6.065 6.282 6.097 5.948 5.624 6.0032 0.2435 0.658
19 5.463 5.876 5.905 5.801 5.847 5.7784 0.1804 0.442
20 5.682 5.475 6.144 6.260 6.760 6.0642 0.5055 1.285
x
Appendix B Data Set: Constructing Control Charts for Boston Rainfall. In Exercises
1416, refer to the daily amounts of rainfall in Boston for one year, as listed in Data Set
10 in Appendix B. Omit the last entry for Wednesday so that each day of the week has ex-
actly 52 values.
14. Boston Rainfall: Constructing a Run Chart Using only the 52 rainfall amounts for
Monday, construct a run chart. Does the process appear to be within statistical control?
15. Boston Rainfall: Constructing an R Chart Using the 52 samples of seven values each,
construct an R chart and determine whether the process variation is within statistical
control. If it is not, identify which of the three out-of-control criteria lead to rejection
of statistically stable variation.
16. Boston Rainfall: Constructing an Chart Using the 52 samples of seven values each,
construct an chart and determine whether the process mean is within statistical con-
trol. If it is not, identify which of the three out-of-control criteria lead to rejection of a
statistically stable mean. If not, what can be done to bring the process within statisti-
cal control?
14-2 BEYOND THE BASICS
17. Constructing an s Chart In this section we described control charts for R and based
on ranges. Control charts for monitoring variation and center (mean) can also be
based on standard deviations. An s chart for monitoring variation is made by plotting
sample standard deviations with a centerline at (the mean of the sample standard de-
viations) and control limits at B
4
and B
3
, where B
4
and B
3
are found in Table 14-2.
Construct an s chart for the data of Table 14-1. Compare the result to the R chart given
in this section.
s s
s
x
x
x
5014_TriolaE/S_CH14pp732-757 11/22/05 6:17 AM Page 747
748 Chapter 14 Statistical Process Control
18. Constructing an Chart Based on Standard Deviations An chart based on standard
deviations (instead of ranges) is made by plotting sample means with a centerline at
and control limits at 1A
3
and 2A
3
, where A
3
is found in Table 14-2 and is the
mean of the sample standard deviations. Use the data in Table 14-1 to construct an
chart based on standard deviations. Compare the result to the chart based on sample
ranges (shown in this section).
14-3 Control Charts for Attributes
Key Concept This section presents a method for constructing a control chart to
monitor the proportion p for some attribute, such as whether a service or manu-
factured item is defective or nonconforming. (Agood or a service is nonconform-
ing if it doesnt meet specifications or requirements; nonconforming goods are
sometimes discarded, repaired, or called seconds and sold at reduced prices.)
The control chart is interpreted by using the same three criteria from Section 14-2
to determine whether the process is statistically stable.
Section 14-2 discussed control charts for quantitative data, but this section de-
scribes the construction of control charts for qualitative data. As in Section 14-2,
we select samples of size n at regular time intervals and plot points in a sequential
graph with a centerline and control limits. (There are ways to deal with samples of
different sizes, but we dont consider them here.)
x
x
s s x s x
x
x x
Definition
Acontrol chart for p (or p chart) is a graph of proportions plotted sequen-
tially over time, and it includes a centerline, a lower control limit (LCL), and
an upper control limit (UCL).
The notation and control chart values are as follows (where the attribute of
defective can be replaced by any other relevant attribute).
Notation
5pooled estimate of the proportion of defective items in the process
5pooled estimate of the proportion of process items that are not defective
n 5size of each sample (not the number of samples)
5 1 2 p
q
5
total number of defects found among all items sampled
total number of items sampled
p
Quality Control
at Perstorp
Perstorp Components, Inc. uses
a computer that automatically
generates control charts to
monitor the thicknesses of the
floor insulation the company
makes for Ford Rangers and
Jeep Grand Cherokees. The
$20,000 cost of the computer
was offset by a first-year
savings of $40,000 in labor,
which had been used to manu-
ally generate control charts to
ensure that insulation thick-
nesses were between the
specifications of 2.912 mm
and 2.988 mm. Through the
use of control charts and other
quality-control methods,
Perstorp reduced its waste by
more than two-thirds.
5014_TriolaE/S_CH14pp732-757 11/22/05 6:17 AM Page 748
14-3 Control Charts for Attributes 749
We use for the centerline because it is the best estimate of the proportion of
defects from the process. The expressions for the control limits correspond to
99.7% confidence interval limits as described in Section 7-2.
p
Control Chart for p
Centerline:
Upper control limit:
Lower control limit:
(If the calculation for the lower control limit results in a negative value, use 0
instead. If the calculation for the upper control limit exceeds 1, use 1 instead.)
p 2 3

p q
n
p 1 3

p q
n
p
EXAMPLE Defective Aircraft Altimeters The Chapter
Problem describes the process of manufacturing aircraft altimeters.
Section 14-2 includes examples of control charts for monitoring the
errors in altimeter readings. An altimeter is considered to be defective if it can-
not be calibrated or corrected to give accurate readings (within 20 ft of the true
altitude). The Altigauge Manufacturing Company produces altimeters in
batches of 100, and each altimeter is tested and determined to be acceptable or
defective. Listed below are the numbers of defective altimeters in successive
batches of 100. Construct a control chart for the proportion p of defective
altimeters and determine whether the process is within statistical control. If
not, identify which of the three out-of-control criteria apply.
Defects: 2 0 1 3 1 2 2 4 3 5 3 7
SOLUTION The centerline for the control chart is located by the value of
Because 5 0.0275, it follows that 5 12 5 0.9725. Using 5 0.0275,
50.9725, and n 5100, we find the control limits as follows:
Upper control limit:
Lower control limit:
p 2 3

p q
n
5 0.0275 2 3

s0.0275ds0.9725d
100
5 20.0216
p 1 3

p q
n
5 0.0275 1 3

s0.0275ds0.9725d
100
5 0.0766
q
p p q p
5
2 1 0 1 1 1
# # #
1 7
12 ? 100
5
33
1200
5 0.0275
p 5
total number of defects from all samples combined
total number of altimeters sampled
p:
continued
Six Sigma
in Industry
Six Sigma is the term used in
industry to describe a process
that results in a rate of no
more than 3.4 defects out of a
million. The reference to Six
Sigma suggests six standard
deviations away from the
center of a normal distribu-
tion, but the assumption of a
perfectly stable process is
replaced with the assumption
of a process that drifts
slightly, so the defect rate is
no more than 3 or 4 defects
per million.
Started around 1985 at
Motorola, Six Sigma pro-
grams now attempt to
improve quality and increase
profits by reducing variation
in processes. Motorola saved
more than $940 million in
three years. Allied Signal
reported a savings of $1.5
billion. GE, Polaroid, Ford,
Honeywell, Sony, and Texas
Instruments are other major
companies that have adopted
the Six Sigma goal.
5014_TriolaE/S_CH14pp732-757 11/22/05 6:17 AM Page 749
750 Chapter 14 Statistical Process Control
Because the lower control limit is less than 0, we use 0 instead. Having found
the values for the centerline and control limits, we can proceed to plot the pro-
portions of defective altimeters. The Minitab control chart for p is shown in the
accompanying display.
INTERPRETATION We can interpret the control chart for p by considering
the three out-of-control criteria listed in Section 14-2. Using those criteria, we
conclude that this process is out of statistical control for this reason: There ap-
pears to be an upward trend. The company should take immediate action to
correct the increasing proportion of defects.
Using Technology
MINITAB Enter the numbers of defects
(or items with any particular attribute) in
column C1. Select the option Stat, then
Control Charts, Attributes Charts, then
P. Enter C1 in the box identified as variable,
and enter the size of the samples in the box
identified as subgroup size, then click OK.
EXCEL Using DDXL: To use the
DDXL add-in, begin by entering the num-
bers of defects or successes in column A,
and enter the sample sizes in column B. For
the example of this section, the first three
items would be entered in the Excel spread-
sheet as shown below.
A B
1 2 100
2 0 100
3 1 100
Click on DDXL, select Process Control,
then select Summ Prop Control Chart (for
summary proportions control chart). A dia-
log box should appear. Click on the pencil
icon for Success Variable and enter the
range of values for column A, such as
A1:A12. Click on the pencil icon for Totals
Variable and enter the range of values for
column B, such as B1:B12. Click OK. Next
click on the Open Control Chart bar and
the control chart will be displayed.
Using Excels Chart Wizard: Enter the sam-
ple proportions in column A. Click on the
Chart Wizard icon, which looks like a bar
graph. For the chart type, select Line. For
the chart subtype, select the first graph in the
second row, then click Next. Continue to
click Next, then Finish. The graph can be
edited to include labels, delete grid lines,
and so on. You can insert the required cen-
terline and upper and lower control limits by
editing the graph. Click on the line on the
bottom of the screen, then click and drag to
position the line correctly.
STATISTICS
IN THE NEWS
High Cost of Low
Quality
The Federal Drug Adminis-
tration recently reached an
agreement whereby a phar-
maceutical company, the
Schering-Plough Corpora-
tion, would pay a record
$500 million for failure to
correct problems in manu-
facturing drugs. According
to a New York Times article
by Melody Petersen, Some
of the problems relate to the
lack of controls that would
identify faulty medicines,
while others stem from out-
dated equipment. They
involve some 200 medicines,
including Claritin, the al-
lergy medicine that is Scher-
ings top-selling product.
Minitab
5014_TriolaE/S_CH14pp732-757 11/22/05 6:17 AM Page 750
14-3 Control Charts for Attributes 751
14-3 BASIC SKILLS AND CONCEPTS
Statistical Literacy and Critical Thinking
1. p Chart What is a p chart, and what is its purpose?
2. Contract Specification The Paper Chase office supply company requires a supplier of
pens to maintain a production process with a defect rate less than 2%. If the supplier
uses a p chart to determine that the manufacturing process is within statistical control,
does that indicate that the defect rate is less than 2%?
3. Control Limits Using the methods of this section, a process is analyzed and the upper
and lower control limits are found to be 0.250 and 20.050 respectively. What upper
and lower control limits are used to construct the control chart?
4. Interpreting a Control Chart When monitoring the process of producing altimeters, a
company finds that the process is out of statistical control because there is a down-
ward pattern of defects that is not random. Should the downward pattern be cor-
rected? What should the company do?
Determining Whether a Process Is in Control. In Exercises 58, examine the given con-
trol chart for p and determine whether the process is within statistical control. If it is not,
identify which of the three out-of-control criteria apply.
5. 6.
7. 8.
Constructing Control Charts for p. In Exercises 912, use the given process data to con-
struct a control chart for p. In each case, use the three out-of-control criteria listed in
Section 14-2 and determine whether the process is within statistical control. If it is not,
identify which of the three out-of-control criteria apply.
9. p Chart for Birth Rate In each of 10 consecutive and recent years, 10,000 people
were randomly selected and the numbers of births they generated were found, with
the results given below (based on data from the National Center for Health Statistics).
How might the results be explained?
Births: 155 152 148 147 145 146 145 144 141 139
10. p Chart for Divorce Rate In each of 10 consecutive and recent years, 10,000 people
were randomly selected and the numbers of divorces were found, with the results
given below (based on data from the National Center for Health Statistics). How
might the results be explained?
Divorces: 48 46 46 44 43 43 42 41 42 40
5014_TriolaE/S_CH14pp732-757 11/22/05 6:17 AM Page 751
752 Chapter 14 Statistical Process Control
11. p Chart for Defective Car Batteries Defective car batteries are a nuisance because
they can strand and inconvenience drivers. A car battery is considered to be defec-
tive if it fails before its warranty expires. Defects are identified when the batteries
are returned under the warranty program. The Powerco Battery corporation manu-
factures car batteries in batches of 1000 and the numbers of defects are listed
below for each of 12 consecutive batches. Does the manufacturing process require
correction?
Defects: 8 6 5 9 10 7 7 4 6 11 5 8
12. Polling When the Infopop polling organization conducts a telephone survey, a call is
considered to be a defect if the respondent is unavailable or refuses to answer ques-
tions. For one particular poll about consumer preferences, 200 people are called each
day, and the numbers of defects are listed below. Does the calling process require cor-
rective action?
Defects: 92 83 85 87 98 108 96 115 121 125 112 127 109 131 130
14-3 BEYOND THE BASICS
13. p Chart for Boston Rainfall Refer to the Boston rainfall amounts in Data Set 10 in
Appendix B. For each of the 52 weeks, let the sample proportion be the proportion of
days that it rained. (Delete the 53rd value for Wednesday.) In the first week, for exam-
ple, the sample proportion is 3 7 5 0.429. Do the data represent a statistically stable
process?
14. Constructing an np Chart A variation of the control chart for p is the np chart in
which the actual numbers of defects are plotted instead of the proportions of defects.
The np chart will have a centerline value of , and the control limits will have values
of 1 3 and 2 3 . The p chart and the np chart differ only in the
scale of values used for the vertical axis. Construct the np chart for the example given
in this section. Compare the result with the control chart for p given in this section.
Review
In Chapter 2 we identified important characteristics of data: center, variation, distribution,
outliers, and changing pattern of data over time. The focus of this chapter is the changing
pattern of data over time. Process data were defined to be data arranged according to some
time sequence, and such data can be analyzed with run charts and control charts. Control
charts have a centerline, an upper control limit, and a lower control limit. Aprocess is sta-
tistically stable (or within statistical control) if it has only natural variation with no pat-
terns, cycles, or unusual points. Decisions about statistical stability are based on how a
process is actually behaving, not how we might like it to behave because of such factors as
manufacturer specifications. The following graphs were described:
Run chart: a sequential plot of individual data values over time
R chart: a control chart that uses ranges in an attempt to monitor the variation in a
process
chart: a control chart used to determine whether the process mean is within sta-
tistical control
p chart: a control chart used to monitor the proportion of some process attribute,
such as whether items are defective
x
!np q np !np q np
np
>
5014_TriolaE/S_CH14pp732-757 11/22/05 6:17 AM Page 752
Review Exercises 753
Statistical Literacy and Critical Thinking
1. Pattern over Time Why is it important to monitor a changing pattern of data over time?
2. Statistical Process Control The title of this chapter is Statistical Process Control.
What does that mean?
3. Manufacturing Run Amok What would be a possible adverse consequence of a Pepsi
bottling plant running a process that is not monitored?
4. Control Charts When monitoring the times it takes technicians to repair computers,
why is it important to use an chart and an R chart together?
Review Exercises
Constructing Control Charts for Consumption of Electricity. The following table lists
amounts of electrical consumption (in kWh) for the authors home, as given in Data Set 9
in Appendix B. Use the data for Exercises 13.
Electrical Consumption (kWh)
Year 1: 1st Half 3375 2661 2073
Year 1: 2nd Half 2579 2858 2296
Year 2: 1st Half 2812 2433 2266
Year 2: 2nd Half 3128 3286 2749
Year 3: 1st Half 3427 578 3792
1. Run Chart Construct a run chart for the 15 values. Does there appear to be a pattern
suggesting that the process is not within statistical control?
2. R Chart Using subgroups of size n 5 3 corresponding to the rows of the table, con-
struct an R chart and determine whether the process variation is within statistical con-
trol. If it is not, identify which of the three out-of-control criteria lead to rejection of
statistically stable variation.
3. Chart Using subgroups of size n 5 3 corresponding to the rows of the table, con-
struct an chart and determine whether the process mean is within statistical control.
Does the process appear to be statistically stable? If it is not, identify which of the
three out-of-control criteria lead to rejection of statistically stable variation.
4. Constructing a Control Chart for Infectious Diseases In each of 13 consecutive and
recent years, 100,000 adults 65 years of age or older were randomly selected and
the number who died from infectious diseases is recorded, with the results given
below (based on data from Trends in Infectious Diseases Mortality in the United
States, by Pinner et al., Journal of the American Medical Association, Vol. 275,
No. 3). Construct an appropriate control chart and determine whether the process is
within statistical control. If not, identify which criteria lead to rejection of statistical
stability.
Number who died: 270 264 250 278 302 334 348 347 377 357 362 351 343
5. Control Chart for Defects The Medassist Pharmaceutical Company manufactures
aspirin tablets. Each day, 100 tablets are randomly selected and tested. A tablet is
considered defective if it has obvious physical deformities or the aspirin content is
less than 490 mg or greater than 510 mg. The numbers of defects are listed below for
x
x
x
5014_TriolaE/S_CH14pp732-757 1/19/07 11:08 AM Page 753
Cooperative Group Activities
754 Chapter 14 Statistical Process Control
consecutive days. Construct an appropriate control chart and determine whether the
process is within statistical control. If not, identify which criteria lead to rejection of
statistical stability.
Defects: 4 2 2 3 5 2 9 12 1 11 3 2 12 14
Cumulative Review Exercises
1. Control Chart for Defective Seat Belts The Flint Accessory Corporation manufac-
tures seat belts for cars. Federal specifications require that the webbing must have a
breaking strength of at least 5000 lb. During each week of production, 200 belts are
randomly selected and tested for breaking strength, and a belt is considered defective
if it breaks before reaching the force of 5000 lb. The numbers of defects are listed be-
low for a sequence of 10 weeks. Use a control chart for p to verify that the process is
within statistical control. If it is not in control, explain why it is not.
6 4 12 3 7 2 3 5 4 2
2. Confidence Interval for Defective Seat Belts Refer to the data in Exercise 1 and, us-
ing all of the data from the 2000 seat belts that were tested, construct a 95% confi-
dence interval for the proportion of defects.
3. Hypothesis Test for Defective Seat Belts Refer to the data in Exercise 1 and, using all
of the data from the 2000 seat belts that were tested, use a 0.05 significance level to
test the claim that the rate of defects is greater than 1%.
4. Using Probability in Control Charts When interpreting control charts, one of the
three out-of-control criteria is that there are eight consecutive points all above or all
below the centerline. For a statistically stable process, there is a 0.5 probability that a
point will be above the centerline and there is a 0.5 probability that a point will be be-
low the centerline. In each of the following, assume that sample values are indepen-
dent and the process is statistically stable.
a. Find the probability that when eight consecutive points are randomly selected, they
are all above the centerline.
b. Find the probability that when eight consecutive points are randomly selected, they
are all below the centerline.
c. Find the probability that when eight consecutive points are randomly selected, they
are all above or all below the centerline.
1. Out-of-class activity Collect your own process data
and analyze them using the methods of this section. It
would be ideal to collect data from a real manufactur-
ing process, but that may be difficult to accomplish. If
so, consider using a simulation or referring to published
data, such as those found in an almanac. Here are some
suggestions:
Shoot five basketball foul shots (or shoot five crum-
pled sheets of paper into a wastebasket) and record
the number of shots made; then repeat this proce-
dure 20 times, and use a p chart to test for statistical
stability in the proportion of shots made.
Your pulse rate can be measured by counting the
number of times your heart beats in 1 min. Measure
your pulse rate four times each hour for several
hours, then construct appropriate control charts.
What factors contribute to random variation?
Assignable variation?
Go through newspapers for the past 12 weeks and
record the closing of the Dow Jones Industrial
5014_TriolaE/S_CH14pp732-757 11/22/05 6:17 AM Page 754
Technology Project 755
Technology Project
a. Simulate the following process for 20 days: Each
day, 200 calculators are manufactured with a 5%
rate of defects, and the proportion of defects is
recorded for each of the 20 days. The calculators for
one day are simulated by randomly generating 200
numbers, where each number is between 1 and 100.
Consider an outcome of 1, 2, 3, 4, or 5 to be a defect,
with 6 through 100 being acceptable. This corre-
sponds to a 5% rate of defects. (See the technology
instructions below.)
b. Construct a p chart for the proportion of defective
calculators, and determine whether the process is
within statistical control. Since we know the process
is actually stable with p 5 0.05, the conclusion that
it is not stable would be a type I error; that is, we
would have a false positive signal, causing us to be-
lieve that the process needed to be adjusted when in
fact it should be left alone.
c. The result from part(a) is a simulation of 20 days.
Now simulate another 10 days of manufacturing cal-
culators, but modify these last 10 days so that the de-
fect rate is 10% instead of 5%.
d. Combine the data generated from parts(a) and (c) to
represent a total of 30 days of sample results. Con-
struct a p chart for this combined data set. Is the pro-
cess out of control? If we concluded that the process
was not out of control, we would be making a type II
error; that is, we would believe that the process was
okay when in fact it should be repaired or adjusted
to correct the shift to the 10% rate of defects.
Technology Instructions for Part(a):
STATDISK Select Data, Uniform Generator, and pro-
ceed to generate 200 values with a minimum of 1 and a
maximum of 100. Copy the data to the data window, then
sort the values using the Data Tools button. Repeat this pro-
cedure until results for 20 days have been simulated.
MINITAB Select Calc, Random Data, then Integer.
Enter 200 for the number of rows of data, enter C1 as the
column to be used for storing the data, enter 1 for the mini-
mum value, and enter 100 for the maximum value. Repeat
this procedure until results for 20 days have been simulated.
EXCEL Click on the fx icon on the main menu bar, then
select the function category Math & Trig, followed by
RANDBETWEEN. In the dialog box, enter 1 for bottom
and 100 for top. A random value should appear in the first
row of column A. Use the mouse to click and drag the lower
right corner of that cell, then pull down the cell to cover the
first 200 rows of column A. When you release the mouse
button, column Ashould contain 200 random numbers. You
can also click drag the lower right corner of the bottom cell
by moving the mouse to the right so that you get 20 columns
of 200 numbers each. The different columns represent the
different days of manufacturing.
TI-83/84 PLUS Press the MATH key. Select PRB,
then select the 5th menu item, randInt(, and proceed to en-
ter 1, 100, 200; then press the ENTER key. Press STO and
L1 to store the data in list L1. After recording the number of
defects, repeat this procedure until results for 20 days have
been simulated.
>
Average (DJIA) for each business day. Use run and
control charts to explore the statistical stability of
the DJIA. Identify at least one practical consequence
of having this process statistically stable, and iden-
tify at least one practical consequence of having this
process out of statistical control.
Find the marriage rate per 10,000 population for
several years. (See the Information Please Almanac
or the Statistical Abstract of the United States.) As-
sume that in each year 10,000 people were randomly
selected and surveyed to determine whether they
were married. Use a p chart to test for statistical sta-
bility of the marriage rate. (Other possible rates:
death, accident fatality, crime.)
Obtain a printed copy of computer results, and write a
report summarizing your conclusions.
2. In-class activity If the instructor can distribute the
numbers of absences for each class meeting, groups of
three or four students can analyze them for statistical
stability and make recommendations based on the con-
clusions.
3. Out-of-class activity Conduct research to identify
Demings funnel experiment, then use a funnel and mar-
bles to collect data for the different rules for adjusting
the funnel location. Construct appropriate control
charts for the different rules of funnel adjustment. What
does the funnel experiment illustrate? What do you
conclude?
5014_TriolaE/S_CH14pp732-757 11/22/05 6:17 AM Page 755
756 Chapter 14 Statistical Process Control
FromData to Decision
Critical Thinking: Are the axial
loads within statistical control?
Is the process of manufacturing
cans proceeding as it should?
Exercises 9 and 10 in Section 14-2 used pro-
cess data from a New York company that
manufactures 0.0109-in. thick aluminum
cans for a major beverage supplier. Refer to
Data Set 15 in Appendix B and conduct an
analysis of the process data for the cans that
are 0.0111 in. thick. The values in the data
set are the measured axial loads of cans, and
the top lids are pressed into place with pres-
sures that vary between 158 lb and 165 lb.
Analyzing the Results
Based on the given process data, should the
company take any corrective action? Write
a report summarizing your conclusions.
Address not only the issue of statistical sta-
bility, but also the ability of the cans to with-
stand the pressures applied when the top lids
are pressed into place. Also compare the
behavior of the 0.0111-in. cans to the behav-
ior of the 0.0109-in. cans and recommend
which thickness should be used.
Control Charts
This chapter introduces different charting tech-
niques used to summarize and study data asso-
ciated with a process along with methods for
analyzing the stability of that process. With the
exception of the run chart, individual data
points are not needed to construct a chart. For
example, the R chart is constructed from sample
ranges while the p chart is based on sample
proportions. This is an important point, as data
collected from third-party sources are often
given in terms of summarizing statistics. Go to
the Elementary Statistics Web site:
http://www.aw.com/Triola.
Locate the Internet Project dealing with control
charts. There you will be directed to data sets
and sources of data for use in constructing con-
trol charts. From the resulting charts you will be
asked to interpret and discuss trends in the
underlying processes.
Internet Project
5014_TriolaE/S_CH14pp732-757 11/22/05 6:18 AM Page 756
Statistics @ Work 757
There is a certain
amount of respect that
is given to someone who
knows statistics and
can explain it to
someone who doesnt
know it.
Dan OToole
Account Executive: A. C. Nielsen
In his work in the Advanced Ana-
lytics Group at A. C. Nielsen, Dan
develops statistical solutions to
help clients like Polaroid, Ocean
Spray, and Gillette understand
which of their marketing vehicles
drives sales most profitably. Dan
has a Masters Degree in Business
Economics from Bentley College.
Statistics @ Work
What concepts of statistics do
you use?
I have worked with analyses as simple as
correlation and general significance
tests, all the way to multiple regression,
factor analysis, correspondence analysis,
and cluster analysis.
How do you use statistics
on the job?
My job is to discover or uncover client is-
sues, and then find out if we can apply
one of our statistical techniques to their
specific issue. If a technique wont help a
client, then you need to know that. An
example of how I use stats: A client may
say, I sell product X, whether it is juice,
bread, or a camera. Right now, they may
control 20% of the market. They may
come to us to see if they can increase
market share by lowering their price. My
job would be to design a study to ana-
lyze this question. To do this I have to
design a study that will take into account
everything that affects the sales of a
product. Using techniques like regres-
sion, if I am able to create a model with
good significance, I will be able to isolate
specific influences on the sales and offer
recommendations. Things like seasonal-
ity distribution as well as any marketing
efforts that may have taken place, must
be included. In addition, I will have to
take into account complementary prod-
ucts price (butter is complementary to
bread; while film is for a camera) and
also competitive products. For instance,
bread may compete with English muffins
(I know it does for me).
Do you feel job applicants are viewed
more favorably if they have studied
some statistics?
By far. There is a certain amount of re-
spect that is given to someone who
knows statistics and can explain it to
someone who doesnt know it (because it
means you really know it and arent recit-
ing from a textbook). Almost every job
uses statistics (particularly correlations
and regressions). People will say things
like, Oh, check if theyre correlated.
Is your use of probability and statis-
tics increasing, decreasing, or re-
maining stable?
It definitely is increasing. In this busi-
ness (consulting), you are constantly
challenged to learn a new technique or
look at an old technique in order to
improve it. In addition, since we are
constantly coming out with new prod-
ucts, our understanding of statistics has
to increase to use these techniques
effectively.
How beneficial do you find your
knowledge of statistics for perform-
ing your responsibilities?
It is not a question of beneficial, but
rather it is a necessity. In fact, we find
that we have to know it so well, so that
we can explain it in laymans terms to
our clients.
5014_TriolaE/S_CH14pp732-757 11/25/05 8:37 AM Page 757