EM 8733-E • June 1999

PERFORMANCE EXCELLENCE $3.00

IN THE WOOD PRODUCTS INDUSTRY

Part 2:
How and Why SPC Works
S. Leavengood and J. Reeb

Part 1 in this series introduced Statistical Process Control (SPC) by

discussing the history, philosophy, and benefits of SPC, by suggesting
how to successfully implement a new SPC program, and by attempting
to alleviate fears of the math associated with SPC.
The goal for Part 2 is to build management understanding and
confidence in SPC as a profit-making tool. It is unreasonable to expect
managers to commit to and support SPC training and implementation
if they do not understand what SPC is and how and why it works. This
publication discusses the importance of understanding and quantifying
process variation and describes how and why SPC works.
Part 3 in the series is a step-by-step approach to building the skills
required to implement SPC. Later publications in the series will
present case histories of SPC use in wood products firms, examining
pitfalls and successful approaches and providing real-world evidence
of SPC’s benefits as a process improvement tool.

Variation—it’s everywhere
Variation is a fact of life. It is everywhere, and it is unavoidable.
Even a brand-new, state-of-the-art machine cannot hold perfectly to the
target setting; there always is some fluctuation around the target.
Attaining consistent product quality requires understanding, monitor-
ing, and controlling variation. Attaining optimal product quality
requires a never-ending commitment to reducing variation.
Where does variation come from? Walter Shewhart, the man whose
work laid the foundations for SPC, recognized that variation has two
broad causes: common (also called chance, random, or unknown)
causes and special (also called assignable) causes.
Common causes of variation are inherent in the process and can be
thought of as the “natural rhythm of the process.” Common causes are

Scott Leavengood, Extension agent, Klamath

County; and James E. Reeb, Extension forest
products manufacturing specialist; Oregon State
University.
STATISTICAL PROCESS CONTROL

evidenced by a stable, repeating pattern of variation. Real quality

improvement requires a continual focus on reducing common-
cause variation.
Special causes of variation are a signal that something has
changed in the process. Special causes are evidenced by a disrup-
tion of the stable, repeating pattern of variation. Special causes of
variation result in unpredictable process performance and must
therefore be identified and removed before taking other steps to
improve quality.
Why is it important to distinguish between these two types of
variation? Because the remedies are completely different. Under-
standing the difference between the two types of variation helps
manufacturers to target quality improvement efforts correctly and
thereby avoid wasted effort and expense.

SPC in a nutshell
Montgomery (1997) defines SPC as “...a powerful collection of
problem-solving tools useful in achieving process stability and
Real quality improving capability through the reduction of variability.” Control
improvement … charts and process capability analysis are the two primary tools of
SPC. Other tools such as histograms, flow charts, cause-and-effect
requires a diagrams, check sheets, and Pareto diagrams also are useful in
continual focus on quality and process improvement.
reducing common- In discussing how and why SPC works, we will:
cause variation. • Describe the distribution of the process
• Estimate the limits within which the process operates under
“normal” conditions
• Determine whether the process is stable
• Continue to monitor and control the process
• Compare process performance to specifications, and
• See how to continuously improve the process

The distribution of the process

In SPC, when we talk about the distribution of the process, we
are referring not to the process itself but to data collected from the
process—for example, data on widths of pieces coming out of a
woodworking machine—and to the way those data are distributed
when plotted on a chart or graph.
Describing the distribution of a process is analogous to evaluat-
ing your marksmanship. How’s your aim? Are you accurate—in
other words, are you on target? Is your aim precise; that is, are all

2
 HOW AND WHY SPC WORKS

Neither precise nor accurate Accurate but not precise

Precise but not accurate Precise and accurate

Figure 1.—Precision and accuracy (adapted from Montgomery, 1997).

the shots clustered around the target, or are they distributed all over
the place (Figure 1)? In manufacturing, the questions are:
• Where is the process centered; in other words, is the process on
or off target and, if it’s the latter, by how much is it off target?
• How much does the process fluctuate about the center?

Histograms
Histograms are visual tools to examine distributions. A histo-
gram is a bar graph that shows how frequently data fall within
specific cells, that is, ranges of values. Histograms make it rela-
tively simple to estimate where the process is centered and how
much fluctuation there is about the center.
Table 1 (Page 4) shows measurements of widths (in inches) of
125 wood components produced by a woodworking machine.
We must have data to know how a process is performing. How-
ever, it is difficult to derive much information from data as pre-
sented in Table 1. The data would provide more information if they
were grouped, organized, and displayed graphically. A histogram
does just that.
We will leave the detailed discussion of how to create histo-
grams for a future publication. For our purposes here, we will

3
STATISTICAL PROCESS CONTROL

simply state that creating a histogram involves developing and

plotting a frequency distribution. A frequency distribution is a tally
of measurements within specific cells. For example, in Table 1
there are 15 measurements in the 2.5395–2.5404 cell, 21 measure-
ments in the 2.5405–2.5414 cell, and so on. The frequency distri-
bution is shown in Table 2, and the histogram is shown in Figure 2.
What does this histogram tell us? First, we can easily estimate
where the process is centered and the amount of spread about the
center. The center of this distribution is approximately 2.542
inches. We can see that the majority of the measurements are
between 2.540 and 2.544 inches.
Knowing the specifications enables us to get more information
from the histogram. If, for example, we knew the specifications
were 2.542 inches ± 0.008 inch, we would say the process was
performing quite well. If, on the other hand, the specifications
were 2.544 inches ± 0.002 inch, the histogram shows that a sub-
stantial amount of material is being produced below the lower
specification and thus is defective.
A histogram is a snapshot of the process. It is useful for examin-
ing the status (centering and spread) of the process at the time the
data were collected and for examining the general shape (for
example, number of peaks and symmetry) of the distribution.
A single histogram, however, does not allow us to evaluate
process performance through time nor does it allow us to deter-
mine whether the process was stable (consistent and predictable

Table 1.—Sample data.

2.541 2.542 2.541 2.542 2.543 2.544 2.539 2.542 2.545 2.543 2.543
2.540 2.544 2.539 2.540 2.542 2.540 2.545 2.543 2.540 2.543 2.541
2.542 2.542 2.541 2.543 2.543 2.543 2.543 2.543 2.543 2.540 2.544
2.541 2.541 2.544 2.544 2.544 2.543 2.542 2.546 2.542 2.542 2.543
2.543 2.537 2.542 2.541 2.540 2.542 2.542 2.542 2.543 2.544 2.541
2.543 2.538 2.543 2.541 2.541 2.541 2.545 2.545 2.543 2.543
2.541 2.540 2.542 2.542 2.541 2.541 2.542 2.540 2.542 2.540
2.541 2.541 2.544 2.543 2.543 2.541 2.542 2.542 2.542 2.539
2.543 2.541 2.543 2.540 2.543 2.542 2.544 2.540 2.542 2.542
2.543 2.543 2.543 2.543 2.543 2.542 2.545 2.541 2.540 2.542
2.541 2.542 2.543 2.540 2.542 2.543 2.539 2.542 2.543 2.542
2.541 2.542 2.540 2.540 2.542 2.542 2.542 2.542 2.544 2.542

4
 HOW AND WHY SPC WORKS

Table 2.—Frequency distribution for data in Table 1.

Cell Cell A histogram. . .
boundaries midpoint Frequency
is a useful snapshot of
2.5365 – 2.5374 2.537 1
the process, but it
2.5375 – 2.5384 2.538 1
2.5385 – 2.5394 2.539 4 doesn’t let us evaluate
2.5395 – 2.5404 2.540 15 the process over time
2.5405 – 2.5414 2.541 21 or determine whether
2.5415 – 2.5424 2.542 35 the process is stable.
2.5425 – 2.5434 2.543 32
2.5435 – 2.5444 2.544 10
2.5455 – 5.5454 2.545 5
2.5455 – 2.5464 2.546 1

performance) when the data were collected. Also, it takes time to

collect the data (mathematicians suggest at least 50 to 100 data
points per histogram) which can become overwhelming if it has to
be done every day.
To be practical for day-to-day process control, we need a sys-
tem in which relatively small samples allow us to decide whether
the process is okay and therefore should be left alone, or whether
problems are beginning to arise and we should take action. SPC
provides such a system through the use of control charts.

35

30

25
Frequency

20

15

10

0
2.537 2.538 2.539 2.540 2.541 2.542 2.543 2.544 2.545 2.546
Widths (inches)
Figure 2.—Histogram for data in Table 1.

5
STATISTICAL PROCESS CONTROL

Control charts
Control charts are an SPC tool used to monitor and control
processes. There are charts for variables data (measurement data
such as length, width, thickness, and moisture content) and charts
for attributes data (“counts” data such as number of defective units
in a sample or number of errors on an invoice). We’ll focus here on
one type of variables control chart and will discuss the other kinds
in future publications.
In general, control charts are used as follows: samples are taken
Control limits… from the process, statistics (for example, average and range) are
tell us how far we can calculated and plotted on charts, and the results are interpreted
expect sample values with respect to process limits—or, as they are known in SPC
to stray from the terminology, control limits. Control limits are the limits within
average, given the which the process operates under normal conditions. They tell us
inherent variability of how far we can expect sample values to stray from the average
the process. given the inherent variability of the process—or, to use the SPC
terms, the magnitude of common-cause variation. Data points
beyond the control limits or other unusual patterns indicate special-
cause variation.

Estimating control limits

Calculating control limits requires only two numbers: an esti-
mate of central tendency (process centering), and an estimate of
process variation.
The average is our best estimate of central tendency. The aver-
age is widely used and is understandable to most people. For
example, golfers and bowlers routinely calculate averages from a
list of scores. To find an average, add all the sample measurements
and divide the sum by the sample size. As an example, let’s use the
first five sample measurements in Table 1:
2.541 + 2.540 + 2.542 + 2.541 + 2.543 = 12.707
× = 12.707 ÷ 5
× = 2.541 (rounded to three decimal places)

where × (pronounced “X-bar”) is the symbol used for the average.

In addition to a measure of central tendency, we need a measure

of variation. The values commonly used to quantify variation are
the standard deviation and the range.
SPC uses the range more often than the standard deviation
because calculating the standard deviation is fairly involved. Also,
the standard deviation usually is a less familiar concept for most

6
 HOW AND WHY SPC WORKS

people. The range (R), on the other hand, is simply the largest
value (Xmax) in the sample minus the smallest value (Xmin) in the
sample. For the five-sample measurements listed above, Xmax =
2.543 and Xmin = 2.540, and therefore:
R = 2.543 – 2.540 = 0.003
We now have the two values we need to calculate the control
limits—a measure of central tendency (the average, × ) and a
measure of variation (the range, R). Recall that control limits tell
us how far we can expect sample values to stray from the average,
given the magnitude of process variation. Therefore, the formula for
the control limits must account for both the average and the range. A
frequency distribution allows us to develop such a formula.

Frequency distributions and probability

Recall that Table 2 was the frequency distribution for the data in
Table 1. The science of statistics provides us with a number of
frequency distributions with known probabilities. In common
language, probabilities quantify the odds or likelihood of a specific
event. For example, the probability of getting the ace of spades in a
single draw from a deck of cards is 1/52 (1 of 52 possible out-
comes), or approximately 0.019. The probability of winning the
lottery is often on the order of 1 in 1,000,000 (0.000001). As you
can see, the smaller the probability, the less likely the event.
Statistics science also enables us to determine the probability
that something will not happen. For example, the probability of not
getting the ace of spades in a single draw from a deck of cards is
1 minus the fraction that indicates the probability of getting the ace
of spades, which is 1 minus 1⁄52 (or 0.019), which is 0.981. For
more information on probability, see OSU Extension publication
EM 8718, An Introduction to Models and Probability Concepts.
In SPC, we use known frequency distributions to establish con-
trol limits with known probabilities. Control limits are established
so that the probability of obtaining a value (that is, a result) beyond
the limits is very small unless the process changes significantly.
Therefore, control limits minimize false alarms—that is, searching
for problems when none exists. Searching for problems is often
expensive because it involves time, effort, and, in many cases,
equipment downtime. Minimizing false alarms and their expense is
a key benefit of SPC and is the prime reason the first book on SPC
was titled “Economic Control of Quality of Manufactured Product”
(Shewhart, 1931).

7
STATISTICAL PROCESS CONTROL

The normal distribution

The normal* distribution is a frequency distribution used exten-
sively in SPC. The normal distribution commonly is called the bell
curve due to the curve’s shape (Figure 3). Only two parameters are
needed to construct a normal distribution: the average and the
standard deviation. The average is designated with the Greek
lowercase letter µ (mu), and the standard deviation is designated
with the Greek lowercase letter σ (sigma). The normal curve’s
peak is at the average, µ, and its spread (the width of the curve) is
reflected by the standard deviation, σ. The parameters µ and σ
generally are unknown, and thus we have to estimate them by
sampling the process. We use × , the sample average, to estimate µ,
and we use R, the sample range, to estimate σ.
In SPC, the normal distribution allows us to determine the
probabilities of getting values beyond control limits. Probability
values for the normal distribution can be found in textbooks on
statistics and quality control. For purposes of our brief discussion
of normal distribution, we simply will state some common proba-
bilities.
The probability that a normally distributed variable will be
within plus or minus 1 standard deviation (1σ, pronounced “one
Frequency

µ – 3σ µ – 2σ µ – 1σ µ µ + 1σ µ + 2σ µ + 3σ
Figure 3.—The normal distribution.

*Note: “Normal” is the name for the distribution and is not to be

confused with common usage of the word “normal” as meaning
“standard” or “commonplace.”

8
 HOW AND WHY SPC WORKS

sigma”) of the average is approximately 0.68. That is, 68 percent

of the values will be within 1 standard deviation of the average.
The probability that a normally distributed variable will be within
±2σ of the average is approximately 0.95, and within ±3σ of the
average is approximately 0.997 (Figure 4).
Frequency

µ – 3σ µ – 2σ µ – 1σ µ µ + 1σ µ + 2σ µ + 3σ

68.0%
95.0%
99.7%

Figure 4.—The normal distribution, showing common probabilities.

Knowing that 99.7 percent of the values will fall within ±3σ of
the average, we can feel confident that a value beyond 3σ would
be highly unlikely unless a significant change (i.e., special-cause
variation) had occurred in the process.
For example, if process output is normally distributed, and the
average for our process is 2.542, and the standard deviation (σ) is
0.002, then the probability of a value between 2.536 and 2.548
[that is, 2.542 ± (3 x 0.002)] inches is 0.997. Conversely, the
probability of observing a value less than 2.536 or greater than
2.548 is 0.003 (1 – 0.997), or approximately 3 chances in 1,000.
Therefore, 2.536 and 2.548 would appear to make good control
limits because a value beyond the limits is statistically rare. There-
fore, it is very likely that special causes of variation are influencing
the process, and so searching for problems is likely to be profitable.
So, can we use 2.536 as the lower control limit and 2.548 for the
upper control limit? We could—if we were examining individual
items from the process. Recall, however, that we are sampling the

9
STATISTICAL PROCESS CONTROL

process and collecting statistics (in this example, the average) for
multiple items* rather than for individual items. Our control chart
therefore should reflect the distribution of averages, not of indi-
vidual data points.

Individuals vs. averages

The variation of averages is significantly less than the variation
of the individual values. Figure 5 demonstrates this point by over-
laying a histogram of averages of five measurements from the data
in Table 1 onto the histogram of individual values.
The larger the sample, the narrower the distribution of sample
averages. Therefore, control limits for averages will be narrower
(closer to the average) than control limits for individual measure-
ments. The values needed to calculate control limits have been
tabulated and are available in textbooks. To calculate control limits,
you need only look up in a table the value that corresponds to
sample size, and then perform simple multiplication and addition.

40

30
Frequency

20

10

0
2.537 2.538 2.539 2.540 2.541 2.542 2.543 2.544 2.545 2.546
Width (inches)
Individual values Averages of 5

Figure 5.—Histogram for individual values and for averages of 5 measurements.

*There are several reasons we prefer to sample multiple items

versus individual items from the process. One reason is that the
Central Limit Theorem allows us to use the normal distribution
regardless of the actual distribution of the process. We will discuss
this and the other reasons in detail in future publications.

10
 HOW AND WHY SPC WORKS

Control limits vs. specification limits

Control limits …
Before leaving the issue of control limits, we must address the
often-asked question, “Why can’t we simply use the specifications represent “what the
for control limits? Why all this bother with control limits based on process can do.”
statistical probabilities, 3σ, etc. when all we really want to know is Specification limits
how much ‘good’ (that is, ‘within the specifications’) product represent “what we
we’re producing?” There are two primary reasons never to use want the process to
specification limits as control limits. do.”
Control limits are established to minimize false alarms. Unless
some significant change has occurred in the process, a sample value
beyond a 3σ control limit is a statistical rarity and thus a signal that
special-cause variation is influencing the process. Searching for
problems is likely to be profitable. Therefore, control limits must be
established as a function of the capabilities of the process. In prac-
tice, specification limits usually are established by engineers and are
not a function of the capabilities of the process. Control limits
represent “what the process can do,” and specification limits repre-
sent “what we want the process to do.” To be useful for quality
control, limits must be based on what the process can do.
Another reason never to use specification limits as control limits
is that the former are for individual items, not for averages. As
noted earlier, the variation for averages is less than the variation
for individual values. Therefore, comparing a sample average to a
specification limit is the proverbial apples-to-oranges comparison.
False alarms (chasing problems that aren’t there) and failing to
detect problems are common when specifications instead of control
limits are used on control charts.

Determine whether the process is stable

How can we be sure that the data we use to establish control
limits are not simply a snapshot of an unstable process? If the
process is not stable, we are shooting at a moving target, and our
control limits are meaningless for long-term process control.
Therefore, we must find a way to determine whether the process is
stable. In SPC terminology, we ask, “Is the process in statistical
control?” Or, more simply, “Is it in control?”
In common usage, the phrase “in control” generally describes a
desirable situation, and it has the same connotation in SPC. On the
other hand, the phrase “out of control” brings forth images of
mayhem—machines on fire, nuts and bolts flying through the air,
etc. An out-of-control process, as the phrase is used in SPC, is not
quite so dramatic.

11
STATISTICAL PROCESS CONTROL

Deming (1982) defines control in SPC:

A stable process, one with no indication of a special cause
of variation, is said to be, following Shewhart, in statistical
control, or stable. It is a random process. Its behavior in the
near future is predictable.
Therefore, out of control can be defined as a process that is
under the influence of special causes of variation. Its behavior is
not predictable. Statistically rare occurrences are signals that a
process is out of control. When a process is out of control, we
should investigate to find and eliminate the special causes of
variation in order to bring the process back in control and thereby
improve consistency of product quality.
In our discussion of the normal distribution, above, we stated
that upper and lower control limits in SPC generally were set at
plus or minus 3 standard deviations (± 3σ) from the average.
Finding a normally distributed value beyond the 3σ limits has a
probability of approximately 0.003, which means it is statistically
rare. Therefore, we can state that a process is out of control if a
sample value falls outside the control limits.
There are other rules in SPC for determining whether a process
is in or out of control. Some companies simply use the “outside the
control limits” rule just mentioned; others use several rules that
involve detecting trends or “runs” of data points above and below
average. For example, eight consecutive data points on the chart on
the same side of the center line (the average) or six consecutive
points on the chart steadily increasing or decreasing may indicate a
change in the process and therefore an out-of-control situation.
These additional rules often help to detect problems sooner than
waiting for a point to fall outside the control limits. For this discus-
sion, we will focus simply on the outside-the-control-limits rule.
Other rules will be discussed in a future publication.
Now, we have most of the information we need to determine
whether the process is in control. However, for practical applica-
tion, one thing is lacking: a visual tool to help us evaluate sample
data and compare them to the control limits. For this purpose,
Shewhart created the control chart, also known as the Shewhart
Control Chart in honor of its inventor.
Control charts provide a graphical view of the process over time.
We will use the data in Table 1 to construct a control chart for the
average, known as an × chart. For purposes of the example, we
group the measurements in Table 1 into 25 samples with 5 mea-
surements per sample. The grouped data are shown in Table 3.

12
 HOW AND WHY SPC WORKS

Table 3. Data from Table 1 organized in samples of 5.

Sample Items in sample* Summary of sample
no. 1 2 3 4 5 × R
1 2.541 2.540 2.542 2.541 2.543 2.541 0.003
2 2.543 2.541 2.541 2.543 2.543 2.542 0.002
3 2.541 2.541 2.542 2.544 2.542 2.542 0.003
4 2.541 2.537 2.538 2.540 2.541 2.539 0.004
5 2.541 2.543 2.542 2.542 2.541 2.542 0.002
6 2.539 2.541 2.544 2.542 2.543 2.542 0.005
7 2.542 2.544 2.543 2.543 2.543 2.543 0.002
8 2.540 2.542 2.540 2.543 2.544 2.542 0.004
9 2.541 2.541 2.542 2.543 2.540 2.541 0.003
10 2.543 2.540 2.540 2.543 2.542 2.542 0.003
11 2.543 2.544 2.540 2.541 2.541 2.542 0.004
12 2.543 2.543 2.543 2.542 2.542 2.543 0.001
13 2.544 2.540 2.543 2.543 2.542 2.542 0.004
14 2.541 2.541 2.541 2.542 2.542 2.541 0.001
15 2.543 2.542 2.539 2.545 2.543 2.542 0.006
16 2.542 2.542 2.545 2.542 2.542 2.543 0.003
17 2.544 2.545 2.539 2.542 2.542 2.542 0.006
18 2.543 2.543 2.546 2.542 2.545 2.544 0.004
19 2.540 2.542 2.540 2.541 2.542 2.541 0.002
20 2.542 2.545 2.540 2.543 2.542 2.542 0.005
21 2.543 2.543 2.542 2.542 2.542 2.542 0.001
22 2.540 2.543 2.544 2.543 2.543 2.543 0.004
23 2.540 2.542 2.544 2.543 2.540 2.542 0.004
24 2.539 2.542 2.542 2.542 2.542 2.541 0.003
25 2.543 2.541 2.544 2.543 2.541 2.542 0.003
Averages for
all samples
× R
2.542 0.003

*In practice, measurements should be grouped in a manner

Shewhart called “rational subgrouping.” Sampling should be done
so that differences between subgroups (samples) are maximized
and differences within subgroups are minimized. We will discuss
this subject in depth in a future publication.

13
 STATISTICAL PROCESS CONTROL

To construct the control chart, we first calculate the average and

range for each sample. We then calculate × (pronounced “×-bar-
bar” or “× double bar”), which is the average of the 25 sample
averages, and we calculate R , which is the average of the 25
sample ranges.
The grand average, × , for the 25 sample averages is 2.542
inches. We draw a horizontal line across the center of the graph at
2.542 to represent the average. R is 0.003 inch. As discussed
previously, to calculate the upper and lower control limits (UCL
and LCL, respectively) we use formulas and table values from
textbooks. Using Table D in Appendix VI in Montgomery (1997),
the table value (A2) for a sample size of 5 is 0.577. The control
limits are then:
UCL = × + A2 R
= 2.542 + (0.577 x 0.003)
= 2.544
LCL = × – A2 R
= 2.542 – (0.577 x 0.003)
= 2.540

We now draw horizontal lines on the graph at the appropriate

locations for the UCL and LCL. Last, data points for each sample
average are plotted on the chart (Figure 6).

2.545
2.544 ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○
Average

2.543
2.542
2.541
2.540
2.539
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

○
UCL ○ ○ ○
LCL Data Average

Figure 6.— × control chart for data in Table 3.

Given that we are attempting to determine whether the process is

in or out of control, what can we learn from this chart? Recall that
a point outside the control limits or some other unusual patterns
indicate the process is out of control. Therefore, Sample 4 indicates
the process is out of control, and Sample 18 is questionable

14
 HOW AND WHY SPC WORKS

because it is on the upper control limit.* Had these data been

collected as the process was operating, we would have looked for a
source of trouble immediately after we plotted the average of
Sample 4. Because the data are historical, we are using the data
only to determine whether the process is in control.
The chart indicates the process was out of control when the data
were collected. Actions should be taken to identify the problems
that led to the low value in Sample 4 and the high value in Sample
18. After these problems have been identified and corrected, we
remove out-of-control samples from the data and recalculate × , R ,
and the control limits. We then redraw the charts and check to see
whether the process is in control now that the out-of-control data
points have been removed. We continue this process until all data
points are in control, and then use the resulting limits as trial limits
for future production.
Had the initial control chart shown that the process was in
control, the initial control limits would serve as trial control limits
for future production.

Continue to monitor and control the process

For day-to-day process monitoring, we will collect samples and
plot the results on a control chart using the trial center line and
control limits developed above. A data point beyond the limits and
unusual patterns will continue to be used as indicators that some
aspect of the process has changed and, therefore, as signals that we
must search for special causes of variation. If special causes are
found, corrective actions must be taken.
Due to the hectic pace of the manufacturing environment, it is
common for long periods to elapse between sample collection and

*In practice, the control chart for variation, the R chart, always
should be constructed and interpreted first to determine whether
the process is in or out of control. However, we have chosen to
sacrifice some technical accuracy in order to reduce the complexity
of discussion about multiple control charts. We chose to discuss the
× chart here rather than the R chart because × charts usually are
easier for newcomers to SPC to understand. An × chart monitors
the fluctuation in the process over time—a pretty straightforward
concept. An R chart, on the other hand, monitors the fluctuation in
process variability over time; in other words, it monitors the
variability of the variability.

15
STATISTICAL PROCESS CONTROL

plotting points on control charts. If an out-of-control situation is

indicated, it is likely that defective product has been produced in
the time between problem occurrence and problem detection; the
longer this period, the more defective product has been produced.
For control charts to be effective, data should be analyzed, plotted,
and interpreted and actions taken as soon as possible. Therefore,
control charts should be constructed and interpreted in real time by
the workers on the mill floor, not in the office by managers.
Finally, because control limits are a function of process varia-
tion, they should be evaluated periodically and revised when
control charts provide evidence that process variation has been
reduced.

Compare process performance to specifications:

Process capability analysis
Up to this point, we have concentrated on monitoring what the
process is doing but have shown little consideration for what we
want the process to do. In other words, we have paid little atten-
tion to how the process conforms to specifications.
The fact is, an in-control process can produce defective product
if the process is off-target or if the common-cause variation is too
high. As stated above, the first steps in an SPC program should be
to establish control. An out-of-control process is unstable, and
therefore estimates of process performance (centering and varia-
tion) are of little use. Only after establishing control can we exam-
ine the process’s ability to meet specifications.
Let’s begin with a graphical look at process variation.
Figure 7 shows normal curves for two process distributions.
LSL and USL are the lower specification limit and upper specifica-
tion limit, respectively. Both processes are centered on the target
dimension. The lower, wider distribution represents a process with
relatively high variation (high standard deviation); the taller,
narrower distribution represents a process with lower variation
(low standard deviation). The shaded area represents material
outside specifications.
The advantages of reducing variation are obvious. The process
with lower standard deviation produces far less defective material.
From a practical standpoint, however, we need to quantify the
relationship between the spread of the process and the spread of
the specifications. A process-capability analysis does this by using
capability indices.

16
 HOW AND WHY SPC WORKS

Defective Defective
product product

LSL µ USL
(Target)
Figure 7.—Effect of process standard deviation on percentage of product that
does not meet specifications.

The first capability index we will discuss is Cp which is

calculated as:
>

Cp = (USL – LSL) ÷ 6 σ
where σ is the process standard deviation and the “^” (pronounced
“hat”) symbol over it means “estimate.”
Recall that σ is the true standard deviation for a normally
distributed variable, and generally the best we can do is to estimate
>

it by sampling the process. The value 6 σ (plus and minus 3σ) is

the total width of process variation. Therefore, Cp is a ratio of the
specification width to total process width.
Continuing the example using the data in Table 1, let’s say our
specifications are 2.543 inches ± 0.003 inch. Therefore, our LSL is
2.540 and our USL is 2.546 inches. To estimate standard deviation
>

( σ ) we divide R (the average range) by d2, a table value found in

SPC textbooks. Appendix VI in Montgomery (1997) lists d2 as
2.326 for samples of size 5 (i.e., samples of five measurements).
>

Recall our estimate of R is 0.003; therefore, σ is 0.003 ÷ 2.326, or

0.001 inch. Cp is then:
Cp = (2.546 – 2.540) ÷ (6 x 0.001) = 1.0
What does this mean? In simple terms, Cp less than 1.0 is “bad”
and greater than 1.0 is “good.” That is because a Cp less than 1.0

17
STATISTICAL PROCESS CONTROL

indicates that process variation is higher than the specification

width, and therefore too much material is defective. A Cp greater
than 1.0 indicates that process variation is less than the specifi-
Cp doesn’t cation width, and therefore the process can meet the specifications
account for… while producing minimal defects. Because our Cp is 1.0, we know
that the process variation is equal to the specification width, and
process centering. therefore the process is (just barely) capable of meeting the
In theory, a process specifications.
could turn out Cp does not account for process centering relative to the target.
100% defective (Theoretically, a manufacturing process could be centered far away
product, yet if from the target specification and therefore producing 100% defec-
process variability tive product; yet if the process variability were low, Cp would
were low, Cp would indicate everything was okay.)
be okay. To account for process variation and for centering relative to
the target, we use another process capability index, Cpk. The
formula for Cpk is:

>
>
>

>
Cpl = ( µ – LSL) ÷ 3 σ , Cpu = (USL – µ ) ÷ 3 σ
Cpk = min{Cpl, Cpu}
where Cpl and Cpu are the lower and upper process capability

>
indices, respectively, relative to the process center; µ is our
estimate of process centering; and “min” indicates that Cpk is the
minimum (lesser) of Cpl and Cpu.
>

To calculate Cpk, we use σ as calculated above, and we use ×

>

as µ . (Recall that, previously, × was calculated as 2.542.) Cpk is

then:
Cpl = (2.542 – 2.540) ÷ (3 x 0.001) = 0.667,
Cpu = (2.546 – 2.542) ÷ (3 x 0.001) = 1.333
Cpk = min{0.667, 1.333} = 0.667
Cpk is interpreted much the same as Cp; that is, below 1.0 is
“bad” and above 1.0 is “good.” We get a bit more information with
Cpk, however. Because Cpl and Cpu are not equal, we know our
process is off-center. More specifically, because Cpl is lower than
Cpu, we know we are centered too close to the lower specification
limit. Therefore, excessive defects are produced below the lower
specification limit. Furthermore, using the normal distribution as
an approximation, we can calculate that about 2.3 percent (about
2.3 defects per 100) of the material will be defective (out of spec).
Calculations of this type will be discussed in a later publication.

18
 HOW AND WHY SPC WORKS

What if we adjusted the process to put it on target; that is, we

shifted the process center ( ×) from 2.542 to 2.543 inches? Cp does
not account for process centering, and so it would not be affected.
Cpk, however, would increase from 0.667 to 1.0, meaning that our
defect rate would decrease from about 2.3 percent to approxi-
mately 0.27 percent (about 2.7 defects per 1,000), a very signifi-
cant reduction.
In addition to adjusting the process to put it on target, what if we
could reduce process variation from 0.001 to 0.0008 inch? This
would result in Cp and Cpk of 1.25. The defect rate would drop to
0.009 percent (approximately 9 defects per 100,000). This is
obviously a very significant improvement in quality. After obtain-
ing cost estimates for scrap and rework, we can calculate the effect
that percentage reductions in defects will have on profit increases
and thus determine the benefits of quality improvement to the
bottom line.
The profit increases due to decreases in variation are substantial.
The question becomes, how do we reduce variation? The answer:
through continuous process improvement.

Continuous process improvement

As stated previously, real quality improvement requires a con-
tinual focus on reducing common-cause variation. Reducing
common-cause variation is possible only after the process has been
brought into control.
Process improvement (which leads to quality improvement)
requires a more systematic and structured approach than usually
required to remove special causes of variation. Companywide
quality improvement requires:
• Determining customer needs
• Setting quality goals
• Developing an improvement strategy
• Providing the necessary training and resources
• Establishing the organizational infrastructure (quality councils,
teams, etc.)
• Reviewing progress, and
• Revising the reward system
For these reasons, quality improvement requires the commit-
ment and involvement of upper management. Years of experience
have shown that “delegating quality” is ineffective. Juran (1989)

19
STATISTICAL PROCESS CONTROL

provides a thorough treatment of management’s role in quality

planning, control, and improvement.

Conclusions
The bottom line is: SPC helps manufacturers increase their
competitiveness and profitability. We have demonstrated SPC’s
foundation in mathematics and statistics to build your understand-
ing and confidence in SPC as a profit-making tool and to overcome
any preconceptions that SPC is yet another management fad.
We hope this publication has convinced you that implementing
SPC will make your company money, and we hope you will
commit to training your personnel to use SPC. To begin the train-
ing process, we refer you to Part 3 in this series, Starting an SPC
Program.

20
 HOW AND WHY SPC WORKS

For more information

Brown, G.M. and C. Horn. 1982. Quality control programs in
action: A southern pine mill. In: T.D. Brown, ed. Quality Control
in Lumber Manufacturing (San Francisco: Miller Freeman).
Brown, T.D. 1979. Determining lumber target sizes and monitoring
sawing accuracy. Forest Products Journal 29(4):48–54.
Brown, T.D. 1982. Determining rough green target size. In: T.D.
Brown, ed. Quality Control in Lumber Manufacturing (San
Francisco: Miller Freeman).
Brown, T.D. 1982. Setting up a size control program. In: T.D.
Brown, ed. Quality Control in Lumber Manufacturing (San
Francisco: Miller Freeman).
Brown, T.D. 1982. Evaluating size control data. In: T.D. Brown,
ed. Quality Control in Lumber Manufacturing (San Francisco:
Miller Freeman).
Brown, T.D. 1986. Lumber size control. Special Publication No.
14, Forest Research Laboratory, Oregon State University,
Corvallis, OR. 16 pp.
Cassens, D.L., J.L. Bankston, and J.S. Friday. 1994. Statistical
process control of hardwood lumber target sizes: Is it time?
Forest Products Journal 44(1):48–50.
Cook, D.F. 1992. Statistical process control for continuous forest
products manufacturing operations. Forest Products Journal
42(7/8):47–53.
Deming, W.E. 1982. Out of the Crisis (Cambridge, MA: Massa-
chusetts Institute of Technology Center for Advanced Engineer-
ing Study). 507 pp.
Deming, W.E. 1993. The New Economics: For Industry,
Government, Education (Cambridge, MA: Massachusetts
Institute of Technology Center for Advanced Engineering
Study). 240 pp.
Dramm, J.R. 1997. Statistical process control and other tools for
continuous improvement. In: Proceedings, Wood Technology
Clinic and Show (San Francisco: Miller Freeman).
Dramm, J.R. 1998. Results-driven approach to improving quality
and productivity. In: Proceedings, Wood Technology Clinic and
Show (San Francisco: Miller Freeman).

21
STATISTICAL PROCESS CONTROL

Duncan, A.J. 1986. Quality Control and Industrial Statistics, 5th ed.
(Homewood, IL: Irwin). 1,123 pp.
Eagan, F.M. 1982. Statistics-based size control in the mill. In: T.D.
Brown, ed. Quality Control in Lumber Manufacturing (San
Francisco: Miller Freeman).
Grant, E.L. and R.S. Leavenworth. 1988. Statistical Quality
Control, 6th ed. (New York: McGraw-Hill). 714 pp.
Juran, J.M. 1989. Juran on Leadership for Quality: An Executive
Handbook (New York: Macmillan, The Free Press). 376 pp.
Leicester, R.H. 1995. Statistical control for stress graded lumber.
In: Proceedings No. 7307, Statistical Process Control
Technologies: State of the Art for the Forest Products Industry
(Madison, WI: Forest Products Society).
Leonard, O.F. 1982. Quality control programs in action: The
medium-sized sawmill. In: T.D. Brown, ed. Quality Control in
Lumber Manufacturing (San Francisco: Miller Freeman).
Minneci, J. 1995. Control chart implementation in glued laminated
timber end joint production. In: Proceedings No. 7307,
Statistical Process Control Technologies: State of the Art for
the Forest Products Industry (Madison, WI: Forest Products
Society).
Moller, D. 1990. Statistical process control (SPC) for dry kiln
operations. In: Proceedings, Western Dry Kiln Association
Joint Meeting, Reno, NV, May 9–11.
Montgomery, D.C. 1997. Introduction to Statistical Quality
Control, 3rd ed. (New York: John Wiley & Sons). 677 pp.
Patterson, D.W. and R.B. Anderson. 1996. Use of statistical
process control in the furniture and cabinet industries. Forest
Products Journal 46(1):36–38.
Peck, J. and R. Myers. 1995. SPC in action: Reduction in variation
keys customer satisfaction. In: Proceedings No. 7307,
Statistical Process Control Technologies: State of the Art for
the Forest Products Industry (Madison, WI: Forest Products
Society).
Schaffer, R.H. and H.A. Thomson. 1992. Successful change
programs begin with results. Harvard Business Review. Jan.–
Feb., pp. 80–89.
Shewhart, W.A. 1931. Economic Control of Quality of Manu-
factured Product (Milwaukee, WI: Quality Press). 501 pp.

22
 HOW AND WHY SPC WORKS

Valg, L. 1965. Analysis of sawing accuracy by statistical quality

control. Research Note No. 51, Faculty of Forestry, University
of British Columbia, Vancouver. 4 pp.
Walton, M. 1986. The Deming Management Method (New York:
Putnam). 262 pp.
Warren, W.G. 1973. How to calculate target thickness for green
lumber. Information Report VP-X-112, Department of the
Environment, Canadian Forestry Service, Western Forest
Products Laboratory, Vancouver, BC. 11 pp.
Western Electric Co. 1956. Statistical Quality Control Handbook.
(Milwaukee, WI: Quality Press). 328 pp.
Whitehead, J.C. 1978. Procedures for developing a lumber size
control system. Information Report VP-X-184, Department of
the Environment, Canadian Forestry Service, Western Forest
Products Laboratory, Vancouver, BC. 15 pp.
Young, T.M. and P.M. Winistorfer. 1999. Statistical process control
and the forest products industry. Forest Products Journal
49(3):10–17.
Yuhas, J. 1982. Quality control programs in action: A large sawmill
complex. In: T.D. Brown, ed. Quality Control in Lumber
Manufacturing (San Francisco: Miller Freeman).

23
STATISTICAL PROCESS CONTROL

Glossary
Assignable causes of variation—See special causes of variation.
Attributes data—Qualitative data that can be counted for recording
and analysis. Results may be recorded as yes/no, go/no go, or
defective/not defective. Examples include percent defective in a
sample and number of blemishes on a surface.
Arithmetic mean—See average.
Average—A measure of location or central tendency which is the
sum of the observed values divided by the number of observa-
tions. Also called the arithmetic mean or, simply, the mean.
Bell curve—Common name for the normal distribution, a name
derived from the shape of the curve.
Cell—A grouping of values between specified upper and lower
boundaries used to create frequency distributions.
Center (centered, centering)—A numerical value that is “typical”
for a set of data. Values used include the average, the median,
and the mode.
Central tendency—See center.
Chance causes of variation—See common causes of variation.
Common causes of variation—Sources of variation that affect all
the individual values of the process output being studied. The
sources generally are numerous and individually of small impor-
tance but cannot be detected or identified. Also called chance,
random, and unknown causes of variation.
Control (statistical)—The condition that exists after a process in
which all special causes of variation have been eliminated and
only common causes remain.
Control chart—A graphic representation of a characteristic of a
process, showing plotted values of some statistic gathered from
the characteristic, a central line, and one or two control limits.
Used to determine whether a process is in statistical control and
to help maintain statistical control.
Control limits—On a control chart, the criteria for signaling the
need for action, or for judging whether a set of data does or does
not indicate a “state of statistical control.” Control limits are
calculated from process data and are not to be confused with
specification limits.

24
 HOW AND WHY SPC WORKS

Distribution—See frequency distribution.

Frequency distribution—A tally of the count, or frequency, of
occurrences of data in specific cells.
Histogram—A bar chart for displaying a frequency distribution.
In control—See control (statistical).
Mean—See average.
Median—The value at the midpoint in the ordered range of values:
half the values are greater than the median value, and half the
values are less than the median value.
Mode—The most frequently observed value.
Normal distribution—A continuous, symmetrical, bell-shaped
frequency distribution for variables data that underlies control
charts for variables.
Out of control—The absence of conditions described in control
(statistical).
Probability—A scientific discipline whose objective is to study
uncertainty. Probability is the likelihood (commonly called the
“odds”) that a specific event will occur.
Process limits—See control limits.
Random causes of variation—See common causes of variation.
Range—A measure of dispersion; the difference between the
largest observed value and the smallest observed value in a
given sample.
Sample—A group of items, observations, test results, or portions of
material taken randomly from a larger collection of items,
observations, test results or quantities of material, which provide
information that may be used as a basis for making a decision
about the larger collection. See also subgroup.
Special causes of variation—Sources of variation that are intermit-
tent, unpredictable, and unstable and that can be detected and
identified.
Specification limits—The engineering requirement for judging
acceptability of a particular characteristic. Specifications are not
to be confused with control limits.
Spread—General term describing the dispersion or variability in a
data set. Commonly measured with the range or standard
deviation.

25
STATISTICAL PROCESS CONTROL

Standard deviation (sample)—A measure of dispersion, calculated

as the square root of the sum of the squared deviations of obser-
vations from their average divided by one less than the number
of observations. The range often is used to estimate the standard
deviation.
Subgroup—In process control applications, generally synonymous
with sample.
Unknown causes of variation—See common causes of variation.
Variables data—Quantitative data, where measurements are used
for analysis. Examples include length, width, thickness, viscos-
ity, strength (e.g., pounds per square inch, or psi), and density.

26
 HOW AND WHY SPC WORKS

PERFORMANCE EXCELLENCE
IN THE WOOD PRODUCTS INDUSTRY

ABOUT THIS SERIES

This publication is part of a series, Performance
Excellence in the Wood Products Industry. The various
publications address topics under the headings of wood
technology, marketing and business management,
production management, quality and process control,
and operations research.
To view and download any of the other titles in the
series, visit the OSU Extension Web site at http://
eesc.oregonstate.edu/ then “Publications & Videos” then
“Forestry” then “Wood Processing” and “Business
Management”. Or, visit the OSU Wood Products Extension
Web site at http://wood.oregonstate.edu/

27
STATISTICAL PROCESS CONTROL

© 1999 Oregon State University

This publication was produced and distributed in furtherance of the Acts of Congress of May 8 and June 30, 1914. Extension
work is a cooperative program of Oregon State University, the U.S. Department of Agriculture, and Oregon counties. Oregon
State University Extension Service offers educational programs, activities, and materials—without regard to race, color, religion,
sex, sexual orientation, national origin, age, marital status, disability, and disabled veteran or Vietnam-era veteran status—as
required by Title VI of the Civil Rights Act of 1964, Title IX of the Education Amendments of 1972, and Section 504 of the
Rehabilitation Act of 1973. Oregon State University Extension Service is an Equal Opportunity Employer.
Published June 1999.

28