23. - 25. 5.

2012, Brno, Czech Republic, EU

ANALYSIS OF NONRANDOM PATTERNS IN CONTROL CHART AND ITS SOFTWARE

SUPPORT
Darja NOSKIEVIOV
FMMI, VB-TU Ostrava, 17. listopadu 15, 708 33 Ostrava Poruba, Czech Republic,
darja.noskievicova@vsb.cz

Abstract
Identification of assignable causes of the process variability and restriction and elimination of their influence
are the main goals of statistical process control (SPC). Identification of these causes is associated with so
called tests for special causes or runs tests. From the time of formulation of the first set of such rules
(Western Electric rules) several different sets have been created (Nelson rules, AIAG rules, Boeing AQS
rules, Trietsch rules). This paper deals with the comparison analysis of these sets of rules and with the
analysis of the computer support of them. At the end of this paper some recommendations for the correct
use of these rules are formulated.
Keywords:, Common causes, special causes, runs tests, natural pattern, unnatural pattern
1.

INTRODUCTION

Statistical process control (SPC) is an approach to process control that has been widely used in any
industrial or non-industrial fields. It is primarily a tool for understanding process variation [1]. SPC is based
on so called Shewharts conception of the process variability. This conception distinguishes variability
caused by obviously effected common causes (process is considered to be statistically stable) from
variability caused by abnormal special (assignable) causes (process is considered not to be statistically
stable) using control charts. The main goals of SPC is an identification of abnormal variability caused by
special (assignable) causes with the aim to make the process stable, minimize the process variability,
improve the process performance. To meet these goals SPC must be built as the problem solving instrument
and the sequence of the subprocesses Out-of control signal revelation Root cause identification Action
acceptance Verification of action must be the axis of the SPC application. In this paper the stress is put on
the first subprocess applying to so called Shewhart control charts.
Control chart is the main SPC instrument for the analysis of the process variation over time. It is a graphical
depiction of the process variation and its natural and unnatural patterns. Control chart displays a value of the
quality characteristic that has been measured or some sample statistics that has been computed from
measured values in sample versus the sample number or time. Central line in the control chart represents
the average value of the quality characteristic corresponding to the statistically stable process (only natural
pattern is presented and only common causes influence the process). Control limits are set so that when the
process is statistically stable, nearly all the points in control chart fall between them (for mostly applied 3
control limits and assumed normal distribution it is 99,73% points). As long as the points are within the
control limits and they show natural random pattern, the process is considered to be statistically stable. But if
some points are out of the control limits or the points inside the limits show nonrandom (unnatural) pattern it
is assumed that assignable cause of the abnormal process variation is present and must be removed from
the process via searching for it and corrective action or some improvement must be realized.
In general the points plotted on a control chart form an irregular up-and-down pattern natural or unnatural.
An advantage of using control charts is that it enables observation of the process behaviour over time to
search for occurring of any unnatural pattern [2] as a symptom of the special (assignable) cause and as a
signal for the reduction of the process variation and for improvement.

23. - 25. 5. 2012, Brno, Czech Republic, EU

Natural pattern is such pattern where the points fluctuate randomly. It can be characterized as follows [3]:
most points lie near the central line; only few of the points spread out and reach the control limits; very rarely
some point goes out of the control limit. When some or all these attributes are missing the pattern will be
classified as unnatural. Many types of unnaturalness can be recognized in control charts. But several generic
patterns that can be detected for various processes were defined. In table 1 we can find mostly
recommended (for instance in [3], [4], [5], [6], [7]) and applied nonrandom patterns for Shewhart control
charts with their description and their typical symptom in control chart. To standardize and simplify the
process of the typical unnatural patterns recognition statistical tests (rules) that quantify the length of the
unnatural patterns were formulated.
Table 1 Unnatural pattern description
No.

Unnatural (nonrandom) pattern

Large shift
(strays, freaks)
Smaller sustained
shift
Trends

Stratification

Mixture

Systematic
Variation
Cycle

1
2

2.

Pattern Description

Symptom in control chart

Sudden and high change

Points near and or beyond control limits

Sustained smaller change

Series of points on the same side of the central

line
Steadily increasing or decreasing run of points

A continuous changes in one

direction
Small differences between values in
a long run, absence of points near
the control limits
Saw-tooth effect, absence of points
near the central line
Regular alternation of high and low
values
Recurring periodic
movement

A long run of points near the central line on the

both sides
A run of consecutive points on both sides of
central line, all far from the central line
A long run of consecutive points alternating up
and down
Cyclic recurring patterns of points

TESTS FOR UNNATURAL PATTERNS RECOGNITION

Tests for unnatural patterns supplement the evidence of the unnatural pattern given by the position of the
points relative to the control limits or central line with evidence based on the statistical theory of runs [8]
except the Shewhart rule (point out of the control limit). They are based on the probability calculations that
tell us the portion of points lying near the central line, near the control limits, etc. We will discuss tests that
cover unnatural patterns no. 1 - 6 (see the former table), i.e. shifts, trends, mixtures, systematic variation and
stratification. Table 2 contains sets of the rules that are the most well known and applied (Shewhart rule
[9], Western Electric rules [3], Nelson tests [10], ISO 2859 tests [11], special sets formulated by large
companies (Boeing [12], AIAG [13]) and the most new set by Trietsch [2].
Rules 1 - 4 were defined to quickly recognize patterns linked to the shifts in the process. Rule 5 is connected
with the trends, rule 6 and 8 with patterns caused by incorrect sampling and rule 7 with abnormal oscillation.
Historically first rule for the detection of the assignable cause developed W. S. Shewhart in table 2 it is
Rule 1. Set of Western Electric Rules added to the Shewhart criterion run tests covering such unnatural
patterns as smaller or sustained shifts (called zone tests), mixtures, stratification and systematic variation. To
be able to apply these tests the region between control limits in control chart must be divided into 6 zones
each of 1 sigma width and the location of the predefined sequences of the points in relation to these zones
must be evaluated. As normal distribution of the applied sample statistics is supposed, in zones C 68,27%
values, in zones B 27,18% values and in zones A 4,28% values are expected when process is statistically

23. - 25. 5. 2012, Brno, Czech Republic, EU

Table 2 Tests for unnatural patterns

Tests for unnatural
patterns
Rule 1

Shewhart

Western
Electric
Rules

Rule

Nelson
Tests

ISO
8258

AIAG
Rules
1

Boeing
ASQ
Rules

Trietsch

9 points

9 points

7 points

One or more points more

than 3 sigma from the
mean (beyond zone A)
Rule 2
2 of 3 consecutive points
between 2 and 3 sigma
from the mean (in zone A
or beyond)
Rule 3
4 of 5 consecutive points
between 1 and 3 sigma
from the mean (in zone B
or beyond)
Rule 4
8 consecutive points on
one side of the mean (in
zone C or beyond)
Rule 5

9 points
3

Six consecutive points

steadily increasing or
decreasing
Rule 6

15 consecutive points both

above and below central
line ( in zones C)
Rule 7

13 points
4

14 consecutive points
alternating up and down
Rule 8
Eight points in a row on
both sides of the central
line with none in zones C

13 points
8

5 points

stable.
Rules 2, 3 and 4 were designed for quicker identification of the shifts from the mean as compared to the test
1 (early warning indicators).
As it can be seen in table 2 the most complex as to the rate of covering all eight defined rules are Nelson
rules [10]. In addition to the Westerns Electric set Nelson defined run test for the oscillation and for the
trends. In rule 4 he changed the length of the run from 8 to 9 points. As compared to the Western Electric
rules the sequence of the rules had been changed by Nelson, too. Standard for Shewhart control charts ISO
8250 has wholly copied the Nelson set of rules. Also Trietsch in his publication works with the Nelson rules
but in rules 6, 7, 8 he has changed the length of the runs to reach better statistical properties of these tests.

23. - 25. 5. 2012, Brno, Czech Republic, EU

Rules of the companies Boeing and AIAG have less complex sets of rules based on the first four Western
Electric zone rules.
3.

METHODOLOGY
RECOGNITION

FOR

APPLICATION

OF

THE

RULES

FOR

UNNATURAL

PATTERN

In literature (see for instance [9], [3], [14], [15], [10], [16], [17], [4]) several strategies for the application of the
rules for unnatural pattern recognition from different authors can be found. They have many differences but
the main idea is common: runs test must not be applied routinely; it is really impractical and simultaneous
application of all known rules brings a danger of unacceptable rise of the false alarm.
As it can be found in the mentioned literature some authors differ in the idea what rules to apply routinely.
But predominantly they recommend rule 1 in conjunction with rule 4. For additional sensitizing of the control
charts some of the authors recommend rules 2 and 3. Wheeler recommends using of these rules only when
increasing sensitivity of a control chart is absolutely necessary. From some strategies it can be seen an effort
to apply separately rules for the identification of patterns for shifts and patterns for subgrouping problems
(for instance Wheeler intents [16] that the main use of Rule 6 is at start-up of SPC than in an on-going
control). Some authors put the stress on the phases of the SPC implementation when selecting suitable
rules. Montgomery means that Sensitizing rules can be helpful when the control chart is first applied and the
focus is on stabilizing an out-of-control process. However, once the process is reasonably stable, the
routine use of these sensitizing rules to detect small shifts or to try to react more quickly to assignable
causes, should be discouraged ([4], p. 177).
4.

TESTS AND TYPES OF CONTROL CHARTS

When selecting suitable rules for the selected control chart it must be considered that ([10], [3], [2], [16.]):

All rules can be applied to x-bar chart and individuals chart supposing normal distribution of the variable;

Rules 1- 4 can be applied to the dispersion charts (R, s) without any modification when sample size is 5
or more (it should lead to the symmetric control limits);

Nonparametric rules (Rule 5 and 7) work sufficiently well for any continuous distribution.

The rules can be applied without any problems to attribute control charts np and c supposing that
normal approximation is valid and control limits are reasonably symmetrical.

The same pays for the control charts p and u with constant control limits.

5.

ANALYSIS OF SELECTED SW PRODUCTS SUPPORTING UNNATURAL PATTERNS

RECOGNITION

As it was mentioned in the previous chapter routine application of runs rules can lead to reducing the
effectiveness of SPC implementation. Now days there are many SW products that support this part of SPC.
They enable to increase effectiveness of the problem solving in SPC but at the same time they bring danger
of routine application of the runs rules with its negative effects.
In this charter the analysis of SW (Statgraphics [18], [19], Minitab [20]) used for the education at the
Department of quality management at the University of Ostrava, Czech Republic and really excellent
statistical SW used by many well known Czech universities (Statistica [21]) will be done in connection with
their support to the control chart unnatural pattern recognition and with respect to the problems discussed in
the previous chapter.
In general all analysed SW products cover the set of Nelson rules (the same used in ISO 8258) with some
little departures. All Nelson rules (all tests for assignable causes) including Shewhart criterion 1 (Rule 1)
contains only SW Minitab. The rest of the analysed SW have rule 1 incorporated into the basic control chart

23. - 25. 5. 2012, Brno, Czech Republic, EU

analysis outside the set of runs tests. Statgraphics products have slightly different definition of run in rule 5,
rule 8 (Statgraphics Plus) and Rule 8 (Statgraphics Centurion). Detailed characterization of every analyzed
SW can be found in table 3.
As to the complexity in general it can be stated that all analysed SW cover the set of Nelson rules (the same
used in ISO 8258) with some little departures. Only SW Minitab contains all Nelson rules (tests for
assignable causes) including Shewhart criterion 1 (rule 1). The rest of the analysed SW have rule 1
incorporated into the basic control chart analysis outside the set of runs tests ( for that reason in these SW
the term runs tests is used). From the point of view of the common types of the unnatural patterns defined
in this paper (large shift, smaller sustained shift, trends, stratification, mixture, systematic variation) all
analysed SW cover all these types except Statgraphics Plus where systematic variation is not considered.
All analyzed SW cover all standard Shewhart control charts for variables and for attributes. Statgraphics
products cover in addition ARIMA residuals chart and CusCore chart.
To use the tests correctly the user needs to have at least basic knowledge of the statistical base of the tests.
This request is to a certain extent solved only in SW Statistica.
Different default runs length definition as compared to the Nelson rules can be found in the Statgraphics SW
products in rule 5, rule 8 (Statgraphics Plus) and rule 7 (Statgraphics Centurion).
The analyzed SW also differ in the default rules. SW Statgraphics Plus and Statistica have preset all defined
rules, covering unnatural patterns as all types shifts in the mean, systematic variation, stratification and
mixture; Statgraphics Centurion only first four (according to the numbering in table 2 - rules 4, 2, 3, 5, i.e.
rules covering different types of shifts).WS Minitab has only first four rules ((according to the numbering in
table 2 - rules 1, 4, 5, 7 covering basic types of the mean shift and systematic variation). For less
experienced user setting all the tests as a default can be confusing. Better access is setting the limited
number of rules. It is not good strategy to apply all tests simultaneously it leads to the higher false alarm
and to more complicated analysis of the process. The limited number of rules as a default is a smaller evil.
In addition every rule is not suitable for each type of the chart. In SW Statgraphics Centurion and Minitab the
user is to some extend protected from incorrect application of runs rules as to the type of control chart by the
limited default set of runs that can be applied to various control charts. But most precisely it is solved in
Minitab where the system itself enables to use only suitable runs tests when it is needed for R, S, moving
Range, np, p, u, c charts only tests 1, 4, 5 and 7 (numbering according table 2) are allowed.
All the tested SW are flexible as to the choice of the run test and the run length. We can select only test or
tests we need. It offers the possibility to set the unique strategy for the analysis of the process instability
related to the process behaviour. The possibility to change the length of the runs can lead to the selection of
the optimal length connected with the acceptable value of false signal. Except SW Minitab all tested SW
enable to change definition of zones. It offers possibility to adopt the rules to control chart that is not based
on the normal distribution. It is sufficiently flexible for the user but the information how to decide about
suitable tests or the length of runs is not appropriately detailed.
Indication of patterns in control chart is solved in all analyzed SW. SW Statistica offers in addition the
following function: when the out-of-control conditions are investigated and an explanation for them is found,
investigator can assign descriptive labels to those out-of-control samples and explain the causes and actions
that have been taken. Having causes and actions displayed in the chart will document that the centre line
and the control limits of the chart are affected by special cause variation in the process.
As it was mentioned above simultaneous application of more rules leads to rise of false alarm and it can lead
to worse and longer identification of really affected uncommon causes. Warning about simultaneous using
more rules is included only in the adviser of SW Minitab and Statistica. The same conclusion can be done as
to the warning about the necessity to have a deep knowledge about the analyzed process. But without
appropriate knowledge about the process the unnatural patterns analysis cant offer acceptable results.

23. - 25. 5. 2012, Brno, Czech Republic, EU

Table 3 Multicriteria analysis of the selected SW

Properties/ SW

Statgraphics
Plus V.15

Statgraphics
Centurion

Minitab

Statistica

v. 15

v. 16

v. XV
Term used for unusual pattern
Term used for tests
Complexity
Statistical base of tests

Unusual pattern

Unusual
sequence

Nonrandom
pattern

Systematic
pattern

Runs chart

Runs tests

Tests for
special causes

Runs tests

6 from 8

7 form 8

8 from 8

7 form 8

no

no

no

yes

Coverage of unnatural pattern

Large shift

yes

yes

yes

yes

Smaller sustained shift

yes

yes

yes

yes

Trends

yes

yes

yes

yes

Stratification

yes

yes

yes

yes

Mixture

yes

yes

yes

yes

Systematic Variation

no

yes

yes

yes

Default number of rules

6 all

First 4

Possibility to choice the rules

yes

yes

yes

yes

Possibility to change the length of runs

yes

yes

yes

yes

Possibility to change definition of zones

yes

yes

no

yes

Indication of patterns in control chart

yes

yes

yes

yes

Record of assignable causes into

control chart

no

no

no

yes

Record of actions into control chart

no

no

no

yes

Warning about simultaneous using more

rules

no

no

yes

yes

Warning about the necessity to have a

deep knowledge about process

no

no

yes

yes

S+0

S +0

yes

yes

yes

yes

Information to that control charts are

rules applicable

Yes (no precise)

Yes (no precise)

partial

partial

Defined potential general assignable

causes

no

no

no

yes

References

yes

yes

yes

yes

Description of rules

yes

yes

yes

yes

Interpretation of rules

no

no

no

yes

Different default runs definition as

compared to Nelson rules

Types of control chart covered (SShewhart, O- other)

The same default set of rules for
different control charts

First 4 (WE)

7 all

23. - 25. 5. 2012, Brno, Czech Republic, EU

Description of the set of the rules can be found in each analyzed SW. But the most detailed description
including possible interpretation of the rules and definition of possible assignable causes offers only SW
Statistica.
The former analysis of the properties of the analyzed SW showed that the best solution in the field of runs
tests application offer SW Minitab and Statistica. But they also suffer from some disadvantages that can lead
little experienced user to incorrect results and to complicate procedure of the process control. It can
unfortunately result in ignoring these runs tests in the analysis of the process stability or failing of the whole
SPC application. For that reason in the next charter simple methodology for the correct application of tests
for special casus is formulated.
6.

RECOMMENDATIONS FOR THE APPLICATION OF RULES FOR UNNATURAL PATTERNS

RECOGNITION

7.

Before applying runs tests or tests for special causes check the default set of tests for selected
control charts.
Never routinely apply all runs tests or tests for assignable causes even if they are preset in your SW.
Check if chosen control charts have reasonably symmetrical control limits because in such situation
next conclusions are valid:
all tests (Rule 1 - Rule 8) can be applied to x-bar chart and individuals chart supposing normal
distribution of the variable;
rules 1- 4 can be applied to the dispersion charts (R, s) without any modification when sample
size is 5 or more;
the rules can be applied without any problems to control charts np and c supposing that normal
approximation is valid and control limits are reasonably symmetrical.
The same pays for control chart p and u with constant control.
Tests 5 and 7 work sufficiently well for any continuous distribution because of their
nonparametric nature.
In the situation when control charts have reasonable symmetrical control limits, select suitable test
according to the following schema:
- Rules 6 and 8 apply at the beginning of the SPC implementation to verify rational subgrouping.
- In the phase of the statistical stability ensuring start with Rule 1 and 4.
- If additional sensitizing of control chart is necessary add Rules 2 and 3 or only one of them.
- When your knowledge of the process has risen during the previous SPC implementation adds
some of the other standard tests or your own fitted test.
- Test 5 apply singularly (its marginal increase in sensitivity to real signals is more than offset by
the greater increase in false alarm ([16], p. 137).
Rules 1 4 apply to both halves of the control chart but separately.
When using a couple of control charts (for instance x-bar and R) start with an analysis of the
dispersion control chart.
Keep in mind that where the run test is violated does not always indicate where the process change
had occurred.
CONCLUSIONS

Based on the theoretical background made in the introductory chapter this paper dealt with the analysis and
comparison of the several known sets of the rules for the special causes identification. Shewhart rule,
Western Electric Rules, Nelson tests, ISO 2859 tests, special sets formulated by large companies (Boeing,
GE) and the most new set by Trietsch were compared. In the light of the conclusions of these analyses the
next multi-criteria analysis of the selected SW product with the focus on the tests for the special causes

23. - 25. 5. 2012, Brno, Czech Republic, EU

identification were done with the aim to reveal the advantages and disadvantages of every discussed SW
from the point of view of the support for meeting the main goal of SPC. Based on this analysis several
recommendations for the correct and effective application of these tests were then formulated.
ACKNOWLEDGEMENT
This paper was elaborated in the frame of the specific research project No.
SP2012/42, which has been solved at the Faculty of Metallurgy and Materials
Engineering, VB-TU Ostrava with the support of Ministry of Education, Youth
and Sports, Czech Republic.
REFERENCES
[1]

STAPENHURST, T. Mastering Statistical Process Control. Oxford: Elsevier, 2005, 460 p. ISBN 0750665297.

[2]

TRIETSCH, D. (1999). Statistical quality control. A loss minimization approach. London: World Scientific. 387 p.
ISBN 9810230311.

[3]

WESTERN ELECTRIC COMPANY, INC. Statistical Quality Control Handbook, 2 ed. Easton: Mack Printing Co., 1958. 328 p.

[4]

MONTGOMERY, D.C. Introduction to Statistical Quality Control. New York: J. Wiley Sons, 2001. 796 s.
ISBN 0-471-31648-2.

[5]

GRIFFITH, G.K. Statistical Process Control Methods for Long and Short Runs. Milwaukee, Wisconsin: ASQC Quality Press,
1996. 250 p. ISBN 0-87389-345-X.

[6]

EVANS, J. R., LINDSAY, W.M. Managing for Quality and Performance Excellence. 7th ed. Mason, OH: Thomson, 2008. 783 p.
ISBN 10:0-324-64686-0.

[7]

BESTERFIELD, D. H. Quality Control. 8 ed. New Jersey: Prentice Hall, 2009, 552 s. ISBN 978 0135 000 953.

[8]

GRANT, E. L., LEAWENHWORTH, R.S. Statistical Quality Control. New York: McGraw Hill, 1996. 764 p. ISBN 0-07-024162-7.

[9]

SHEWHART, W. A. 1931 Economic Control of Quality of Manufactured Product. New York: Van Nostrand, 1931. 501 p. ISBN
0-873890-76-0. (republished by Milwaukee: ASQ, 1980).

[10]

NELSON, L. S. 1984, The Shewhart Control Chart Tests For Special Causes. Journal Of Quality Technology,1984, V.16, No.4,
pp. 337-339.

[11]

SN ISO 8258 Shewhartovy regulan diagramy. Praha: SNI, 1994.

[12]

D1-9000 ASQ Boeing Company, 2000.

[13]

AIAG SPC-3, 2005.

[14]

BURR, I., W.: Statistical Quality Control Methods. New York: Marcel Dekker Inc., 1976. 532 p. ISBN 9780824763442.

[15]

OTT, E. R. Analysis of Means A Graphical Procedure. Journal of Quality Technology. V.15, pp. 10-18. 1983.

[16]

WHEELER, D. Advanced Topics in Statistical Process Control. Knoxville: SPC Press, Inc., 2004. ISBN 0-945320-63-9.

[17]

MITRA, A. Fundamentals of Quality Control and Improvement. New York: MacMillan Publishing Company, 1993, 651 s. ISBN
0-02-381791-7.

[18]

STATGRAPHICS PLUS v. 5.1. Manugistics Inc., 2000.

[19]

STATGRAPHICS CENTURION,v XV. StatPoint Technologies, 2006.

[20]

MINITAB v. 16. Minitab Inc., 2010.

[21]

STATISTICA v. 10, StatSoft, Inc., 2010.

nd

th