10-1 Quality Control

Run Tests
 Run test – a test for randomness
 Control charts test for points that are too extreme to be
considered random.
 However, even if all points are within the control
limits, the data may still not reflect a random process.
 Any sort of pattern in the data would suggest a non-
random process.
 The presence of patterns, such as trends, cycles, or bias
in the output indicates that assignable, or nonrandom,
cause of variation exist.
 Analyst often supplement control charts with a run
test, which is another kind of test for randomness.

10-2 Quality Control

Nonrandom Patterns in Control charts

Figure 10.11
 Trend: sustained upward or downward
movement.
 Cycles: a wave pattern

 Bias: too many observations on one side of

the center line
 Mean shift: A shift in the average

 Too much dispersion: the values are too

spread out
 10-3 Quality Control

Run Test
 A run is defined as a sequence of observations with a
certain characteristic, followed by one or more
observations with a different characteristic.
 The characteristic can be anything that is observable.

 For example, in a series AAAB, there are two runs; a

run of three A’s followed by a run of one B.
 The series AABBBA , indicates three runs; a run of
two A’s followed by a run of three B’s, followed by
a run of one A.

10-4 Quality Control

Run test
 There are two types of run test:
1. Runs up and down
2. Runs above and below the median
 In order to count these runs, the data are transformed
into a series of U’s and D’s (for up and down) and into
a series of A’s and B’s (for above and below the
median).
 There are three U/D and four A/B runs for the data:
25 29 42 40 35 38
- U U D D U
B B A A B A
Where the median is 36.5
10-5 Quality Control

Counting Runs

Figure 10.12 Counting Above/Below Median Runs (7 runs)

B A A B A B B B A A B

Figure 10.13 Counting Up/Down Runs (8 runs)

U U D U D U D U U D

10-6 Quality Control

Run test procedure

 To determine whether any patterns are present in
control charts, one must do the following:
1. Transform the data into both A’s and B’s and U’s
and D’s, and then count the number of runs in each
case.
2. Compare the number of runs with the expected
number of runs in a completely random series, which
is calculated as follows:
N
E ( r ) med  1
2
2N 1
E (r )u / d 
3
Where: N is the number of observations or data
points, and E(r) is the expected number of runs
10-7 Quality Control

Run test procedure (cont.)

3. Calculate the standard deviations of the runs as:
N  1
 med 
4
16 N  29
 u / d 
90

4. Calculate the test statistic (Ztest) as following:

observed number of runs – expected number of runs
Ztest standard deviation of number of runs
N If the Ztest is
r  (  1 )
Z test  2 For the median within ± 2 or ± 3;
N  1
4
then the process
2 N  1 is random;
r  ( )
Z test  3 Up and down otherwise, it is
16 N  29
not random
90

10-8 Quality Control

Run test
 Example sample mean sample Mean
Twenty sample means 1 10 11 10.7
have been taken from a
process. The means are 2 10.4 12 11.3
shown in the following 3 10.2 13 10.8
table. Use median and 4 11.5 14 11.8
up/down run test with
5 10.8 15 11.2
z = 2 to determine if
assignable causes of 6 11.6 16 11.6
variation are present. 7 11.1 17 11.2
Assume the median is 11. 8 11.2 18 10.6
9 10.6 19 10.7
10 10.9 20 11.9
10-9 Quality Control

Run test
 Solution
sample mean A/B U/D Sample Mean A/B U/D
1 10 B - 11 10.7 B D
2 10.4 B U 12 11.3 A U
3 10.2 B D 13 10.8 B D
4 11.5 A U 14 11.8 A U
5 10.8 B D 15 11.2 A D
6 11.6 A U 16 11.6 A U
7 11.1 A D 17 11.2 A D
8 11.2 A U 18 10.6 B D
9 10.6 B D 19 10.7 B U
10 10.9 B U 20 11.9 A U

10-10 Quality Control

Run test
Solution (cont.)
1. A/B: 10 runs and U/D: 17 runs
2. Expected number of runs for each test is:
N 20
E (r ) med  1   1  11
2 2 Although the median
2 N  1 2(20)  1
E (r ) u / d    13 test doesn’t reveal any
3 3
pattern, because its Ztest
3. The standard deviations are: value is within ±2, the
N 1 20  1 up/down test does; its
 med    2 . 18
4 4 value exceed +2.
16 N  29 16 ( 20 )  29 consequently,
 u/d    1 .8
90 90 nonrandom variations
4. The ztest values are: are probably present in
Z 
10  11
  0 . 46
the data and, hence, the
med
2 . 18
17  13
process is not in control
Z u/d   2 . 22
1 .8