Professional Documents
Culture Documents
2004 - Wu - Using Target Costing Concept in Loss Function and Process Capability Indices To Set Up Goal Control Limits
2004 - Wu - Using Target Costing Concept in Loss Function and Process Capability Indices To Set Up Goal Control Limits
DOI 10.1007/s00170-003-1547-8
O R I GI N A L A R T IC L E
Hsin-Hung Wu
Received: 30 September 2002 / Accepted: 30 September 2002 / Published online: 30 June 2004
Springer-Verlag London Limited 2004
Abstract This paper depicts the relationship among the rT: The standard deviation from the target va-
loss function, process capability indices and control lue of a population
charts to establish goal control limits by extending the r^: The estimated standard deviation ofr
target costing concept. The specification limits derived ^0 :
r The new process standard deviation when
from the reflected normal loss function is linked through 2D’ are applied
the Cpm value, computed either directly from the raw n: The sample size of the subgroup
data or given by management or engineers, to conven- d2: The parameter used to estimate r 2,
^ ¼ R=d
tional control charts to obtain goal control limits. The determined by n
target value can be taken into consideration directly. D 4, D 3,
The parameters in Xand Rcontrol charts,
The advantages of applying the target costing philoso- and A2: determined by n
phy are also discussed. This paper explains, from a c4: the parameter used to estimate r ^ ¼ S c4 ,
quantitative approach, that reducing process variation is determined by n
not enough to solve quality problems. In fact, reducing B 4 , B 3, The parameters in X and Scontrol charts,
process variation should be used along with bringing the and A3: determined by n
process mean to the target value. L(y): The general loss function
L1(y): The general loss function when the quality
Keywords Reflected normal loss function Æ Target improvement is implemented
costing concept Æ Process capability index Æ Goal control h: The parameter used to determine L1(y),
limits where L1(y)=hK
f(y): The probability density function
A list of symbols
K: The maximum-loss parameter in the re-
flected normal loss function
c: The shape parameter in the reflected normal 1 Introduction
loss function, D/4
T: The target value Loss functions are often used to measure the loss of a
D: The distance from the target value to the product characteristic deviated from a desired target
point where K first occurs (tolerance or value, largely due to the Taguchi philosophy. Taguchi [1]
specification limit) has used a modified quadratic loss function to assess the
E(L(y)): The expected loss associated with the re- loss associated with deviations of a product character-
flected normal loss function istic from the target value. As the process departs from
l: The average value (mean) of a population the target value, a loss is incurred. Moreover, the loss
r: The standard deviation of a population increases as the process is further away from the target
value. Therefore, the purpose of loss functions is to re-
H.-H. Wu flect the economic loss associated with variation in, and
Department of Business Administration, deviations from, the process target or the target value of
National Changhua University of Education, a product characteristic [2].
No. 2 Shda Road, Changhua City, The loss function has been widely used in many areas.
Changhua, Taiwan 500
E-mail: hhwu@thmu.edu.tw
For instance, Chen [3] has applied the Taguchi loss
Tel.: +886-4-7232105 ext. 7405 function towards setting up specification limit(s) when
Fax: +886-4-7211162 the product characteristic follows an exponential distri-
207
n o
bution. In addition, Chen [4] developed a specification Þ2
Lð y Þ ¼ K 1 exp ðyT
2c 2
limit under the membership function by implementing n o ð1Þ
the Taguchi loss function. Sauers [5] also applied the 8ðyT Þ2
¼ K 1 exp D2 ;
Taguchi loss function along with process capability
indices and control charts to set up goal control limits by where y represents the quality measurement, K is the
considering the target costing philosophy. Spiring and maximum-loss parameter, T is the target value and c is a
Yeung [2], on the other hand, have developed a general shape parameter. The target value, shape and maximum-
class of loss functions with industrial examples to dem- loss parameters allow a customisation of the loss func-
onstrate the advantage of using loss functions. tion to meet practitioners’ requirements. In addition, c is
The traditional loss function has been criticised by defined as D/4, where D is the distance from the target
practitioners and researchers because it fails to provide a value to the point where K first occurs.
quantifiable maximum loss and magnitude of losses The reflected normal loss function is asymptotic to
associated with extreme deviations from the target value the maximum loss and will attain this value only at ±¥.
[6, 7]. On the other hand, Spiring [8] has pointed out that The equation of c=D/4 ensures that the loss function at
things such as production resources, cost of identifica- the points T±D will be 0.9997K, which can be consid-
tion, scrap or rework and liability generally have a ered to be K. The expected loss associated with the re-
maximum loss. As a result, the traditional loss function flected normal loss function defined by Spiring [8] is:
with an infinite maximum loss is inadequate to describe Z !
the loss associated with a product characteristic. ðy T Þ2
EðLð y ÞÞ ¼ K K exp f ð y Þdy ; ð2Þ
For an organization’s viewpoints, it is always eco- 2c2
nomical to reduce unit-to-unit performance variation
around the target value even if the products are within where f(y) is the associated probability density function.
specification limits [9]. Continuous quality improvement If the quality characteristic follows a normal distribution
and cost reduction are necessary for an organisation to with a mean of l and a standard deviation of r, the
stay in a competitive economy. Besides, quality impro- expected loss becomes:
vement requires the continuous reduction of variation in R 1 n 2
o
1 ðyT Þ ðylÞ2
product and/or process performance around the target ELð y Þ ¼ K K pffiffiffiffi exp 2 c2
þ 2 dy
2pr
r ð3Þ
value. Therefore, the purpose of this paper is to apply c ðlT Þ2
target costing philosophy discussed by Sauers [5] along ¼ K 1 pffiffiffiffiffiffiffiffiffi
2 2
exp 2
2ðr þc Þ2 ;
r þc
with Spiring’s loss function with finite maximum loss in
goal control limits to reduce common causes of variation. where the minimum happens at l=T.
The paper is organised as follows: first, the loss Spiring and Yeung [2] stated that using this type of
function developed by Spiring [8] and Spiring and Yeung loss function provides the optimal setting for the process
[2] is reviewed. Then, control charts and process capa- from an economic consideration. When the process
bility indices and the target costing philosophy are dis- deviates from the target value, the loss increases. In
cussed. The framework based upon the target costing addition, the loss function can be used as a control chart
philosophy is then developed. Finally, an example and to monitor the loss from a process. Though the control
conclusions are provided. charts and charts developed by the loss function might
be similar, the standard deviation of the economic loss is
usually larger than those of control charts. Moreover,
2 Spiring’s loss function the subgroup averages for the economic loss and mea-
surements are different, which provide practitioners with
Spiring [8] has developed the reflected normal loss alternatives for assessing process performance.
function based upon the Taguchi philosophy. The gen-
eral equation and figure of this reflected normal loss
function are shown in Eq. 1 and Fig. 1: 3 Control charts and process capability indices
;
CLð RÞ ¼ R ð5Þ USL LSL D
Cp ¼ ¼ ; ð16Þ
6r 3r
;
LCLð RÞ ¼ D3 R ð6Þ
USL l l LSL
Cpk ¼ min ; ; ð17Þ
r^ R 3r 3r
UCLðX Þ ¼ X þ 3 pffiffiffi ¼ X þ 3 pffiffiffi ¼ X þ A2 R
; ð7Þ
n d2 n
and
CLðX Þ ¼ X ; ð8Þ USL LSL USL LSL D
Cpm ¼ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ;
6rT
and 6 r 2 þ ðl T Þ2 3 r 2 þ ðl T Þ2
r^
R ð18Þ
LCLðX Þ ¼ X 3 pffiffiffi ¼ X 3 pffiffiffi ¼ X A2 R
; ð9Þ
n d2 n
where USL and LSL are the upper specification limit
where UCL, CL and LCL stand for upper control limit, and lower specification limit, respectively, and 2D is the
centre line and lower control limit, respectively, distance between USL and LSL. In Eqs. 16, 17, and 18,
2 , n is the sample size of the subgroup, and the
^ ¼ R=d
r l is the mean which is the sum of the numerical values of
parameters of D4, D3 and A2 can be found in the work of the measurement divided by the number of items
Gitlow, Oppenheim and Oppenheim [9]. examined, r is the standard deviation which is the square
For X and S control charts, the formulas are as follows: root of the average squared deviates from the mean and
rT is the standard deviation from the target value which
UCLðS Þ ¼ B4 S ; ð10Þ is the square root of the average squared deviation from
the target value.
CLðS Þ ¼ S ; ð11Þ The values of l, r and rT are usually unknown, and
estimations from the sample data are required. If control
LCLðS Þ ¼ B3 S ; ð12Þ
charts, such as X -R and X -S control charts, are imple-
r^ S mented prior to the process capability analysis, Eqs. 16,
UCLðX Þ ¼ X þ 3 pffiffiffi ¼ X þ 3 pffiffiffi ¼ X þ A3 S ; ð13Þ 17 and 18 can be revised as follows:
n c4 n
^ p ¼ USL LSL ¼ D ;
C ð19Þ
CLðX Þ ¼ X ; ð14Þ 6^
r 3^
r
!
and USL X X LSL
^
Cpk ¼ min ; ; ð20Þ
r
^ S 3^
r 3^
r
LCLðX Þ ¼ X 3 pffiffiffi ¼ X 3 pffiffiffi ¼ X A3 S ; ð15Þ
n c4 n
and
where r^ ¼ S c4 , n is the sample size of the subgroup,
and the parameters of B4, B3 and A3 can be found in the ^ pm ¼ USLLSL ¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
USLLSL D
C ffi ¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2ffi ;
work of Gitlow, Oppenheim and Oppenheim [9]. 6^
rT 2
6 r 2
^ þ X T 2
^ þ X T
3 r
After the process is in statistical control, a process
capability analysis can be conducted to further examine if ð21Þ
the process is capable of producing high quality prod-
ucts. Typically, Cp, Cpk and Cpm are the well-known where X is available directly from the X control chart,
process capability indices (PCIs) for process capability ^ depends on the X -R and X -S control charts. For
while r
analysis. These PCIs are widely used throughout industry
the X -R charts, R/d 2 is used to replace r 4 is
^, while S/c
to quantify the ability of a process within specification used to substitute r ^ for the X -S charts [14]. The third
limits when the populations are normally distributed. approach is to use a sample standard deviation,
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
The Cp index, classified as the first generation, only Pn .
ðXi X Þ n1, and X from the overall samples
2
considers the relationship between the spread and the S¼
i¼1
specification limits. On the other hand, the Cpk index, ^ and X from the overall samples to replace r
to replace r ^
classified as the second generation, considers the loca-
and X in Eqs. 19, 20 and 21.
tion of a process with the specification limits, while the
Cpm index, also viewed as the second generation, further
considers the departures of a process mean from the
target value. In fact, the Cpm index is more sensitive to 4 The target costing philosophy
determine if the process is centred at the target value.
This paper only focuses on normal process data. For The ‘target costing’ philosophy has been applied to set the
non-normality-based PCIs, please refer to [10, 11, 12, price of products by many companies [15, 16]. Sauers [5],
13]. The formulas of these normality-based Cp, Cpk, and on the other hand, has discussed how the target costing
Cpm indices are described as follows: philosophy should be applied to combine the philosophy
209
of the Taguchi loss function, the process capability index to force manufacturers to reduce the total variation
and the X -R charts and developed the procedure for the relentlessly, even when SPC charts illustrate no special
implementation in practice. The philosophy is as follows: causes of variation in the process.
By starting with the anticipated acceptable market price,
the companies subtract the desired profit margin to obtain
a target manufacturing cost. Design and manufacturing 5 The implementation of target costing concept
engineers are responsible to bring the product into being
at this cost. By applying this technique, price concerns can Suppose the target costing concept is applied and a
be driven down to the process level. Continuous quality improvement program is implemented; the gen-
improvement can be carried out by listening to the price eral loss is expected to be reduced to L1(y)=hK, where
concerns of the marketplace [5]. 0<h<1. If the loss function L1(y) in Eq. 1 is defined in
Roth [17] also described a similar concept to estimate the interval [TD’, T+D’], at the point yat which
scrap and rework costs from the process analysis. Dif- 8(yT)2=D’ we have:
ferent settings in process means would result in different
8D0
rework costs and scrap costs, and the total cost is the L1 ð y Þ ¼ hK ¼ K 1 exp 2 ð22Þ
sum of the rework and scrap costs. Typically, the rela- D
tionship between the total cost and the process settings and
can be represented as an increasing function ƒ as a sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
quadratic function with a minimum value at some cer-
D2
tain point during process settings. If the setting is far 8D0 ¼ lnð1 hÞ : ð23Þ
away from the best setting point with the minimum 8
value, the total cost increases. If the process is stable and ^p, C
^ pk and C
^ pm can be calculated from
The values of C
the function ƒ is known, this technique can be utilised to
the raw data. If the management decides that the
estimate the total cost based upon the actual production
product quality should be maintained at least as good as
run long before the accounting department begins to ^p, C^ pk and C^ pm values from the raw data after a
the C
generate any data. If the process performance does not
quality improvement program is implemented from the
meet the management’s expectations, corrected actions
economical consideration, the variation in the denomi-
are required to bring the total cost to the best setting
nator should be proportionally reduced when tighter
point with the minimum value.
specification limits 2D’ in the numerator are applied. In
If the target costing concept is applied, the implied
fact, higher capability index values encourage companies
tolerance or specification limits can be derived from loss
to further reduce the variation in the denominator with
functions. Later, the specification limits can be linked
new specification limits 2D’ in the numerator.
through predetermined process capability index values, ^ p value remains constant when the quality
If the C
calculated from the original process data, to conventional
program is implemented along with the new specifica-
SPC charts to provide goal control limits to reduce com-
tion limits 2D’ from the loss function, the relationship is
mon causes of variation [5]. The goal control limits form
established as follows:
the foundation for directed, continuous improvement
efforts by considering the price concerns from the mar- 0 0
^ p ¼ 2D ¼ D
C ð24Þ
ketplace. Generally, Sauers [5] only applied the Taguchi 6^
r 0
r0
3^
loss function along with the Cp index and X -R control
charts to set up goal control limits by applying the target and
costing philosophy. To further strengthen and broaden D0
^0 ¼
r : ð25Þ
the use of the target costing philosophy, this paper takes ^p
3C
into account Spiring’s loss function and the Cpm index on
both X -R and X -S control charts. ^ pm value, either computed directly from the
If the C
The goal control limits derived from the target cost- raw data or given by management or engineers, is
ing concept are always tighter than the traditional con- achieved for new specification limits 2D’ from the loss
trol limits. That is, the values of r^ , S and R
in Eqs. 4, 5, function, the relationship is:
6, 7, 8, 9, 10, 11, 12, 13, 14, and 15 for the target costing 2D0 D0
concept are usually smaller compared with those values ^ pm ¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
C ¼ r ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð26Þ
2 2
for traditional control limits. For instance, the points 6 r 02
^ þ X T 3 r 02
^ þ X T
slightly below the upper control limit, which are in
control without any special causes of variation in SPC and
charts, might be indicated as the points outside the up- vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
per goal control limit. Taguchi and Deming have ex- u 02 2 u 2
u D u D2
pressed a similar idea that continuously reducing process ^¼
r 0 t X T ¼t
ln ð1 hÞ X T ;
variation in a product and process performance around 9C^2 72C ^2
pm pm
the nominal value for quality improvement is economi-
cal [9]. Therefore, the application of goal control limits is ð27Þ
210
Table 1 The modified data from Grant and Leavenworth [18] ^ pm value of 1.03 computed from the
by maintaining the C
raw data, then the value of D’ in [37D’, 37+D’] region
Subgroup Sample size for each Average Range
subgroup (n=5) ðX Þ (R) becomes:
1 36 35 34 33 32 34.0 4 8D02
2:99 ¼ 3:00 1 exp 2
2 32 32 33 33 31 32.6 4 13
3 33 33 35 33 35 33.8 2
4 32 33 33 32 35 33.0 3 and
5 32 34 37 37 35 35.0 5 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
6 32 32 31 33 33 32.2 2 ffi
7 33 33 36 32 31 33.0 5 0 169 2:99
D ¼ ln 1 ¼ 10:98 :
8 29 33 34 33 34 32.6 5 8 3
9 36 36 35 31 31 33.8 5
10 32 32 32 34 34 32.8 2 Since D’ is 10.98 and C ^ pm is maintained at 1.03, the new
11 34 38 35 34 38 35.8 4
12 32 34 36 35 36 34.6 4
process standard deviation r ^0 can be computed as fol-
13 36 37 34 30 33 34.0 7 lows by Eq. 27:
14 36 35 37 34 33 35.0 4 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
15 30 37 33 34 35 33.8 7 10:982
16 31 34 36 36 36 34.6 5 r 0
^ ¼ ð33:70 37Þ2 ¼ 1:318 ;
17 33 30 34 33 35 33.0 5 ð9Þ1:032
18 30 31 33 31 35 32.0 5
19 36 37 30 28 33 32.8 9 where D’ should be greater than or equal to 3(1.03)
20 33 35 35 39 36 35.6 6 |33.7037|=10.20 in this case. If D’ is smaller than
Total 674.0 93 ^0 becomes meaningless.
10.20, r
Average 33.70 4.65 Because r^0 is computed, the goal control limits for the
X and R charts can be calculated using Eqs. 28, 29, 30,
normal loss function, while the other is to set a new, and 31:
higher Cpm value to improve the quality. Besides, these
two approaches actually can work together to bring the UCLðX Þ ¼ pffiffi
33:70 þ 3 1:318
5
¼ 35:47
process mean to the target value by considering the cost CLðX Þ ¼ 33:70
factor as well as reducing process variation. LCLðX Þ ¼ pffiffi
33:70 3 1:318
5
¼ 31:93
UCLð RÞ ¼ ð2:115Þð2:326Þð1:318Þ ¼ 6:48
CLðRÞ ¼ ð2:326Þð1:318Þ ¼ 3:07
6 An example
and
An example borrowed from Grant and Leavenworth
[18] has been modified to demonstrate the proposed LCLð RÞ ¼ ð0Þð2:326Þð1:318Þ ¼ 0
concept. Assume the data represent deviations from the ^ p and C
^ pk values based upon the target
The new C
target value (T) of 37. The lower specification limit and
costing concept are
upper specification limit are set to 24 and 50, respec-
tively. The modified data are summarised in Table 1.
^ p ¼ 2D ¼ 2ð10:98Þ ¼ 2:78
C
Assume the process data are in statistical control and r0 6ð1:318Þ
6^
normally distributed, and assume that the loss function
in Eq. 1 has been established from observed data. The and
upper control limit, the centre line and the lower control
^ pk ðT þD0 ÞX X ðT þD0 Þ
limit of the X control chart using Eqs. 7, 8, and 9 are C ¼ min r0
3^
; 3^r0
36.38, 33.70 and 31.02, respectively, while 9.83, 4.65 and
0 are the values for the upper control limit, the centre ¼ min 47:9833:70 33:7026:02
3ð1:318Þ ; 3ð1:318Þ
line and the lower control limit of the R control chart ¼ min ð3:61; 1:94Þ ¼ 1:94
using Eqs. 4, 5, and 6, where r 2 ¼ 1:999,
^ ¼ R=d
d2=2.326, A2=0.577, D4=2.115, and D3=0 for n=5.
The values of C^p, C
^ pk and C
^ pm using Eqs. 19, 20, and 21 Table 2 The comparison between the traditional and goal control
are 2.17, 1.55 and 1.03, respectively. limits
Suppose the maximum-loss parameter K is $3.00 Traditional Goal
when D equals 13, which is (USLT) or (TLSL). The control limits control limits
formula of the loss function is
( !) UCLðX Þ 36.38 35.47
8ðy 37Þ2 CLðX Þ 33.70 33.70
Lð y Þ ¼ 3:00 1 exp : LCLðX Þ 31.02 31.93
132
UCL(R) 9.83 6.48
CL(R) 4.65 3.07
If the company decides to improve the quality by LCL(R) 0 0
reducing the original L(y)=$3.00 to L1(y)=$2.99 and
212
^ pm and r^ values
Table 3 The relationship between different C means the product quality has a smaller variation from
^ pm
C r^ the mean, and a smaller X T means to bring the
1.03 1.3178
process mean to the target value. It is important to note
1.04 1.2227 that if X T remains unchanged, the minimum D’ is
1.05 1.1226
1.06 1.0159 10.20, D’‡3(1.03)|33.7037|=10.20. Without bringing
1.07 0.9001 the process mean to the target value, the product quality
1.08 0.7711 cannot be relentlessly improved. Based upon Eq. 36, the
1.09 0.6203 ^ pm value is 10.98/(3|33.7037|)=1.109. The
1.10 0.4251 maximum C
1.11 <0 relationship between different C ^ pm and r^ is depicted in
Table 3 and Fig. 2, where the horizontal axis represents
the C^ pm value, and the vertical axis is the r^ value.
Clearly, if the target costing concept is applied by ^ pm value is set to 1.33 without bringing
reducing the original L(y)=$3.00 to L1(y)=2.99 and can If the new C
be achieved, the overall quality would be improved since the process mean to the target value, D’ is at 13.17, which
the new C^ p and C ^ pk values, 2.78 and 1.94, are higher than is impossible. However, if the process mean can be
the original C ^ p and C ^ pk values of 2.17 and 1.55. If the brought to, say, 35, then D’ using Eq. 36 is at 7.98, which
reflected normal loss function is used to reduce the has plenty of room for improvement. Figure 3 illustrates
variation from the target value, the C ^ p and C^ pk values are the maximum C ^ pm value for different X T values
expected to be increased because of smaller variation. when D’ equals 10.98, where the horizontal axis is the X
The goal control limits are tighter than the traditional value, and the vertical axis represents the maximum C ^ pm
X and R control limits, as shown in Table 2. The man- value. If the X value moves closer to the target value, the
ufacturer can apply goal control limits directly on the ^ pm value can be increased dramatically. In this case,
C
process to reduce the variation from the target value. In reducing variation might be as important as bringing the
this case, the range values of subgroups 13, 15 and 19 are process mean to the target value if the target costing
above UCL(R). The average values of subgroups 11 and philosophy is implemented.
20 are beyond UCLðX Þ: Therefore, an investigation
should be conducted.
From Eqs. 26 and 27, to increase the C ^ pm value, r^
7 Conclusions
and/or X T should become smaller. A smaller r^
This paper depicts the relationship among the loss
function, process capability indices and control charts
such that the goal control limits can be established to
reduce common causes of variation. This study has ex-
tended the target costing concept discussed by Sauers [5]
by further implementing the target value, the reflected
normal loss function to quantify the maximum loss and
the Cpm index, as well as incorporating the philosophy to
both X -R and X -S control charts.
Two major advantages of applying the target costing
philosophy are provided to relentlessly improve product
quality. The first advantage is to take the target value
and cost into consideration directly. The other advan-
^ pm values result in different r^ values
Fig. 2 Different C tage is to discuss that reducing process variation is not
enough to deal with quality problems. Moreover,
reducing process variation should be worked out along
with bringing the process mean to the target value.
References
3. Chen CH (1999) Specification limit under a quality loss func- 12. Wu H-H (2001) Performance problems of families of non-
tion. J Appl Stat 26(8):903–908 normal process capability indices. Qual Engin 13(3):383–388
4. Chen CH (2000) Specification limit under membership func- 13. Wu H-H (2001) Process capability indices for skewed process
tion. Qual Engin 12(4):519–521 data. Int J Indust Engin 8(3):210–219
5. Sauers DG (2000) Using the Taguchi loss function to reduce 14. Porter LJ, Oakland JS (1991) Process capability indices—an
common-cause variation. Qual Engin 12(2):245–252 overview of theory and practice. Qual Reliab Engin Int 7:437–
6. Leon RV, Wu CFJ (1992) A theory of performance measures in 448
parameter design. Statistica Sinica 2(2):335–357 15. Tse KK (1985) Marks & Spencer: an anatomy of Britain’s most
7. Tribus M, Szonyi G (1989) An alternative view of the Taguchi efficiently managed company. Pergamon, Oxford
approach. Qual Prog 22(5):46–52 16. Briggs A (1984) Marks & Spencer: 1884-1984: a century history
8. Spiring FA (1993) The reflected normal loss function. Canad J of Marks & Spencer: originator of penny bazaars. Octopus
Stat 21(3):321–330 Books, London
9. Gitlow H, Oppenheim A, Oppenheim R (1995) Quality man- 17. Roth G (1998-99) Process cost analysis via normal curve. Qual
agement: tools and methods for improvement: 2nd edition. Engin 11(1):11–13
McGraw-Hill/Irwin, New York 18. Grant EL, Leavenworth RS (1980) Statistical process control,
10. Kotz S, Lovelace CR (1998) Process capability indices in theory 5th edition. McGraw-Hill, New York
and practice. Arnold, Baltimore, MD
11. Wu H-H, Swain JJ (2001) A Monte Carlo comparison of
capability indices when processes are non-normally distributed.
Qual Reliab Engin Int 17(3):219–231