Journal of Applied Statistics

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/cjas20

A homogeneously weighted moving average

control chart for Conway–Maxwell Poisson
distribution

Olatunde Adebayo Adeoti, Jean-Claude Malela-Majika, Sandile Charles

Shongwe & Muhammad Aslam

To cite this article: Olatunde Adebayo Adeoti, Jean-Claude Malela-Majika, Sandile Charles
Shongwe & Muhammad Aslam (2021): A homogeneously weighted moving average
control chart for Conway–Maxwell Poisson distribution, Journal of Applied Statistics, DOI:
10.1080/02664763.2021.1937582

To link to this article: https://doi.org/10.1080/02664763.2021.1937582

Published online: 15 Jun 2021.

Submit your article to this journal

Article views: 66

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=cjas20
JOURNAL OF APPLIED STATISTICS
https://doi.org/10.1080/02664763.2021.1937582

A homogeneously weighted moving average control chart for

Conway–Maxwell Poisson distribution

Olatunde Adebayo Adeotia , Jean-Claude Malela-Majika b , Sandile Charles

Shongwe c and Muhammad Aslam d
a Department of Statistics, Federal University of Technology Akure, Akure, Nigeria; b Department of Statistics,
Faculty of Natural and Agricultural Sciences, University of Pretoria, Pretoria, South Africa; c Department of
Mathematical Statistics and Actuarial Sciences, Faculty of Natural and Agricultural Sciences, University of the
Free State, Bloemfontein, South Africa; d Department of Statistics, Faculty of Science, King Abdulaziz
University, Jeddah, Saudi Arabia

ABSTRACT ARTICLE HISTORY

The homogeneously weighted moving average (HWMA) control Received 21 September 2020
chart is a new memory-type chart that allocates a specific weight to Accepted 27 May 2021
the current sample and the remaining weight is distributed equally KEYWORDS
to the previous samples. In this paper, the HWMA control chart Attribute chart; count data;
is proposed for monitoring count data. This chart is based on the over-dispersed data;
Conway–Maxwell (COM) distribution, which can be used to model run-length characteristics;
under-spread and over-spread count data. The performance of the under-dispersed data
proposed chart is evaluated in terms of the average run-length (ARL),
standard deviation of the run-length (SDRL), median run-length
(MRL) as well as the expected ARL, SDRL and MRL values for both loca-
tion and dispersion shifts in the process. The sensitivity of the new
control chart is compared with those of some existing well-known
COM-Poisson memory-type control charts in terms of the out-of-
control ARL. The results of this study reveal that the proposed control
chart performs competitively well with the existing charts in detect-
ing shifts in the location and dispersion parameters in many situa-
tions. Numerical examples are given to demonstrate the application
of the proposed chart.

1. Introduction
Control charts are considered as powerful statistical process monitoring (SPM) tools for
maintaining the quality of products. A modern control chart was originally proposed by
W.A. Shewhart in the 1920s to monitor product quality characteristics. The Shewhart con-
trol charts used in the industry for process monitoring are classified into variable and
attribute charts. Attribute control charts are needed for monitoring count data while vari-
able control charts are needed for monitoring process on the basis of measurement of the
quality of interest. For variable Shewhart control charts, the most commonly used control
charts are the mean (X̄) and median (X̃) charts for monitoring process location and the
R, S and S2 charts are commonly used for monitoring process dispersion, see for instance

CONTACT Sandile Charles Shongwe [email protected]

© 2021 Informa UK Limited, trading as Taylor & Francis Group

2 O. A. ADEOTI ET AL.

[22]. However, attribute charts such as p, np, c and u charts are used for monitoring pro-
cess characteristics that are classified as conforming and non-conforming. While the p and
np control charts are constructed based on the assumption that the process characteristics
follow a binomial distribution; the c and u charts assume that the process characteristic of
interest follows a Poisson distribution, see for instance [14].
The Shewhart control charts are efficient in the detection of large shifts in the process
parameter; however, they are inefficient in the detection of small and moderate shifts. This
is because Shewhart control charts use only current information to make a decision about
the process. On the contrary, memory-type control charts use the current and past obser-
vations in making a decision about the process. In the SPM literature, there are four types of
memory-type control charts (i.e. cumulative sum (CUSUM), exponentially weighted mov-
ing average (EWMA), generally weighted moving average (GWMA) and homogenously
weighted moving average (HWMA)), these were respectively proposed by [1,24,26,34].
Quality control charts for monitoring attribute data have been developed by many
authors, see for instance [30] and [36]. It is important to note that control charts based
on the Poisson distribution are effective at monitoring process data that are equally dis-
persed (see for instance, [16] and [20]), but are ineffective when data are over-dispersed
or under-dispersed. Data are over-dispersed if the variance is greater than the mean; or
under-dispersed if the mean is greater than the variance; see the review by [30] for some
early contributions to the SPM literature on dispersed count data. The Conway–Maxwell
Poisson (also known as COM-Poisson) distribution is an expanded Poisson distribution
that was introduced by [18] to efficiently handle over-dispersed and under-dispersed data.
Sellers et al. [33] presented a survey of methods and applications of the COM-Poisson
model for count data.
Many authors have developed control charts based on the COM-Poisson distribution.
Sellers [32] proposed the Shewhart attribute control chart based on the COM-Poisson dis-
tribution using the k-sigma limits. Saghir et al. [31] illustrated the pitfalls of using [32]’s
symmetric k-sigma limits and proposed a COM-Poisson control chart based on asymmet-
ric probability limits. Next, Saghir and Lin [29,28] proposed the EWMA and CUSUM
control charts based on COM-Poisson distribution. For multivariate attribute data, [27]
proposed the corresponding multi-attributes COM-Poisson distribution control chart. The
aforementioned control charts are based on single sampling; hence, [10] and [12] proposed
COM-Poisson control charts based on repetitive sampling and multiple dependent state
sampling, respectively. Note that a modified EWMA statistic for monitoring the COM-
Poisson process is investigated in [15]. Alevizakos and Koukouvinos [9] proposed the use
of a progressive mean chart to monitor the COM-Poisson distribution. Next, [8] and [13]
proposed the double and hybrid EWMA control charts for monitoring the COM-Poisson
processes, respectively. Note that the double (hybrid) model entails the use of the same (dif-
ferent) smoothing parameter(s), twice. More recently, [17] investigated the performance
of the basic and double GWMA control charts to monitor the COM-Poisson distribution.
While the latter control charts are designed using classical statistics methodology, [11] pro-
posed a COM-Poisson distribution control chart using neutrosophic statistics, which can
be applied in uncertain environments.
Since the HWMA chart is the focus of this paper, it is worth mentioning that there is no
attribute HWMA chart that exists at the moment. So far, there is close to a dozen of publica-
tions on variable HWMA charts in the SPM literature. The first publication on the HWMA
 JOURNAL OF APPLIED STATISTICS 3

control charts (for monitoring process mean from a normally distributed data) is discussed
by [1], to overcome the problem of varying weights that are associated with the EWMA
chart. That is, the HWMA scheme allocates a specific weight to the current observation and
equal weights for all previous observations. Instead of simple random sampling, [23] pro-
posed the use of structured sampling techniques (i.e. ranked set sampling (RSS), extreme
RSS, median, RSS and neoteric RSS) to construct the HWMA chart to monitor the process
location. Next, [6] investigated the use of an auxiliary variable in the form of a regres-
sion estimator as an unbiased estimate of the process mean. As improvements to the basic
HWMA chart, [4] and [7] proposed the double and hybrid HWMA charts for monitor-
ing process mean, respectively. For a more efficient detection of small shifts in the process
mean, [3] proposed the use of a mixed HWMA-CUSUM chart. For linear profiles, [19] pro-
posed the HWMA chart to monitor the intercept, slope and variance parameters. For non-
parametric charts, [25] proposed the HWMA sign and signed-rank charts to monitor data
generated from skewed and symmetric distributions. Finally, for a multivariate scenario,
[5] and [2] proposed the HWMA control chart for monitoring process mean vector when
process parameters are known and unknown, respectively. Note that the HWMA-type
chart is recommended when the current observation (or information) of a quality charac-
teristic is more important and the contribution of the previous (or past) information is also
needed but treated in an equal proportion (see for example, [1]). For instance, when one is
interested to monitor the number of successful surgeries in a hospital, a monitoring scheme
that allocates more weight to the recent information without neglecting the past informa-
tion is needed. In this example, even though the contribution of past information is needed,
the current information is more important and needs more consideration. Thus, one can
decide to assign a specific weight to the current information and divide the rest equally
between the previous information. Thus, the HWMA-type chart can be used to solve the
problem.
Therefore, in this paper, a HWMA control chart based on a COM-Poisson distribution
is proposed for count data; this is hereafter denoted as the CMP-HWMA chart. The run-
length (RL) performance of the new control chart in detecting shifts on the process location
and dispersion, individually as well as simultaneously, is provided. The RL characteristics
of the proposed control chart are evaluated using Monte Carlo simulations and are also
compared with some existing COM-Poisson memory-type control charts.
The rest of the article is organized as follows. A brief review of some existing memory-
type COM-Poisson charts and the design of the proposed CMP-HWMA chart are given in
Section 2. Section 3 presents the performance evaluation of the proposed control chart to
detect shifts in the location parameter, the dispersion parameter and the latter two simul-
taneously. The comparison of the proposed CMP-HWMA control chart with the existing
COM-Poisson memory-type control charts as well as its application using real-life and
simulated data are given in Section 4 and finally, Section 5 provides concluding remarks,
limitations and future research ideas.

2. Control charts based on COM-Poisson distribution

This section presents a brief review of some control charts for monitoring count data based
on the COM-Poisson distribution and the design of the proposed control chart.
4 O. A. ADEOTI ET AL.

2.1. COM-Poisson distribution

The COM-Poisson distribution originally proposed by [18] has two parameters; namely,
the location parameter λ and the scale parameter η. The probability mass function (PMF)
of a COM-Poisson distribution is given by

λx
P(X = x; λ, η) = , λ > 0, η ≥ 0, x = 0, 1, 2, 3, . . . (1)
(x!)η Z(λ, η)

∞
where Z(λ, η) = λs /(s!)η is a normalizing constant, see also [32,33]. The COM-Poisson
s=0
distribution reduces to the Poisson distribution when η = 1, to the Bernoulli distribution
with parameter λ/(1 + λ) when η → ∞ and to the geometric distribution when η = 0
(with parameter 1 − λ, provided λ < 1). The mean and variance of the COM-Poisson
distribution are approximately given by

 1  1/
/η η−1 2 λ η
μx = λ − and σx = (2)
2η η

which are accurate for η ≤ 1 and λ > 10η . The COM-Poisson distribution is undefined
when η = 0 and λ ≥ 1; while it is under-dispersed when η > 1 and over-dispersed when
η < 1.

2.2. COM-Poisson EWMA control chart

Saghir and Lin [29] proposed a generalized EWMA control chart based on the COM-
Poisson distribution for monitoring count data. Let X be a random variable that follows
the COM-Poisson distribution with location parameter λ and dispersion parameter η. The
COM-Poisson EWMA (hereafter CMP-EWMA) charting statistic is defined as

Qi = ωXi + (1 − ω)Qi−1 , i = 1, 2, 3, . . . (3)

where ω (with 0 < ω ≤ 1) is the sensitivity or smoothing parameter.

Let λ0 and η0 be the in-control (IC) process parameters. Thus, the mean and variance
of the COM-Poisson distribution are given by
 1 
/
E(Xi ) = μ0 = λ0 η0 − η0 −1
2η0
and (4)
1/
λ0 η0
Var(Xi ) = σ02 = η0 ,

respectively. Thus, the mean and variance of the CMP-EWMA charting statistic are
respectively given by
μQ = μ0
and (5)
ω
σQ2 i = 2−ω {1 − (1 − ω)2i }σ02 ,
 JOURNAL OF APPLIED STATISTICS 5

where μ0 and σ02 are defined in (4). The time-varying control limits of the CMP-EWMA
control chart are given by

ω
UCLQi /LCLQi = μ0 ± LQ σ0 (1 − (1 − ω)2i ). (6)
2−ω
When the process has been running for a long time, i.e. when i → ∞, the control limits in
(6) reduce to:

ω
UCLQ /LCLQ = μ0 ± LQ σ0 , (7)
2−ω
where LQ is the control limit coefficient selected to attain a pre-determined IC average
run-length (ARL) value. The CMP-EWMA statistic is plotted against the control limits. A
process is considered to be IC if the plotted statistic lies within the control limits; otherwise,
the process is declared as out-of-control (OOC).

2.3. COM-Poisson CUSUM control chart

Saghir and Lin [28] proposed the CUSUM charts for the COM-Poisson distribution (here-
after CMP-CUSUM) for monitoring count data. CUSUM charts which consider shifts in
both location and dispersion parameters are proposed to detect shifts in either location
parameter λ or dispersion parameter η for the COM-Poisson process. Let X1 , X2 , X3 , . . . be
independent and identically distributed (IID) observations that follow the COM-Poisson
distribution with PMF defined in (1). Thus, the standard CMP-CUSUM charting statistics
at the ith sampling time are defined by

Ci+ = Max[0, (Xi − μ0 − KC ) + Ct−1

+
]
and (8)
Ci− = max[0, (μ0 − KC − Xi ) + Ct−1
−
],
⎛  1/2
⎞
1/
η
where KC = ⎝kc λ0 0 /η0 ⎠ is the reference parameter and the starting values C+
0

and C0− are typically taken to be equal to zero, i.e. C0+ = C0− = 0. The CMP-CUSUM chart
 1 1/2
+ − /η
gives a signal if Ci ≥ HC (or Ci ≥ HC ) where HC = hC λ0 /η0 0 .
However, [28]’s CMP-CUSUM control chart to detect shifts in a location parameter λ
(denoted λ-CUSUM chart) is based on the charting statistics

SLi = Min(0, KSi + SLi−1 )

and (9)
SUi = Max(0, KSi + Si−1 )
U

for i = 1, 2, 3, . . ., where SU L
i and Si are the upper and lower charting statistics, with
S0 = S0 = 0 and KSi is the reference value. The reference value KSi is defined as
U L

 
λ1 Z(λ0 , η0 )
KSi = xi Log + Log , (10)
λ0 Z(λ0 , η1 )
6 O. A. ADEOTI ET AL.

where (λ0 , η0 ) and (λ1 , η1 ) are the pairs of IC and OOC process parameters. The λ-
CUSUM chart for COM-Poisson distribution signals when SU i > HUS or Si < HLS where
U

HUS and HLS are the upper and lower control limits, respectively, which are determined to
achieve a pre-specified IC ARL (ARL0 ).
Also, the CMP-CUSUM control chart to detect shifts in the dispersion parameter η
(denoted η-CUSUM chart) is based on the charting statistics

WiL = Min(0, TWi + Wi−1

L )

and (11)
WiU = Max(0, TWi + Wi−1
U ),

where i = 1, 2, 3, . . ., WiU and WiL are the upper-sided and lower-sided CMP-CUSUM
statistics for the η-CUSUM chart, respectively, and W0U = W0L = 0. The reference value
TWi is given by
Z(λ0 , η0 )
TWi = Log(Xi !)(η0 − η1 ) + Log . (12)
Z(λ0 , η1 )
The η-CUSUM chart for COM-Poisson distribution gives a signal when WiU > HUW
or WiL < HLW where HUW and HLW are the upper and lower control limits, respectively,
called decision interval, which are determined to achieve a pre-specified ARL0 .

2.4. COM-Poisson GWMA chart

The COM-Poisson GWMA chart (denoted as CMP-GWMA) is the extension of the CMP-
EWMA chart characterized by the smoothing parameter q and an adjustment parameter
α, see [17]. When α = 1 and q = 1 − ω, the CMP-GWMA chart is equivalent to the CMP-
EWMA chart introduced in Section 2.2. The charting statistic of the CMP-GWMA chart
is given by


i
α α α
Gi = (q(j−1) − qj )Xi−j+1 + qj X0 , 0 ≤ q < 1, 0 < α ≤ 1. (13)
j=1

When the process is IC, the mean and variance of the CMP-GWMA statistic Gi are
defined by
1/
E(Gi ) = μG = λ0 η0 − η2η0 −1
0
and (14)
1/
η0
Var(Gi ) = σG2i = λ0η0 Qi ,

i α α
respectively, where Qi = (q(j−1) − qj )2 . Thus, the control limits of the CMP-GWMA
j=1
chart are given by
LCLGi /UCLGi = μG ± LG σGi , (15)
where LG > 0 represents the control chart coefficient selected to achieve a pre-specified
ARL0 . The CMP-GWMA chart gives a signal if the charting statistic defined in (13) plots
 JOURNAL OF APPLIED STATISTICS 7

beyond the control limits defined in (15). For more details on the GWMA-type control
charts, readers are referred to [21].

2.5. Design of the proposed COM-Poisson HWMA control chart

In this section, it is assumed that the number of nonconformities (defects) of a process
follows a COM-Poisson distribution with parameters λ0 and η0 when the process is IC.
Proceeding in a similar fashion as [29] and [10], given a random sample size of size n
at sampling point i, where the number of non-conformities is denoted as Xi , the CMP-
HWMA charting statistic is calculated as follows:

Hi = ωXi + (1 − ω)X̄i−1 , (16)

where ω is the smoothing constant (0 < ω ≤ 1) and X̄i−1 is the process mean of the pre-
vious (i − 1) charting statistics, where X̄0 = μ0 whenever i = 1; however, for i > 1, it is
given by

1 
i−1
X̄i−1 = Xt . (17)
i−1
t=1
To be able to derive the E(Hi ) and Var(Hi ), the following preliminary derivations are
done:
Using (4), i.e. E(Xi ) = μ0 and Var(Xi ) = σ02 , and since the COM-Poisson distributed
observations are IID, it means that there is no cross-correlation among the observations;
hence, Cov(Xi , Xr ) = 0, whenever i = r, where i and r are positive integers. Similarly,
Cov(Xi , X̄r ) = 0, whenever i = r. Thus, whenever i > 1, it follows that
 
1
E(X̄i−1 ) = E (X1 + X2 + . . . + Xi−1 )
i−1
1 1
= (E(X1 ) + E(X2 ) + . . . + E(Xi−1 )) = (i − 1)μ0 = μ0 ,
i−1 i−1
 2  2
1 1 σ2
Var(X̄i−1 ) = Var(X1 + X2 + . . . + Xi−1 ) = (i − 1)σ02 = 0 .
i−1 i−1 i−1
(18)

More specifically, at i = 1,

E(X̄0 ) = E(μ0 ) = μ0
and (19)
Var(X̄0 ) = Var(μ0 ) = 0.

From (4), (18) and (19), it follows that E(Hi ) at any sampling point i is given by

E(Hi ) = E(ωXi + (1 − ω)X̄i−1 ) = ωE(Xi ) + (1 − ω)E(X̄i−1 ) = (ω + 1 − ω)μ0 = μ0 .

(20)
Using (4) and (19), the variance at sampling point i = 1 is given by

Var(H1 ) = Var(ωX1 + (1 − ω)X̄0 ) = ω2 Var(X1 ) + (1 − ω)2 Var(μ0 ) = ω2 σ02 ; (21)

8 O. A. ADEOTI ET AL.

however, using (4) and (18), the variance at any sampling point i > 1 is given by

σ02
Var(Hi ) = ω2 Var(Xi ) + (1 − ω)2 Var(X̄i−1 ) = ω2 σ02 + (1 − ω)2 . (22)
i−1
Thus, it follows that the control limits of the proposed CMP-HWMA chart are given by
⎧ 
⎨ μ0 + LH ω2 σ02 if i = 1
UCLi = 
⎩μ + L ω2 σ 2 + (1 − ω)2 σ02 if i > 1
0 H 0 i−1
and ⎧  (23)
⎨ μ0 − LH ω2 σ02 if i = 1
LCLi = 
⎩μ − L ω2 σ 2 + (1 − ω)2 σ02 if i > 1,
0 H 0 i−1

with LH > 0 representing the control chart coefficient selected to achieve a pre-specified
ARL0 . Consequently, a process is declared as IC if LCLi < Hi < UCLi ; however, it is
declared as OOC if Hi ≤ LCLi or Hi ≥ UCLi .

3. Performance evaluation
In this section, the IC and OOC performances of the proposed CMP-HWMA control chart
for monitoring shifts in the process location, dispersion and both location and dispersion
simultaneously are investigated under all three scenarios, i.e. for under, equally and over-
dispersed data.

3.1. Performance metrics

In order to evaluate the performance of the proposed control chart, the ARL, i.e. the
expected number of sample points to be plotted on the chart before the first OOC sig-
nal, is used. When the process is IC, a large ARL value is desired; whereas, a small ARL
value is desirable when the process has shifted. In this paper, the pre-specified ARL0 is
fixed to 200. Most of the SPM research works developed control charts for COM-Poisson
distribution to monitor changes in the location parameter λ0 assuming that the disper-
sion parameter η0 is fixed. However, in this paper, we are concerned about monitoring
changes in the location parameter λ0 for a fixed value of η0 , changes in dispersion param-
eter η0 for a fixed value of λ0 and changes in both location and dispersion parameters. The
case of λ0 = 4 and η0 = 0.5 for over-dispersed data, 1 for equally dispersed data and 5 for
under-dispersed data is considered when the process is IC (i.e. λ1 = λ0 and η1 = η0 ) and
OOC (i.e. λ1 = δλ0 for shifts in the location parameter and η1 = γ η0 for shifts in the scale
parameter). For the shift sizes, we consider a shift δ < 1 and δ > 1 as well as γ < 1 and
γ > 1, respectively.
The ARL, standard deviation of the run-length (SDRL) and the median run-length
(MRL) of the proposed control chart are obtained using Monte Carlo simulations with
50,000 replications using SAS® 9.4-IML15.2. The characteristics of the computer used are:
Intel (R) Core (TM) i7-10510U CPU@ 1.80–2.30 GHz, 8GB of RAM and 64-bit operat-
ing system. The computation time of the IC characteristics of the RL is estimated at 9
 JOURNAL OF APPLIED STATISTICS 9

min on average, while, the one for the OOC characteristics of the RL is between 1 and
8 min depending on the size of the shift in the process mean. The computation of the RL
characteristics can be described as follows:
Evaluation of the attained ARL0 value:

(1) Specify the ARL0 , process parameters λ0 and η0 and number of simulations, r (here
r = 50,000).
(2) (a) Set the initial value of LH and calculate the control limits using Equation (23).
(b) If needed, increase (or decrease) LH and recalculate the control limits.
(3) Generate a random sample of size n (n = 1) following the COM-Poisson distribution
with parameters λ0 and η0 i.e. Xi ∼ COM–Poisson(λ0 , η0 ).
(4) Compute the CMP-HWMA charting statistic, Hi , using Equation (16).
(5) At the ith sampling time, if the charting statistic Hi lies within the control limits (i.e.
LCLi < Hi < UCLi ), go to Step 3; otherwise, if Hi ≤ LCLi or Hi ≥ UCLi , record the
number of samples until the OOC signal. This is one value of the RL vector.
(6) Perform Steps 3–5 repeatedly r times (i.e. 50,000 times). Calculate the attained ARL0
r
as ARL0 = RL0i /r where r is the number of simulations and the RL0i is the ith value
i=1
of the IC RL distribution. If the attained ARL0 value is much closer or equal to the pre-
specified ARL0 , record the value of LH and the control limits. Otherwise, go back to
Step 2(b) by increasing (decreasing) LH if the attained ARL0 is significantly greater
than the pre-specified value of the ARL0 .

Evaluation of OOC ARL (ARL1 ) values.

(7) Generate a random sample of size n (n = 1) from COM-Poisson distribution with a

shift in the location parameter λ1 = δλ0 and fixed η1 = η0 in case of a shift in the pro-
cess location only (η1 = γ η0 and λ1 = λ0 in case of a shift in the dispersion parameter
only) where δ and γ are shifts in location and dispersion parameters, respectively. Note
that λ1 = δλ0 and fixed η1 = γ η0 in case of a shift in both location and dispersion
parameters.
(8) Compute the CMP-HWMA charting statistic, Hi , using Equation (16).
(9) At the ith sampling time, if the charting statistic, Hi , lies within LCLi < Hi < UCLi , go
back to Step 7. However, if Hi ≤ LCLi or Hi ≥ UCLi , record the number of samples
until the occurrence of the OOC signal in the OOC RL (RL1 ) vector.
(10) Repeat Steps 7–9 r times (i.e. 50,000 times) and compute the ARL1 as ARL1 =
r
RL1i /r.
i=1

Note that the SDRL and MRL can be computed using the RL vector (or distribution). In
this paper, we used ω =0.05, 0.1, 0.25, 0.5 and 0.9 with pre-specified ARL0 value of 200 to
evaluate the performance of the proposed control chart for monitoring shifts in the loca-
tion parameter, shifts in dispersion parameter and shifts both in location and dispersion
parameters.
The above steps are summarized in Figure 1. Thus, Figure 1 shows how to compute the
characteristics of the RL distribution for a pre-specified ARL0 of 200.
10 O. A. ADEOTI ET AL.

Figure 1. Algorithm of the computation of the characteristics of the run-length of the proposed scheme.

To investigate the performance of the proposed CMP-HWMA chart for a range of shifts
(or overall performance), we made use of the expected ARL (EARL) metric. The EARL is
mathematically defined by (see for example [35])
δmax
1 
EARL = ARL(δ), (24)
δ=δmin
 JOURNAL OF APPLIED STATISTICS 11

where δmin and δmax are the lower and upper bound of the shift (δ) parameter, respectively,
ARL(δ) is the ARL1 value for a specific shift δ and represents the number of increments
between δmin and δmax . The expected MRL and SDRL (i.e. EMRL and ESDRL) can be cal-
culated in a similar way as the EARL, with the ARL(δ) replaced by MRL(δ) and SDRL(δ),
respectively. Where applicable, within the tables of this paper, overall performance metrics
denote the EARL, ESDRL and EMRL profile.

3.2. Proposed chart for monitoring shifts in the location parameter

In this section, the IC and OOC performances of the CMP-HWMA control chart are
investigated assuming that the dispersion parameter remains fixed IC and only a shift
in the process location has occurred (i.e. λ1 = δλ0 and η1 = η0 ) when δ = 0.6 (0.1)
2 which correspond to λ1 = 2.4 (0.4) 8, respectively, for λ0 = 4. The RL character-
istics of the CMP-HWMA chart for shifts in the location parameter (where λ0 = 4)
and a fixed dispersion parameter with three different levels of dispersion parameters
(where η0 ∈ {0.5, 1, 5}) are given in Tables 1–3, respectively, where the IC profiles
are shaded in grey. Note that these parameters were chosen in order to investigate the
performance of the proposed chart for all three scenarios; that is, under, equally and
over-dispersed data. It was also ensured that the parameters are similar to the ones used
in some related articles to make the comparison easy. The IC performance of the pro-
posed control chart is obtained when both shifts in the parameters are equal to one (i.e.
δ = 1 and γ = 1). From Tables 1–3 and Figures 2 and 3, the following findings are
observed:

• The CMP-HWMA chart is ARL-biased, i.e. the ARL1 values are larger than the attained
ARL0 corresponding values, for large ω values under over-dispersed data for very small
downward shifts 0.7 < δ < 1 corresponding to 2.8 < λ1 < 4. For equally and under-
dispersed data, the CMP-HWMA chart is ARL-biased for moderate and large values of
ω when 0.9 ≤ δ < 1 corresponding to 3.6 ≤ λ1 < 4 (see Tables 1–3 and Figure 2).
• The ARL1 values of CMP-HWMA control chart for over-dispersed data (i.e. η0 = 0.5)
are smaller than the ones of the CMP-HWMA chart for under and equally dispersed
data regardless of the size of the shift in the location parameter (see Tables 1–3 and
Figure 3).
• In terms of the ARL values, for specific δ the ARL1 values increase as ω increases.
When ω < 0.25, the CMP-HWMA chart is more sensitive under equally dispersed data
than under under-dispersed data when λ1 < 4. However, when λ1 > 4, the converse
is true. When ω > 0.25, the CMP-HWMA chart is sensitive under equally dispersed
data than under under-dispersed data except for very small downward shifts 0.9≤ δ <1
corresponding to 3.6 ≤ λ1 < 4 (see Tables 1–3 and Figures 2 and 3).
• In terms of the ARL, SDRL and MRL values, the proposed chart is efficient for detect-
ing upward and downward shifts as the smoothing parameter decrease. For instance,
under over-dispersed data, for δ = 1.1 corresponding to λ1 = 4.4 (i.e. upward shift),
when ω = 0.05 and 0.1, the proposed chart gives a signal on the 13th and 16th sub-
groups, respectively. This shows that the CMP-HWMA chart is more sensitive to small
smoothing parameters. However, for δ = 0.9 corresponding to λ1 = 3.6 (i.e. downward
 12
O. A. ADEOTI ET AL.
Table 1. The ARL, SDRL and MRL values of the CMP-HWMA chart with λ0 = 4 and η0 = 0.5 when ω ∈{0.05, 0.1, 0.25, 0.5, 0.9} for different shifts in the location
parameter with an ARL0 value of 200.
ω = 0.05 ω = 0.1 ω = 0.25 ω = 0.5 ω = 0.9

Shift(δ) ARL SDRL MRL ARL SDRL MRL ARL SDRL MRL ARL SDRL MRL ARL SDRL MRL
0.6 2.47 1.18 3.00 3.05 1.06 3.00 3.16 0.96 3.00 3.33 1.51 3.00 20.97 18.55 16.00
0.7 3.45 1.57 3.00 4.16 1.60 4.00 4.47 1.78 4.00 5.68 3.42 5.00 162.43 182.42 98.50
0.8 5.69 3.12 5.00 7.02 3.50 6.00 8.15 4.31 7.00 15.52 12.40 12.00 651.30 641.08 459.00
0.9 15.53 11.30 13.00 19.48 12.04 17.00 27.84 20.56 22.00 118.46 114.37 84.00 1274.18 1276.87 883.00
1.0 199.72 177.89 158.00 199.80 160.97 164.00 200.68 191.91 143.00 198.76 189.99 138.00 199.87 197.18 138.00
1.1 12.78 10.64 10.00 15.71 11.41 13.00 16.63 12.41 14.00 21.34 19.03 16.00 34.26 33.52 24.00
1.2 4.47 3.12 4.00 5.29 3.48 5.00 5.60 3.56 5.00 6.10 4.75 5.00 9.86 9.41 7.00
1.3 2.56 1.62 3.00 2.90 1.73 3.00 3.02 1.72 3.00 2.98 1.96 2.00 3.93 3.47 3.00
1.4 1.75 1.10 1.00 1.96 1.18 1.00 2.00 1.10 2.00 1.89 1.04 2.00 2.12 1.48 2.00
1.5 1.31 0.73 1.00 1.44 0.82 1.00 1.47 0.73 1.00 1.40 0.63 1.00 1.46 0.80 1.00
1.6 1.12 0.46 1.00 1.17 0.53 1.00 1.20 0.49 1.00 1.17 0.41 1.00 1.18 0.45 1.00
1.7 1.04 0.26 1.00 1.06 0.30 1.00 1.07 0.27 1.00 1.06 0.25 1.00 1.07 0.27 1.00
1.8 1.01 0.13 1.00 1.01 0.14 1.00 1.02 0.15 1.00 1.02 0.14 1.00 1.02 0.14 1.00
1.9 1.00 0.06 1.00 1.00 0.05 1.00 1.01 0.08 1.00 1.00 0.06 1.00 1.00 0.07 1.00
2.0 1.00 0.03 1.00 1.00 0.02 1.00 1.00 0.04 1.00 1.00 0.03 1.00 1.00 0.02 1.00
OPM 3.94 2.52 3.43 4.73 2.70 4.14 5.55 3.44 4.71 13.00 11.43 9.64 154.70 154.90 107.04
LH 2.115 2.502 2.784 2.870 2.902
Note: The IC profiles are shaded in grey.
 JOURNAL OF APPLIED STATISTICS 13

Table 2. The ARL, SDRL and MRL values of the CMP-HWMA chart with λ0 = 4 and η0 = 1 when
ω ∈{0.05, 0.1, 0.25} for different shifts in the location parameter with an ARL0 value of 200.
ω = 0.05 ω = 0.1 ω = 0.25

Shift(δ) ARL SDRL MRL ARL SDRL MRL ARL SDRL MRL
0.6 8.09 4.62 7 10.29 5.33 9 13.77 8 12
0.7 12.9 8.52 11 16.23 9.58 14 25.76 17.77 21
0.8 24.24 18.35 20 31.11 20.86 27 72.86 66.76 53
0.9 67.95 59.94 52 91.18 74.5 73 826.6 998.24 464.5
1.0 200.13 178.03 159.00 199.57 176.26 155.00 200.03 188.09 146.00
1.1 57.50 51.40 45.00 60.12 47.37 50.00 62.57 54.44 48.00
1.2 23.44 20.67 18.00 26.46 20.63 22.00 28.13 22.27 23.00
1.3 12.71 10.79 10.00 15.21 11.33 13.00 16.39 12.31 14.00
1.4 8.42 6.75 6.00 9.76 7.31 8.00 10.94 7.86 9.00
1.5 6.14 4.64 5.00 7.09 5.07 6.00 7.95 5.44 7.00
1.6 4.76 3.37 4.00 5.45 3.87 5.00 6.05 4.04 5.00
1.7 3.95 2.69 3.00 4.42 3.03 4.00 4.84 3.15 4.00
1.8 3.34 2.21 3.00 3.69 2.46 3.00 4.00 2.56 4.00
1.9 2.89 1.89 3.00 3.20 2.12 3.00 3.42 2.09 3.00
2.0 2.57 1.65 3.00 2.78 1.82 3.00 3.00 1.84 3.00
OPM 17.06 14.11 13.57 20.50 15.38 17.14 77.59 86.20 47.89
LH 2.098 2.483 2.831
Note: The IC profiles are shaded in grey; Bold represents the OOC events.

Table 3. The ARL, SDRL and MRL values of the CMP-HWMA chart with λ0 = 4 and η0 = 5 when
ω ∈{0.05, 0.1, 0.25} for different shifts in the location parameter with an ARL0 value of 200.
ω = 0.05 ω = 0.1 ω = 0.25

Shift(δ) ARL SDRL MRL ARL SDRL MRL ARL SDRL MRL
0.6 33.25 31.94 24 34.8 31.56 26 30.54 37.31 17
0.7 53.5 55.27 37 53.85 52.2 39 57.56 99.97 26
0.8 88.31 97.34 58 87.98 102.81 56 148.33 366.68 33
0.9 147.51 180.5 85 149.76 216.73 79 245.83 528.86 41
1.0 200.20 260.26 114.00 198.99 324.51 222.00 200.03 412.60 42.00
1.1 55.34 50.31 43 154.07 215.96 82 120 256.66 36
1.2 22.56 19.74 17 103.24 117.96 67 69.94 118.06 31
1.3 12.39 10.49 9 75.24 78.35 53.5 48.62 67.79 27
1.4 8.16 6.69 6 57.98 55.35 42 36.3 42.08 23
1.5 5.96 4.48 5 46.96 41.59 38 29.8 33.19 20
1.6 4.68 3.35 4 40.42 34.6 32 25.59 25.51 17
1.7 3.81 2.61 3 33.89 27.86 28 22.3 21.25 15
1.8 3.21 2.15 3 29.95 23.86 24 19.97 18.56 14
1.9 2.81 1.79 3 26.69 21.37 21 17.71 15.95 13
2.0 2.52 1.62 2 24.36 19.55 19 16.14 14.26 11
OPM 31.72 33.45 21.36 65.66 74.27 43.32 63.47 117.58 23.14
LH 2.056 2.209 2.204
Note: The IC profiles are shaded in grey.

shift), when ω = 0.05 and 0.1, the proposed chart gives a signal on the 16th and 19th
subgroups, respectively.
• The CMP-HWMA chart is more sensitive for upward shifts as compared to the
downward shifts for specific smoothing parameters. For instance, when ω = 0.1 with
δ = 0.9 and 1.1 corresponding to λ1 = 3.6 and 4.4 (i.e. for downward and upward
shifts), the CMP-HWMA chart gives a signal on the 19th and 16th subgroups, respec-
tively. The same pattern is also observed in terms of the SDRL and MRL values as well
(see Tables 1–3).
14 O. A. ADEOTI ET AL.

Figure 2. OOC performance of the CMP-HWMA chart when γ = 1, η0 ∈ {0.5,1,5} and λ0 = 4 for
different location shifts. (a) (λ0 , η0 ) = (4, 0.5), (b) (λ0 , η0 ) = (4, 1), (c) (λ0 , η0 ) = (4, 5).

• In terms of the ARL, SDRL and MRL values, the CMP-HWMA chart performs worst for
large smoothing parameters regardless of the type of data and the process parameters.
These findings remain valid in terms of the overall performance under equally and over-
dispersed data. However, for under-dispersed data, the performance in terms of the ARL
and EARL values increases in the interval 0.1 < ω ≤ 0.25.

3.3. Proposed chart for monitoring shifts in dispersion parameter

In this subsection, the IC and OOC performances of the CMP-HWMA control chart are
evaluated assuming the location parameter λ0 is known and there is a shift in the dispersion
parameter from η0 to η1 = γ η0 where γ < 1 and γ > 1. The RL characteristics of the
proposed control chart for different shifts in dispersion parameter η0 ∈{0.5, 1, 5} assuming
that the IC location parameter is fixed to 4 are displayed in Tables 4–6 and Figure 4. From
Tables 4–6 and Figure 4, the findings are summarized as follows:

• The proposed control chart for over-dispersed data (η0 = 0.5) is ARL-biased for large
smoothing parameters under small upwards shifts, i.e. when 1< γ ≤ 1.3 corresponding
to 0.5 < η1 ≤ 0.65 (see Table 4). However, the situation gets worst under equally and
under-dispersed data where the CMP-HWMA chart is ARL-biased for 0.25 ≤ ω ≤ 1
(Table 4 and Figure 4).
• The CMP-HWMA control chart is more sensitive in detecting both upward and down-
ward shifts as the smoothing constant decreases; but the proposed chart is more sensitive
in detecting downward shifts than upward shifts in the dispersion parameter. For
instance, when ω = 0.05, for an upward and downward dispersion shift of 1.1 and 0.9
corresponding to η1 = 0.55 and 0.45 for η0 = 0.5, the CMP-HWMA chart gives a signal
on the 11th and 6th sampling time, respectively. This is conclusion is also valid in terms
of the SDRL and MRL values.
• In terms of the EARL and ESDRL values (i.e. overall performance), the proposed control
chart is more sensitive under over-dispersed data followed by equally dispersed data.
The CMP-HWMA performs worst for under-dispersed data. The overall performance
decreases as the ω increases for upward shifts in the dispersion parameter. However, the
performance for specific shifts, i.e. in terms of the ARL, SDRL and MRL values, remains
 JOURNAL OF APPLIED STATISTICS 15

Figure 3. Performance comparison of the CMP-HWMA chart when λ0 = 4 and η0 ={0.5, 1, 5}, ω = 0.05
and 0.25 for different shifts in the location parameter. (a) ω = 0.05, (b) ω = 0.25.
16 O. A. ADEOTI ET AL.

Figure 4. OOC ARL performance of the CMP-HWMA scheme when γ = 1, η0 ∈ {0.5,1,4} and λ0 = 4 for
different dispersion shifts. (a) (λ0 , η0 ) = (4, 0.5), (b) (λ0 , η0 ) = (4, 1), (c) (λ0 , η0 ) = (4, 5).

almost the same under over-dispersed data for the downward shifts in the dispersion
parameter.
• In terms of the MRL values, when 0.05 ≤ ω ≤ 0.25, for under-dispersed data, the CMP-
HWMA control chart is sensitive for large values of ω. However, for equally and over-
dispersed data, the CMP-HWMA chart performs better for small ω values for upward
dispersion shifts; while its sensitivity under downward shifts is the same.
• The overall performance in terms of the EMRL values, for over- and equally dispersed
data, the CMP-HWMA chart is more sensitive for small smoothing parameters and less
sensitive for larger values. However, the converse is true for under-dispersed data.

3.4. The CMP-HWMA chart for monitoring both location and dispersion parameters
In this section, the performance of the proposed control chart when there is a shift in both
the location and dispersion parameters is investigated using simulation with 50,000 repli-
cations. Table 7 displays the RL characteristics of the CMP-HWMA control chart when the
location parameter has shifted from λ0 to λ1 and scale parameter from η0 to η1 where δ ∈
{0.7,0.8,0.9,1.0,1.1,1.2} and γ ∈{0.6,0.7,0.8,0.9,1,1.1,1.2,1.3,1.4,2}. The findings in Table 7
can be summarized as follows:

• The proposed control chart is ARL-biased for over-dispersed data for small upward
shifts when ω ≥ 0.5. This is also observed for equally and under-dispersed data (not
shown here to conserve space).
• In terms of the ARL, SDRL and MRL values, the CMP-HWMA control chart is more
sensitive to downward shifts compared to upward shifts in both process location and
dispersion considered simultaneously.
• Regardless of the type of data (here we only show for over-dispersed data to conserve
writing space), in terms of the overall performance (i.e. EARL, ESDRL and EMRL
values), the larger the smoothing parameter, the less sensitive the chart. The overall
performance is worst for very large smoothing parameters.
• For downward shifts, when the shift in the location parameter is kept fixed, the higher
(smaller) the shift in the dispersion parameter, the higher (smaller) the ARL, SDRL and
MRL values. However, for upward shifts, when the shift in the location parameter is kept
Table 4. The ARL, SDRL and MRL values of the CMP-HWMA chart with λ0 = 4 and η0 = 0.5 when ω ∈{0.05, 0.1, 0.25, 0.5, 0.9} for different shifts in the scale
parameter with an ARL0 value of 200.
ω = 0.05 ω = 0.1 ω = 0.25 ω = 0.5 ω = 0.9

Shift(γ ) ARL SDRL MRL ARL SDRL MRL ARL SDRL MRL ARL SDRL MRL ARL SDRL MRL
0.6 1.00 0.00 1.00 1.00 0.00 1.00 1.00 0.00 1.00 1.00 0.01 1.00 1.00 0.00 1.00
0.7 1.02 0.20 1.00 1.03 0.23 1.00 1.04 0.20 1.00 1.04 0.19 1.00 1.04 0.19 1.00
0.8 1.70 1.08 1.00 1.88 1.16 1.00 1.92 1.08 2.00 1.81 1.03 2.00 2.00 1.38 2.00
0.9 5.71 4.35 5.00 6.89 4.86 6.00 7.25 4.96 6.00 7.97 6.43 6.00 12.52 12.07 9.00
1 199.72 177.89 158.00 199.80 160.97 164.00 200.68 191.91 143.00 198.76 189.99 138.00 199.87 197.18 138.00
1.1 11.49 7.51 10.00 14.62 8.33 13.00 19.72 13.01 17.00 80.65 77.79 56.00 2285.21 2277.89 1595.50
1.2 5.22 2.55 5.00 6.42 2.89 6.00 7.46 3.58 7.00 14.69 11.26 12.00 1232.67 1259.37 845.00
1.3 3.68 1.51 4.00 4.40 1.57 4.00 4.82 1.81 4.00 6.90 4.23 6.00 529.62 590.17 336.00
1.4 2.98 1.21 3.00 3.59 1.10 3.00 3.77 1.18 4.00 4.61 2.36 4.00 73.59 66.41 55.00
1.5 2.58 1.08 3.00 3.18 0.92 3.00 3.26 0.85 3.00 3.52 1.53 3.00 44.65 41.83 32.00
1.6 2.28 1.03 3.00 2.90 0.83 3.00 2.95 0.69 3.00 2.98 1.10 3.00 30.81 29.33 22.00
1.7 2.07 0.98 2.00 2.70 0.81 3.00 2.75 0.62 3.00 2.64 0.86 2.00 22.43 20.89 16.00

JOURNAL OF APPLIED STATISTICS

1.8 1.88 0.94 2.00 2.56 0.81 3.00 2.64 0.58 3.00 2.43 0.68 2.00 17.53 16.15 13.00
1.9 1.72 0.89 1.00 2.43 0.84 3.00 2.52 0.56 3.00 2.27 0.56 2.00 14.13 12.54 10.00
2 1.60 0.83 1.00 2.35 0.84 3.00 2.44 0.54 2.00 2.18 0.48 2.00 11.88 10.61 9.00
OPM 3.21 1.73 3.00 4.00 1.80 3.79 4.54 2.12 4.21 9.62 7.75 7.29 305.65 309.92 210.46
LH 2.115 2.502 2.784 2.870 2.902
Note: The IC profiles are shaded in grey.

17
18 O. A. ADEOTI ET AL.

Table 5. The ARL, SDRL and MRL values of the CMP-HWMA chart with λ0 = 4 and η0 = 1 when
ω ∈{0.05, 0.1, 0.25} for different shifts in the scale parameter with an ARL0 value of 200.
ω = 0.05 ω = 0.1 ω = 0.25

Shift(γ ) ARL SDRL MRL ARL SDRL MRL ARL SDRL MRL
0.6 1.73 1.13 1 1.77 1.18 1 1.83 1.13 1
0.7 3.16 2.3 3 3.45 2.51 3 3.7 2.52 3
0.8 7.18 5.93 6 8.48 6.6 7 8.8 6.38 7
0.9 26.28 24.03 20 29.14 22.97 24 29.56 23.86 24
1.0 200.13 178.03 159.00 199.57 176.26 155.00 200.03 188.09 146.00
1.1 47.8 38.57 38 62.46 45.67 52 1821.5 4008.97 265
1.2 20.08 13.43 17 26.03 15.61 23 55.71 42.3 44
1.3 12.59 7.52 11 16.24 8.73 15 27.91 19 23
1.4 9.35 4.98 8 11.94 5.86 11 17.85 10.29 16
1.5 7.61 3.67 7 9.62 4.22 9 13.52 6.96 12
1.6 6.56 2.94 6 8.15 3.34 8 10.98 5.08 10
1.7 5.85 2.37 6 7.24 2.73 7 9.51 4.18 9
1.8 5.3 2.04 5 6.57 2.3 6 8.37 3.39 8
1.9 4.96 1.79 5 6.09 2 6 7.6 2.92 7
2.0 4.66 1.57 4 5.72 1.77 5 6.99 2.59 7
OPM 11.65 8.02 9.79 14.49 8.96 12.64 144.56 295.68 31.14
LH 2.098 2.483 2.831
Note: The IC profiles are shaded in grey.

Table 6. The ARL, SDRL and MRL values of the CMP-HWMA chart with λ0 = 4 and η0 = 5 when
ω ∈{0.05, 0.1, 0.25, 0.5, 0.9} for different shifts in the scale parameter with an ARL0 value of 200.
ω = 0.05 ω = 0.1 ω = 0.25

Shift(γ ) ARL SDRL MRL ARL SDRL MRL ARL SDRL MRL
0.6 10.06 11.78 5 11.76 10.3 9 8.25 7.4 6
0.7 21.43 24.53 13 21.55 19.99 15 14.16 14.2 9
0.8 44.65 52.56 29 41.81 42.12 30 25.96 30.92 15
0.9 104.11 123.64 62 90.68 108.11 57 58.7 98.79 26
1.0 200.20 260.26 114.00 198.99 324.51 222.00 200.03 412.60 42.00
1.1 169.29 194.15 107 191.36 286.58 94.5 527.51 1118.2 55
1.2 114.2 114.35 85 124.28 140.88 81 724.54 2036.11 56
1.3 87.82 78.84 69 89.92 86.62 65 255.83 711.31 52
1.4 75.73 62.55 62 73.13 64.37 56 128.33 240.57 47
1.5 66.88 53.97 56 65.28 54.5 53 96.42 156.22 42
1.6 61.93 48.09 53 60.03 48 48 79.78 114.51 41
1.7 59.13 44.62 51 56.96 43.36 48 69.08 90.05 39
1.8 57.18 42.3 50 54.81 41.82 46 64.77 81.02 36
1.9 55.9 41.43 49 53.06 39.7 45 62.67 76.97 36
2.0 54.97 40.09 48.5 52.26 39.13 45 60.12 70.86 36
OPM 70.23 66.64 52.82 70.49 73.25 49.46 155.44 346.22 35.43
LH 2.056 2.209 2.204
Note: The IC profiles are shaded in grey.

fixed, the higher (smaller) the shift in the dispersion parameter, the smaller (higher) the
ARL, SDRL and MRL values.

4. Numerical comparison and applications

In this section, a comparative study of the performance of the proposed CMP-HWMA
control chart with the CMP-EWMA, CMP-CUSUM and CMP-GWMA charts for location,
 JOURNAL OF APPLIED STATISTICS 19

Table 7. The ARL, SDRL and MRL values of the CMP-HWMA chart with λ0 = 4 and η0 = 0.5 when
ω ∈{0.05, 0.1, 0.25, 0.5} for different shifts in both location and scale parameters with an ARL0 value
of 200.
ω = 0.05 ω = 0.1 ω = 0.25 ω = 0.5

Shift(δ) Shift(γ ) ARL SDRL MRL ARL SDRL MRL ARL SDRL MRL ARL SDRL MRL
0.70 0.60 1.91 1.29 1.00 2.08 1.34 1.00 2.10 1.29 2.00 1.96 1.21 2.00
0.70 0.70 12.32 11.50 9.00 14.71 12.04 12.00 14.32 11.23 12.00 14.92 13.27 11.00
0.70 0.80 18.59 15.44 14.00 23.99 17.24 20.00 32.01 25.33 26.00 96.30 97.16 65.00
0.70 0.90 5.27 2.90 5.00 6.56 3.28 6.00 7.42 3.93 7.00 12.65 9.84 10.00
0.80 0.60 1.07 0.35 1.00 1.09 0.39 1.00 1.10 0.35 1.00 1.09 0.30 1.00
0.80 0.70 2.39 1.64 2.00 2.72 1.76 3.00 2.73 1.70 3.00 2.61 1.76 2.00
0.80 0.80 18.67 17.27 13.00 21.60 17.30 18.00 21.42 17.10 17.00 23.85 22.14 17.00
0.80 0.90 19.14 14.94 15.00 24.25 16.64 21.00 34.65 26.95 28.00 136.38 137.65 94.00
0.90 0.60 1.00 0.03 1.00 1.00 0.05 1.00 1.00 0.04 1.00 1.00 0.04 1.00
0.90 0.70 1.26 0.67 1.00 1.35 0.74 1.00 1.36 0.67 1.00 1.31 0.57 1.00
0.90 0.80 3.39 2.36 3.00 3.95 2.60 4.00 4.06 2.65 4.00 4.09 2.99 3.00
0.90 0.90 46.12 43.16 35.00 49.62 39.27 42.00 48.10 40.85 38.00 55.65 53.43 39.00
1.00 1.00 199.72 177.89 158.00 199.80 160.97 164.00 200.68 191.91 143.00 198.76 189.99 138.00
1.10 1.10 56.87 46.57 46.00 75.27 55.33 62.00 182.41 173.46 129.00 805.34 841.33 544.00
1.10 1.20 8.74 5.15 8.00 10.96 5.72 10.00 14.17 8.50 12.00 48.69 43.89 36.00
1.10 1.30 4.78 2.20 4.00 5.83 2.38 5.00 6.63 2.97 6.00 12.63 9.30 10.00
1.10 1.40 3.53 1.42 3.00 4.25 1.45 4.00 4.59 1.65 4.00 6.48 3.83 6.00
1.10 2.00 1.74 0.89 1.00 2.48 0.82 3.00 2.54 0.55 3.00 2.30 0.57 2.00
1.20 1.10 55.79 47.92 45.00 60.95 45.81 51.00 68.06 57.85 53.00 93.66 90.09 65.00
1.20 1.20 21.38 15.00 18.00 27.46 17.16 24.00 47.52 36.89 38.00 445.39 490.65 284.00
1.20 1.30 6.87 3.64 6.00 8.56 4.07 8.00 10.58 5.55 9.00 31.81 27.37 24.00
1.20 1.40 4.36 1.83 4.00 5.29 2.03 5.00 5.92 2.46 5.00 10.58 7.26 9.00
1.20 2.00 1.90 0.94 2.00 2.60 0.79 3.00 2.66 0.55 3.00 2.45 0.68 2.00
OPM 13.50 10.78 10.77 16.21 11.28 13.86 23.43 19.21 18.27 82.32 84.33 55.82
LH 2.115 2.502 2.784 2.870
Note: The IC profiles are shaded in grey.

dispersion and both location and dispersion shifts in the process parameters are presented
in terms of the ARL profile. Moreover, illustrative examples using real-life and simulated
data are provided to demonstrate the application and implementation of the proposed
CMP-HWMA chart.

4.1. Comparison study

In this comparison, the ARL values of CMP-EWMA and CMP-HWMA chart are computed
when ω = 0.05 and 0.1; while for the CMP-CUSUM chart, kC = 0.225 and 0.5 and for
the CMP-GWMA chart, (q, α) = (0.9, 07) and (0.95, 0.1) for a pre-specified ARL0 value
of 200. Tables 8–10 display the performance comparison of the competing charts when a
shift has occurred in the location, dispersion and both location and dispersion parameters,
respectively, as well as the optimal design parameters for over-dispersed data when λ0 = 4
and η0 = 0.5. From these Tables, the following is observed:

• Table 8 shows that when a shift occurs in location parameter only, the CMP-EWMA
chart is more sensitive than the competing charts for small and moderate location shifts
including large downward shifts followed by the proposed CMP-HWMA chart and
the CMP-GWMA chart, with ω = 0.05 and (q, α) = (0.9,0.7), respectively. For large
upward shifts in the process location, the CMP-EWMA and CMP-HWMA charts are
20 O. A. ADEOTI ET AL.

Table 8. Performance comparison of the CMP-HWMA chart and the existing CMP-EWMA, CMP-CUSUM
and CMP-GWMA charts when λ0 = 4 and η0 = 0.5 for different shifts in the location parameter with an
ARL0 value of 200.
EWMA HWMA CUSUM GWMA EWMA HWMA CUSUM GWMA
λ1 ω = 0.05 ω = 0.05 kC = 0.225 (q, α) =(0.95,0.1) ω = 0.1 ω = 0.1 kC = 0.5 (q, α) =(0.9,0.7)
2.4 2.19 2.47 5.20 10.89 2.47 3.05 3.82 2.75
2.8 3.17 3.45 6.65 22.41 3.56 4.16 5.16 3.96
3.2 5.54 5.69 9.89 60.36 6.41 7.02 8.70 7.01
3.6 15.75 15.53 22.27 276.52 19.71 19.48 29.24 19.98
4.0 202.38 199.72 199.43 201.66 200.66 199.80 202.05 201.63
4.4 12.72 12.78 18.81 29.47 14.10 15.71 19.21 13.93
4.8 4.11 4.47 7.92 8.84 4.59 5.29 6.35 4.72
5.2 2.25 2.56 4.96 3.82 2.44 2.90 3.69 2.57
5.6 1.56 1.75 3.56 2.15 1.65 1.96 2.58 1.73
6.0 1.24 1.31 2.80 1.47 1.28 1.44 2.01 1.34
6.4 1.09 1.12 2.31 1.19 1.11 1.17 1.66 1.15
6.8 1.03 1.04 1.99 1.06 1.04 1.06 1.39 1.05
7.2 1.01 1.01 1.75 1.02 1.01 1.01 1.19 1.02
7.6 1.00 1.00 1.52 1.00 1.00 1.00 1.08 1.00
8.0 1.00 1.00 1.31 1.00 1.00 1.00 1.02 1.00
EARL 3.83 3.94 6.50 30 4.38 4.73 6.22 4.52
LQ = 2.275 LH = 2.115 H = 7.282 LG = 3.274 Lq = 2.477 LH = 2.502 H = 4.193 LG = 2.571
Note: The IC profiles are shaded in grey.

Table 9. Performance comparison of the CMP-HWMA chart and the existing CMP-EWMA, CMP-CUSUM
and CMP-GWMA charts when λ0 = 4 and η0 = 0.5 for different shifts in the scale parameter with an
ARL0 value of 200.
EWMA HWMA CUSUM GWMA EWMA HWMA CUSUM GWMA
η1 ω = 0.05 ω = 0.05 kC = 0.225 (q, α) =(0.95,0.1) ω = 0.1 ω = 0.1 kC = 0.5 (q, α) =(0.9,0.7)
0.30 1.00 1.00 1.01 1.00 1.00 1.00 1.00 1.00
0.35 1.02 1.02 1.71 1.04 1.02 1.03 1.22 1.03
0.40 1.52 1.70 3.41 1.97 1.61 1.88 2.47 1.66
0.45 5.47 5.71 9.68 10.92 6.08 6.89 8.08 6.00
0.50 202.38 199.72 199.43 201.66 200.66 199.80 202.50 201.63
0.55 11.96 11.49 17.57 234.18 14.26 14.62 20.75 15.42
0.60 5.08 5.22 9.28 66.80 5.81 6.42 8.02 6.56
0.65 3.36 3.68 6.96 33.16 3.84 4.40 5.47 4.33
0.70 2.65 2.98 5.90 21.22 2.99 3.59 4.47 3.39
0.75 2.28 2.58 5.31 15.14 2.56 3.18 3.91 2.86
0.80 2.03 2.28 4.94 11.87 2.28 2.90 3.55 2.55
0.85 1.86 2.07 4.67 9.74 2.09 2.70 3.34 2.34
0.90 1.76 1.88 4.45 8.37 1.96 2.56 3.19 2.19
0.95 1.65 1.72 4.30 7.43 1.88 2.43 3.09 2.09
1.00 1.59 1.60 4.18 6.66 1.80 2.35 3.02 2.02
EARL 3.09 3.21 5.96 30.68 3.51 4.00 5.11 3.82
LQ = 2.275 LH = 2.115 H = 7.282 LG = 3.274 LQ = 2.477 LH = 2.502 H = 4.193 LG = 2.571
Note: The IC profiles are shaded in grey.

equivalent in performance. These charts perform better for small smoothing parame-
ters.
• Table 9 shows that when a shift occurs in the dispersion parameter only, regardless
of the size of the shift, the CMP-EWMA and CMP-HWMA charts are equivalent for
upward shifts in the process dispersion for small smoothing parameters followed by the
CMP-GWMA chart. However, for downward shifts, the competing charts are similar
 JOURNAL OF APPLIED STATISTICS 21

Table 10. Performance comparison of the CMP-HWMA chart and the existing CMP-EWMA, CMP-CUSUM
and CMP-GWMA charts when λ0 = 4 and η0 = 0.5 for different shifts in both location and scale
parameters with an ARL0 value of 200.
EWMA HWMA CUSUM GWMA EWMA HWMA CUSUM GWMA
λ1 η1 ω = 0.05 ω = 0.05 kC = 0.225 (q, α) =(0.95,0.1) ω = 0.1 ω = 0.1 kC = 0.5 (q, α) =(0.9,0.7)
3.20 0.30 1.05 1.07 1.88 1.06 1.05 1.09 1.35 1.07
3.20 0.35 2.10 2.39 4.55 3.02 2.08 2.72 3.34 2.31
3.20 0.40 18.14 18.67 25.39 28.46 18.17 21.6 24.32 18.14
3.20 0.45 19.86 19.14 26.99 263.97 19.77 24.25 35.93 23.90
3.60 0.30 1.00 1.00 1.18 1.00 1.00 1.00 1.02 1.00
3.60 0.35 1.19 1.26 2.47 1.33 1.19 1.35 1.8 1.25
3.60 0.40 3.07 3.39 6.25 5.17 3.10 3.95 4.74 3.43
3.60 0.45 45.76 46.12 56.52 63.38 45.37 49.62 59.29 44.51
4.00 0.50 202.38 199.72 199.43 201.66 200.66 199.80 202.5 201.63
4.40 0.55 65.08 56.87 80.50 820.56 90.04 75.27 161.38 80.07
4.40 0.60 8.88 8.74 14.02 179.28 10.59 10.96 14.57 11.66
4.40 0.65 4.6 4.78 8.61 61.63 5.28 5.83 7.21 6.00
4.40 0.70 3.24 3.53 6.74 32.90 3.69 4.25 5.28 4.18
4.40 0.75 1.67 1.74 4.32 21.21 1.68 2.48 3.10 3.34
4.80 0.80 59.86 55.79 75.90 21.28 69.29 60.95 94.21 3.26
4.80 0.85 22.97 21.38 30.25 15.72 29.88 27.46 50.78 2.81
4.80 0.90 6.86 6.87 11.54 12.71 8.18 8.56 11.04 2.53
4.80 0.95 4.14 4.36 7.97 10.52 4.77 5.29 6.61 2.34
4.80 1.00 1.78 1.90 4.48 9.07 1.77 2.60 3.21 2.21
EARL 15.07 14.39 20.53 86.24 17.61 17.18 27.18 11.89
LQ = 2.275 LH = 2.115 H = 7.282 LG = 3.274 LQ = 2.477 LH = 2.502 H = 4.193 LG = 2.571
Note: The IC profiles are shaded in grey.

in performance, except for the CMP-CUSUM chart with kC = 0.225. Note that the
CMP-GWMA chart performs worst when (q, α) = (0.95,0.1).
• Table 10 shows that when shifts occur in both location and dispersion parameters,
for upward shifts, the CMP-HWMA chart is superior to the competing charts; while,
for downward shifts, the CMP-EWMA and CMP-HWMA charts are similar in perfor-
mance. However, as the shift in the process location increases, the CMP-GWMA chart
with (q, α) = (0.9,0.7) becomes superior to all competing charts. It is worth mention-
ing that, in general, the GWMA charts’ design is relatively complex than the HWMA
and EWMA charts as there is an additional adjustment design parameter that needs to
be accounted for in the model; see [34] or [21].

Tables 11 and 12 present the performance summary of the competing charts considered
in this paper. For simplicity, the CMP-HWMA, CMP-EWMA and CMP-CUSUM charts
are denoted by H, E and C, respectively. In these tables, the symbol ‘H&E’ indicates that
the CMP-HWMA and CMP-EWMA charts are equivalent in performance and they out-
perform the CMP-CUSUM chart. The word ‘Similar’ indicates that all competing charts
perform similarly. Since the CMP-EWMA and the CMP-GWMA charts are equivalent for
α = 1, only the CMP-EWMA chart is considered in Tables 11 and 12. The comparison is
made when n = 1, 0 < ω ≤ 0.1 and 0.1 < ω ≤ 0.25 for a pre-specified ARL0 of 200. From
Tables 11 and 12, it can be noticed that the CMP-HWMA chart performs better in many
situations followed by the CMP-EWMA chart. Thus, the CMP-HWMA chart is mostly
recommended for under-dispersed data. For equally dispersed and over-dispersed data,
22 O. A. ADEOTI ET AL.

both the CMP-HWMA and CMP-EWMA charts are recommended. For more details in
the choice of the most effective chart, readers are recommended to used Tables 11 and 12.

4.2. Illustrative examples

In this section, two examples are provided to illustrate the application of the CMP-HWMA
chart. The first example is based on real-life data and the second one is based on simulated
data. Moreover, the sensitivity of the proposed control chart is compared to the ones of the
competing charts used in Section 4.1.

4.2.1. Illustrative example using real-life data

To illustrate the application and implementation of the proposed CMP-HWMA chart as
well as the one of the standard CMP-EWMA, CMP-CUSUM and CMP-GWMA charts,
data from Montgomery [22]’s page 310 are used. The data is based on monitoring the
number of nonconformities observed in 26 samples of 100 printed circuit boards. Before
the implementation of the proposed chart, it is important to investigate whether the Pois-
son distribution is appropriate for this example. Thus, the chi-square goodness-of-fit test
of a Poisson distribution was used for this purpose. It was found that the test statistic and p-
value were equal to 4.424 and 0.109, respectively. Therefore, at 5% level of significance the
null hypothesis of a Poisson distribution was not rejected since the p-value > 0.05, which
shows that the data follows a Poisson distribution. But, since the mean and variance are
different, the COM-Poisson distribution will be more appropriate. The parameters of the
COM-Poisson distribution are found using the maximum likelihood estimation technique
that yields λ̂0 = 2.8711 and η̂0 = 0.3652 showing over-dispersion in the data; therefore,
the standard Poisson distribution is not appropriate because of its assumption of equi-
dispersion. Thus, the COM-Poisson distribution is used and the mean and variance are
found to be 18.8506 and 48.88, respectively. For a smoothing constant of 0.5, kC = 0.5 and
(q, α) = (0.9, 0.7), the optimal design parameters of the CMP-HWMA, CMP-EWMA,
CMP-CUSUM and CMP-GWMA chart are found to be equal to 2.879, 2.817, 4.231 and
2.571, which yield attained ARL0 values of 200.33, 199.20, 201.44 and 201.63, respectively.
The charting statistics and control limits of the competing charts are displayed in Table 13
and Figure 5. Thus, from Table 13 and Figure 5, it can be observed that the three control
charts do not give a signal. However, at the 20th sampling time, the charting statistic of
the CMP-HWMA chart is much closer to its corresponding UCL. However, no charting
statistics of the CMP-EWMA and CMP-CUSUM charts are as close as possible to their cor-
responding control limits. This reveals that even though the three charts did not signal, the
CMP-HWMA chart is more sensitive than the CMP-EWMA and CMP-CUSUM charts in
this situation. However, the CMP-GWMA chart gives a signal on the 20th subgroup. This
shows that the CMP-GWMA chart is superior to the competing charts in this particular
case.

4.2.2. Illustrative example using simulated data

In this example, simulated data from [17] are used to illustrate the detection ability of the
CMP-HWMA chart. Moreover, the sensitivity of the proposed CMP-HWMA chart is also
compared to the ones of the standard CMP-EWMA, CMP-CUSUM and CMP-GWMA
charts for a pre-specified ARL0 value of 200. The quality characteristic is assumed to be
Table 11. Performance summary of CMP-HWMA, CMP-CUSUM and CMP-EWMA control charts when 0 < ω ≤ 0.1.
Shift in the location parameter only

Small Moderate Large Small-to-moderate Moderate-to-large Small-to-large

Type of data Downward Upward Downward Upward Downward Upward Downward Upward Downward Upward Downward Upward
Under-dispersed H&E H H&E H H E H&E H C H H&E H
Equally dispersed H H H&E E H&E E H&E H H&E E H H&E
Over-dispersed H&E H&E H&E H&E H&E H&E H&E H&E H&E H&E H&E H&E
Shift in the dispersion parameter only
Under-dispersed H&E H H H H&E H H&E H H H H&E H
Equally dispersed E H H&E H H&E H&E H H H&E H&E H&E H&E
Over-dispersed H&E E Similar E H&E H&E H&E H&E H&E H&E H&E H&E
Shift in both the location and dispersion parameters
Under-dispersed H&E H H&E H H E H&E H H H H&E H

JOURNAL OF APPLIED STATISTICS

Equally dispersed H H E H E H H&E H E H H H
Over-dispersed H&E H&E H&E H&E H&E H&E H&E H&E H&E H&E H&E H&E
Note: CMP-HWMA = H, CMP-EWMA = E, CMP-CUSUM = C, Similar = All competing charts are almost equivalent, H&E = CMP-HWMA and CMP-EWMA are similar, the GWMA and the EWMA are
equivalent for α = 1.

23
 24
O. A. ADEOTI ET AL.
Table 12. Performance summary of CMP-HWMA, CMP-CUSUM and CMP-EWMA control charts when 0.1 < ω ≤ 0.25.
Shift in the location parameter only

Small Moderate Large Small-to-moderate Moderate-to-large Small-to-large

Type of data Downward Upward Downward Upward Downward Upward Downward Upward Downward Upward Downward Upward
Under-dispersed H H&E H&E H H H H&E H C H H&E H
Equally dispersed C H H E H E H H H E C H
Over-dispersed H H&E E H&E H&E H&E H&E H&E H&E H&E H&E H&E
Shift in the dispersion parameter only
Under-dispersed H&E H H&E H H E H&E H H H H&E H
Equally dispersed H E Similar H Similar H H&E E Similar H H&E H
Over-dispersed H&E H Similar H&E Similar E H&E H H&E E H&E H
Shift in both the location and dispersion parameters
Under-dispersed H&E H&E H&E H H&E H H&E H H H H&E H
Equally dispersed H H E E E H E H&E E H E H
Over-dispersed H H Similar E Similar H H H H H H H
Note: CMP-HWMA = H, CMP-EWMA = E, CMP-CUSUM = C, Similar = All competing charts are almost equivalent, H&E = CMP-HWMA and CMP-EWMA are similar, the GWMA and the EWMA
are equivalent for α = 1.
Table 13. Illustrative example of the CMP-EWMA, CMP-CUSUM, CMP-GWMA and CMP-HWMA charts using real-life data from Montgomery ([22], page 310) when
ω = 0.5 for a pre-specified ARL0 value of 200.
CMP-EWMA CMP-CUSUM CMP-GWMA CMP-HWMA

No. Hi LCLQi Qi UCLQi Signal Ci+ H Ci+ Signal LCLGi Gi UCLGi Signal LCLi Hi UCLi Signal
1 21 9.00 19.93 28.70 No 0.00 29.67 0.00 No 15.05 16.95 17.95 No 8.79 19.93 28.91 No
2 24 7.84 21.96 29.86 No 0.00 29.67 1.67 No 14.82 17.51 18.18 No 4.62 22.50 33.08 No
3 16 7.57 18.98 30.13 No 0.00 29.67 0.00 No 14.69 17.09 18.31 No 6.52 19.25 31.18 No
4 12 7.50 15.49 30.20 No 3.32 29.67 0.00 No 14.61 16.54 18.39 No 7.23 16.17 30.47 No
5 15 7.49 15.25 30.22 No 3.64 29.67 0.00 No 14.54 16.52 18.46 No 7.60 16.63 30.10 No
6 5 7.48 10.12 30.22 No 13.96 29.67 0.00 No 14.49 15.44 18.51 No 7.83 11.30 29.88 No
7 28 7.48 19.06 30.22 No 1.28 29.67 5.67 No 14.45 17.09 18.55 No 7.98 21.75 29.72 No
8 20 7.48 19.53 30.22 No 0.00 29.67 3.34 No 14.41 17.08 18.59 No 8.09 18.64 29.61 No
9 31 7.48 25.27 30.22 No 0.00 29.67 12.00 No 14.38 18.33 18.62 No 8.18 24.31 29.53 No
10 25 7.48 25.13 30.22 No 0.00 29.67 14.67 No 14.36 18.49 18.64 No 8.24 22.06 29.46 No
11 20 7.48 22.57 30.22 No 0.00 29.67 12.34 No 14.33 18.27 18.67 No 8.30 19.85 29.41 No
12 24 7.48 23.28 30.22 No 0.00 29.67 14.01 No 14.31 18.65 18.69 No 8.34 21.86 29.36 No
13 16 7.48 19.64 30.22 No 0.00 29.67 7.67 No 14.30 18.10 18.70 No 8.38 18.04 29.33 No
14 19 7.48 19.32 30.22 No 0.00 29.67 4.34 No 14.28 18.16 18.72 No 8.41 19.38 29.29 No
15 10 7.48 14.66 30.22 No 5.32 29.67 0.00 No 14.27 17.25 18.73 No 8.43 14.86 29.27 No
16 17 7.48 15.83 30.22 No 3.64 29.67 0.00 No 14.26 17.44 18.74 No 8.46 18.03 29.24 No

JOURNAL OF APPLIED STATISTICS

17 13 7.48 14.42 30.22 No 5.96 29.67 0.00 No 14.25 17.03 18.75 No 8.48 15.97 29.22 No
18 22 7.48 18.21 30.22 No 0.00 29.67 0.00 No 14.24 17.68 18.76 No 8.49 20.29 29.21 No
19 18 7.48 18.10 30.22 No 0.00 29.67 0.00 No 14.23 17.59 18.77 No 8.51 18.39 29.19 No
20 39 7.48 28.55 30.22 No 0.00 29.67 16.67 No 14.22 19.69 18.78 Yes 8.52 28.87 29.18 No
21 30 7.48 29.28 30.22 No 0.00 29.67 24.34 No 14.21 20.00 18.79 Yes 8.54 24.88 29.16 No
22 24 7.48 26.64 30.22 No 0.00 29.67 26.00 No 14.21 19.86 18.79 Yes 8.55 22.12 29.15 No
23 16 7.48 21.32 30.22 No 0.00 29.67 19.67 No 14.20 19.14 18.80 Yes 8.56 18.20 29.14 No
24 19 7.48 20.16 30.22 No 0.00 29.67 16.34 No 14.20 19.09 18.80 Yes 8.57 19.61 29.13 No
25 17 7.48 18.58 30.22 No 0.00 29.67 11.01 No 14.19 18.81 18.81 Yes 8.58 18.58 29.12 No
26 15 7.48 16.79 30.22 No 0.32 29.67 3.67 No 14.19 18.43 18.81 Yes 8.59 17.52 29.11 No
Note: Bold represents the OOC events.

25
26 O. A. ADEOTI ET AL.

over-dispersed with (λ0 , η0 ) = (4, 0.5). The design parameters of the CMP-EWMA, CMP-
CUSUM, CMP-GWMA and CMP-HWMA charts are (ω, LQ ) = (0.05, 2.231), (kC , h) =
(0.5, 4.193), (q,α,LG ) = (0.9,0.7,2.571) and (ω, LH ) = (0.05, 2.115) which yield attained
ARL0 values of 202.38, 202.05, 201.63 and 199.72, respectively. The charting statistics and
control limits of the competing charts are displayed in Table 14 and Figure 6. Thus, from
Table 14 and Figure 6, it can be seen that the CMP-EWMA, CMP-CUSUM, CMP-GWMA
and CMP-HWMA charts give a signal for the first time on the 35th, 46th, 46th and 28th
subgroups, respectively. These findings reveal that in this case, the CMP-HWMA chart is
more sensitive than the competing charts followed by the CMP-EWMA chart.

5. Conclusions, limitations and future research ideas

Most of the industrial and non-industrial processes are monitored using variable control
charts. When the quality characteristic of the process is not measurable, attribute control
charts can be used to monitor the process. In this paper, a CMP-HWMA chart is proposed
to monitor processes based on count data for equally, over- and under-dispersed data. The
performance of the proposed chart in detecting shifts in the location or / and dispersion
parameter(s) is investigated in terms of specific and overall performance measures based
on the ARL, SDRL and MRL metrics. The results reveal that the proposed chart performs

Figure 5. Comparison of the sensitivities of the CMP-EWMA, CMP-CUSUM, CMP-GWMA and CMP-
HWMA charts using real-life data for a pre-specified ARL0 value of 200. (a) CMP-EWMA chart, (b)
CMP-CUSUM chart, (c) CMP-GWMA chart, (d) CMP-HWMA chart.
 JOURNAL OF APPLIED STATISTICS 27

better for small smoothing parameters and its performance is worst under large smooth-
ing parameters. This confirms the findings reported by many other researchers (see, for
example, [8] and [9]) for other different types of memory-type charts. The CMP-HWMA
chart is more sensitive for over-dispersed data and insensitive for under-dispersed data.
When compared to a number of existing COM-Poisson memory-type control charts, it
is found that the proposed CMP-HWMA control chart outperforms the CMP-EWMA,
CMP-CUSUM and CMP-GWMA charts in many cases. Note that the CMP-GWMA chart
can be adjusted using the extra design parameter α so that its performance reaches an
acceptable level, but the additional parameter brings in complexity in the design of the
GWMA chart. The results from this paper also confirm the findings in the SPM litera-
ture that the performances of the CMP-CUSUM, CMP-EWMA and other COM-Poisson
memory-type charts are relatively worst under equally dispersed data, see for instance,
[30]. This is also observed on the proposed CMP-HWMA chart. Thus, it is recom-
mended to use the proposed control chart for over-dispersed data using small smoothing
parameters.
This study was carried out under the assumption that the quality characteristic of the
process follows a COM-Poisson distribution with under, equally, or over-dispersed data.
In case this assumption is not met, the performance of the proposed chart must be re-
examined.

Figure 6. Comparison of the sensitivities of the CMP-EWMA, CMP-CUSUM, CMP-GWMA and CMP-
HWMA charts using simulated data when ω = 0.05 for a pre-specified ARL0 of 200. (a) CMP-EWMA chart,
(b) CMP-CUSUM chart, (c) CMP-GWMA chart, (d) CMP-HWMA chart.
28 O. A. ADEOTI ET AL.

Table 14. Illustrative example of the CMP-EWMA, CMP-CUSUM, CMP-GWMA and CMP-HWMA charts
using simulated data when ω = 0.05 for a pre-specified ARL0 value of 200.
CMP-EWMA CMP-CUSUM CMP-GWMA CMP-HWMA

No. Xi LCLQi Qi UCLQi OOC Ci+ H Ci+ OOC LCLGi Gi UCLGi OOC LCLi Hi UCLi OOC
1 12 15.86 16.28 17.14 No 1.67 23.72 0.00 No 15.05 16.05 17.95 No 15.90 16.28 17.10 No
2 30 15.61 16.96 17.39 No 0.00 23.72 10.67 No 14.82 17.59 18.18 No 5.12 12.90 27.88 No
3 12 15.44 16.71 17.56 No 1.67 23.72 3.34 No 14.69 16.62 18.31 No 8.44 20.55 24.56 No
4 17 15.30 16.73 17.70 No 0.00 23.72 1.01 No 14.61 16.74 18.39 No 9.91 17.95 23.09 No
5 15 15.19 16.64 17.81 No 0.00 23.72 0.00 No 14.54 16.55 18.46 No 10.79 17.61 22.21 No
6 23 15.10 16.96 17.90 No 0.00 23.72 3.67 No 14.49 17.24 18.51 No 11.38 17.49 21.62 No
7 21 15.02 17.16 17.98 No 0.00 23.72 5.34 No 14.45 17.41 18.55 No 11.82 18.31 21.18 No
8 13 14.96 16.95 18.04 No 0.67 23.72 0.00 No 14.41 16.79 18.59 No 12.16 18.29 20.84 No
9 25 14.90 17.36 18.10 No 0.00 23.72 5.67 No 14.38 17.69 18.62 No 12.44 18.23 20.56 No
10 9 14.85 16.94 18.15 No 4.67 23.72 0.00 No 14.36 16.55 18.64 No 12.66 18.18 20.34 No
11 18 14.81 16.99 18.19 No 0.34 23.72 0.00 No 14.33 16.90 18.67 No 12.86 17.72 20.14 No
12 21 14.77 17.19 18.23 No 0.00 23.72 1.67 No 14.31 17.30 18.69 No 13.02 17.89 19.98 No
13 23 14.73 17.48 18.27 No 0.00 23.72 5.34 No 14.30 17.73 18.70 No 13.16 18.25 19.84 No
14 19 14.70 17.56 18.30 No 0.00 23.72 5.01 No 14.28 17.63 18.72 No 13.29 18.42 19.71 No
15 15 14.67 17.43 18.33 No 0.00 23.72 0.69 No 14.27 17.25 18.73 No 13.40 18.26 19.60 No
16 17 14.65 17.41 18.35 No 0.00 23.72 0.00 No 14.26 17.26 18.74 No 13.50 18.14 19.50 No
17 19 14.63 17.49 18.37 No 0.00 23.72 0.00 No 14.25 17.42 18.75 No 13.60 18.17 19.40 No
18 28 14.61 18.01 18.39 No 0.00 23.72 8.67 No 14.24 18.41 18.76 No 13.68 18.67 19.32 No
19 17 14.59 17.96 18.41 No 0.00 23.72 6.34 No 14.23 17.90 18.77 No 13.76 18.64 19.24 No
20 27 14.58 18.41 18.42 No 0.00 23.72 14.01 No 14.22 18.74 18.78 No 13.82 19.05 19.18 No
21 16 14.56 18.29 18.44 No 0.00 23.72 10.69 No 14.21 18.11 18.79 No 13.89 18.90 19.11 No
22 6 14.55 17.68 18.45 No 7.67 23.72 0.00 No 14.21 16.87 18.79 No 13.95 18.26 19.05 No
23 16 14.54 17.60 18.46 No 5.34 23.72 0.00 No 14.20 17.13 18.80 No 14.00 18.20 19.00 No
24 23 14.53 17.87 18.47 No 0.00 23.72 3.67 No 14.20 17.81 18.80 No 14.06 18.46 18.94 No
25 23 14.52 18.12 18.48 No 0.00 23.72 7.34 No 14.19 18.17 18.81 No 14.10 18.65 18.90 No
26 16 14.51 18.02 18.49 No 0.00 23.72 4.01 No 14.19 17.75 18.81 No 14.15 18.47 18.85 No
27 24 14.50 18.32 18.50 No 0.00 23.72 8.69 No 14.18 18.37 18.82 No 14.19 18.78 18.81 No
28 20 14.50 18.40 18.50 No 0.00 23.72 9.36 No 14.18 18.31 18.82 No 14.23 18.77 18.77 Yes
29 16 14.49 18.28 18.51 No 0.00 23.72 6.03 No 14.17 17.95 18.83 No 14.27 18.61 18.73 No
30 21 14.49 18.42 18.51 No 0.00 23.72 7.70 No 14.17 18.27 18.83 No 14.31 18.77 18.69 Yes
31 14 14.48 18.19 18.52 No 0.00 23.72 2.37 No 14.17 17.72 18.83 No 14.34 18.50 18.66 No
32 18 14.48 18.19 18.52 No 0.00 23.72 1.04 No 14.16 17.84 18.84 No 14.37 18.55 18.63 No
33 19 14.47 18.23 18.53 No 0.00 23.72 0.72 No 14.16 17.95 18.84 No 14.40 18.58 18.60 No
34 21 14.47 18.36 18.53 No 0.00 23.72 2.39 No 14.16 18.21 18.84 No 14.43 18.70 18.57 Yes
35 25 14.47 18.70 18.53 Yes 0.00 23.72 8.06 No 14.16 18.77 18.84 No 14.46 18.96 18.54 Yes
36 8 14.47 18.16 18.53 No 5.67 23.72 0.00 No 14.16 17.43 18.84 No 14.49 18.29 18.51 No
37 17 14.46 18.10 18.54 No 2.34 23.72 0.00 No 14.15 17.65 18.85 No 14.51 18.45 18.49 No
38 20 14.46 18.20 18.54 No 0.00 23.72 0.67 No 14.15 17.94 18.85 No 14.54 18.56 18.46 Yes
39 13 14.46 17.94 18.54 No 0.67 23.72 0.00 No 14.15 17.39 18.85 No 14.56 18.25 18.44 No
40 14 14.46 17.74 18.54 No 0.34 23.72 0.00 No 14.15 17.20 18.85 No 14.58 18.17 18.42 No
41 18 14.45 17.75 18.55 No 0.00 23.72 0.00 No 14.15 17.42 18.85 No 14.61 18.26 18.39 No
42 13 14.45 17.52 18.55 No 0.67 23.72 0.00 No 14.14 17.00 18.86 No 14.63 18.00 18.37 No
43 12 14.45 17.24 18.55 No 2.34 23.72 0.00 No 14.14 16.66 18.86 No 14.65 17.84 18.35 No
44 25 14.45 17.63 18.55 No 0.00 23.72 5.67 No 14.14 17.71 18.86 No 14.67 18.35 18.33 Yes
45 21 14.45 17.80 18.55 No 0.00 23.72 7.34 No 14.14 17.84 18.86 No 14.69 18.30 18.31 Yes
46 38 14.45 18.81 18.55 Yes 0.00 23.72 26.01 Yes 14.14 19.72 18.86 Yes 14.70 19.21 18.30 No
47 20 14.45 18.87 18.55 Yes 0.00 23.72 26.69 Yes 14.14 19.05 18.86 Yes 14.72 18.72 18.28 Yes
48 23 14.45 19.07 18.55 Yes 0.00 23.72 30.36 Yes 14.14 19.23 18.86 Yes 14.74 18.90 18.26 Yes
49 27 14.45 19.47 18.55 Yes 0.00 23.72 38.03 Yes 14.14 19.77 18.86 Yes 14.75 19.18 18.25 Yes
50 19 14.45 19.45 18.55 Yes 0.00 23.72 37.70 Yes 14.14 19.33 18.86 Yes 14.77 18.94 18.23 Yes
Note: The IC profiles are shaded in grey; Bold represents the OOC events.

For future research purpose, we intend to investigate the performance of the double
HWMA (DHWMA) and hybrid HWMA (HHWMA) charts to monitor differently dis-
persed count data and check whether they will outperform the proposed and other existing
COM-Poisson memory-type control charts.
 JOURNAL OF APPLIED STATISTICS 29

Acknowledgements
The authors would like to thank the associate editor and reviewers for their valuable suggestions
that improved the paper. The second author also thanks the Department of Research and Innovation
(DRI) at the University of Pretoria for their support in terms of Research Development Programme.

Disclosure statement
No potential conflict of interest was reported by the author(s).

ORCID
Jean-Claude Malela-Majika http://orcid.org/0000-0001-7236-7678
Sandile Charles Shongwe http://orcid.org/0000-0003-2243-8196
Muhammad Aslam http://orcid.org/0000-0003-0644-1950

References
[1] N. Abbas, Homogeneously weighted moving average control chart with an application in substrate
manufacturing process. Comput. Ind. Eng. 120 (2018), pp. 460–470.
[2] N. Abbas, M. Riaz, S. Ahmad, M. Abid, and B. Zaman, On the efficient monitor-
ing of multivariate processes with unknown parameters. Mathematics 8 (2020), pp. 823.
doi:10.3390/math8050823.
[3] M. Abid, S. Mei, H.Z. Nazir, M. Riaz, and S. Hussain, A mixed HWMA-CUSUM mean chart
with an application to manufacturing process. Qual. Reliab. Eng. Int. 37 (2020), pp. 618–631.
[4] M. Abid, A. Shabbir, H.Z. Nazir, R.A.K. Sherwani, and M. Riaz, A double homogeneously
weighted moving average control chart for monitoring of the process mean. Qual. Reliab. Eng.
Int. 36 (2020), pp. 1513–1527.
[5] N.A. Adegoke, S.A. Abbasi, A.N.H. Smith, M.J. Anderson, and M.D.M. Pawley, A multivariate
homogeneously weighted moving average control chart. IEEE. Access. 7 (2019), pp. 9586–9597.
[6] N.A. Adegoke, A.N.H. Smith, M.J. Anderson, R.A. Sanusi, and M.D.M. Pawley, Efficient
homogeneously weighted moving average chart for monitoring process mean using an auxiliary
variable. IEEE. Access. 7 (2019), pp. 94021–94032.
[7] O.A. Adeoti and S.O. Koleoso, A hybrid homogeneously weighted moving average control chart
for process monitoring. Qual. Reliab. Eng. Int. 36 (2020), pp. 2170–2186.
[8] V. Alevizakos and C. Koukouvinos, A double exponentially weighted moving average control
chart for monitoring COM-Poisson attributes. Qual. Reliab. Eng. Int. 35 (2019), pp. 2130–2151.
[9] V. Alevizakos and C. Koukouvinos, A progressive mean control chart for COM-Poisson distribu-
tion. Commun. Stat. Simul. Comput. (2020). doi:10.1080/03610918.2019.1659361.
[10] M. Aslam, L. Ahmad, C.-H. Jun, and O.H. Arif, A control chart for COM–Poisson distribution
using multiple dependent state sampling. Qual. Reliab. Eng. Int. 32 (2016), pp. 2803–2812.
[11] M. Aslam and A.H. Al-Marshadi, Design of a control chart based on the COM-
Poisson distribution for the uncertainty environment. Complexity (2019), pp. 8178067.
doi:10.1155/2019/8178067.
[12] M. Aslam, M. Azam, and C.-H. Jun, A control chart for COM-Poisson distribution using
resampling and exponentially weighted moving average. Qual. Reliab. Eng. Int. 32 (2016), pp.
727–735.
[13] M. Aslam, N. Khan, and C.-H. Jun, A hybrid exponentially weighted moving average chart for
COM-Poisson distribution. Trans. Inst. Meas. Control 40 (2018), pp. 456–461.
[14] M. Aslam, A. Saghir, and L. Ahmad, Introduction to Statistical Process Control, 1st ed., John
Wiley & Sons, New York, 2020.
[15] M. Aslam, A. Saghir, L. Ahmad, C.-H. Jun, and J. Hussain, A control chart for COM-Poisson
distribution using a modified EWMA statistic. J. Stat. Comput. Simul. 87 (2017), pp. 3491–3502.
30 O. A. ADEOTI ET AL.

[16] C.M. Borror, C.W. Champ, and S.E. Rigdon, Poisson EWMA control charts. J. Qual. Technol. 30
(1998), pp. 352–361.
[17] J.-H. Chen, A double generally weighted moving average chart for monitoring the COM-Poisson
processes. Symmetry 12 (2020), pp. 1014. doi:10.3390/sym12061014.
[18] R.W. Conway and W.L. Maxwell, A queuing model with state dependent service rates. J. Ind. Eng.
12 (1962), pp. 132–136.
[19] A. Dawod, N.A. Adegoke, and S.A. Abbasi, Efficient linear profile schemes for monitoring bivari-
ate correlated processes with applications in the pharmaceutical industry. Chemom. Lab. Syst.
206 (2020), pp. 104137. doi:10.1016/j.chemolab.2020.104137.
[20] F.F. Gan, Monitoring Poisson observations using modified exponentially weighted moving average
control charts. Commun. Stat. Simul. Comput. 19 (1990), pp. 103–124.
[21] K. Mabude, J.-C. Malela-Majika, P. Castagliola, and S.C. Shongwe, Generally weighted moving
average monitoring schemes – overview and perspective. Qual. Reliab. Eng. Int. 37 (2021), pp.
409–432.
[22] D.C. Montgomery, Statistical Quality Control: A Modern Introduction, 7th ed., John Wiley &
Sons, Singapore, 2013.
[23] T. Nawaz and D. Han, Monitoring the process location by using new ranked set sampling based
memory control charts. Qual. Technol. Quant. Manag. 17 (2020), pp. 255–284.
[24] E. Page, Continuous inspection schemes. Biometrika 41 (1954), pp. 100–111.
[25] M. Raza, T. Nawaz, and D. Han, On designing distribution-free homogeneously weighted moving
average control charts. J. Test. Eval. 48 (2020), pp. 3154–3171.
[26] S.W. Roberts, Control chart tests based on geometric moving averages. Technometrics. 1 (1959),
pp. 239–250.
[27] A. Saghir and Z. Lin, Control chart for monitoring multivariate COM-Poisson attributes. J. Appl.
Stat. 41 (2014), pp. 200–214.
[28] A. Saghir and Z. Lin, Cumulative sum charts for monitoring the COM-Poisson processes.
Comput. Ind. Eng. 68 (2014), pp. 65–77.
[29] A. Saghir and Z. Lin, A flexible and generalized exponentially weighted moving average control
chart for count data. Qual. Reliab. Eng. Int. 30 (2014), pp. 1427–1443.
[30] A. Saghir and Z. Lin, Control charts for dispersed count data: An overview. Qual. Reliab. Eng.
Int. 31 (2015), pp. 725–739.
[31] A. Saghir, Z. Lin, S.A. Abbasi, and S. Ahmad, The use of probability control limits of the COM-
Poisson charts and their applications. Qual. Reliab. Eng. Int. 29 (2013), pp. 759–770.
[32] K.F. Sellers, A generalized statistical control chart for over-or under-dispersed data. Qual. Reliab.
Eng. Int. 28 (2012), pp. 59–65.
[33] K.F. Sellers, S. Borle, and G. Shmueli, The COM-Poisson model for count data: A survey of
methods and applications. Appl. Stoch. Models. Bus. Ind. 28 (2012), pp. 104–116.
[34] S.-H. Sheu and T.-C. Lin, The generally weighted moving average control chart for detecting small
shifts in the process mean. Qual. Eng. 16 (2003), pp. 209–231.
[35] S.C. Shongwe, J.-C. Malela-Majika, and P. Castagliola, A combined mixed-s-skip sampling strat-
egy to reduce the effect of autocorrelation on the X̄scheme with and without measurement errors.
J. Appl. Stat. 48 (2021), pp. 1243–1268.
[36] W.H. Woodall, Control charts based on attribute data: Bibliography and review. J. Qual. Technol.
29 (1997), pp. 172–183.