You are on page 1of 11

Simulation

Methodology
Simulation: Transactions of the Society for
Modeling and Simulation International
2017, Vol. 93(2) 91–101
Modeling and simulation of customer Ó The Author(s) 2016
DOI: 10.1177/0037549716678477
dissatisfaction in waiting lines and its journals.sagepub.com/home/sim

effects

Jorge A Alvarado-Valencia1, Gabriela C Tueti Silva1 and Jairo R Montoya-Torres2

Abstract
A combination of discrete-event and agent-based simulation analysis using a field-tested psychological model for evaluat-
ing the effects of customer dissatisfaction in waiting lines beyond balking and reneging was developed. The proposed
model assessed the effects that different psychological parameter values and business decisions in waiting lines have on
the evaluation of waiting and, therefore, in customer satisfaction and mid-term profit. This model hence becomes a
decision-support tool for businesses wanting to model their costs of customer dissatisfaction due to waiting lines in
repetitive and competitive environments.

Keywords
Customer behavior, hybrid simulation, decision-making modeling, customer dissatisfaction, waiting lines, queue analysis

1. Introduction usually assign probability distributions to the impatience


of users or define fixed rates of balking without consider-
In situations where waiting is inevitable, understanding ing the psychological or psychophysical mechanisms of
the psychological factors that influence the evaluation of wait evaluation as developed in psychology.
waiting is of great importance in order to avoid customer There is a long tradition of queuing system analysis
dissatisfaction and the costs associated with it. Waiting through discrete-event simulation techniques. As a conse-
evaluation, even more than the actual time waited, is a quence, code/software to simulate virtually any queuing
key factor associated with customer dissatisfaction.1–4 system configuration is easily available and attachable to
Customer dissatisfaction leads to loss of loyalty and nega- other models, and performance measures are calculated
tive references, which impacts income and profits.5–7 quickly at the service level. On the other hand, agent-
However, it is challenging for businesses to quantify the based models are more appropriate to model behaviors at
aforementioned costs of dissatisfaction and, therefore, it is the customer level, which is a lower level of the system.
difficult to determine how these costs affect their results From these lower levels, the emergence of properties to
and their operations. the higher level and the dynamics over the time of the
Developing service evaluation models based on Quality whole queuing system can be discovered or illustrated.10
of Experience (QoE), rather than on Quality of Service Developing a hybrid (agent-based and discrete-event)
(QoS), has recently attracted the highest interest of acade- model allows one to simulate specific customer behavior
mia and industry.8 The literature has proposed frameworks at the micro level with agents, while the analysis at the
to evaluate customer behavior given their QoE through
simulation (e.g., Montreuil et al.9). However, queuing sys-
tems analysis is not frequently focused on modeling wait- 1
Industrial Engineering Department, Pontificia Universidad Javeriana, Cra.
ing time evaluation or customer dissatisfaction. Models in 7 No. 40–62, Bogotá, D.C., Colombia
the literature typically focus on ‘‘impatient’’ customers 2
School of Management, Universidad de los Andes, Calle 21 # 1-20,
who balk and/or renege; with scarce exceptions, these Bogotá, D.C., Colombia
models do not include the evaluation of waiting times or
Corresponding author:
other possible behaviors of customer dissatisfaction when Jorge A Alvarado-Valencia, Industrial Engineering Department, Pontificia
waiting evaluation is negative, beyond balking or reneging. Universidad Javeriana, Cra. 7 No. 40–62, Bogotá, D.C., Colombia.
Furthermore, simulation models of impatient customers Email: jorge.alvarado@javeriana.edu.co
92 Simulation: Transactions of the Society for Modeling and Simulation International 93(2)

service level is facilitated, using the power and standardi- affected the acceptable wait time (maximum time that a
zation of discrete-event simulation. customer is willing to wait). Evaluation of waiting time is
For the aforementioned reasons, this manuscript shows determined by the difference between the perceived wait-
the development of a hybrid (agent-based and discrete- ing time and the acceptable waiting time. However,
event) simulation model for waiting lines with impatient Antonides et al.15 proposed that the perceived time can be
customers contributing three important results. Firstly, a better modeled using a psychophysical function, where the
psychological waiting evaluation mechanism was inte- evaluation of waiting time was a function of the difference
grated into the queue model based on psychological between the logarithm of acceptable waiting time (expec-
research findings. Secondly, we developed a response to tation) and the logarithm of the perceived waiting time.
the evaluation of waiting time in terms of several customer This function also included an asymmetry that depends on
behaviors beyond balking and reneging. Thirdly, we inte- whether the final evaluation was positive or negative,
grated a cost function that includes hidden costs of cus- transforming it in a non-linear non-continuous function. In
tomer dissatisfaction due to waiting evaluation, based on turn, perceived time was affected by the actual waiting
the modeled behavior responses. We believe that these time, time fillers, and environmental factors. We base our
contributions can be helpful for any business to quantify work on the model of Antonides et al.15
the losses generated by customer dissatisfaction in the Different research studies have been conducted to test
long term and provide businesses with a tool to support the effects of time fillers on evaluating waiting times.
their decision-making. Time filler effects seem to be context-dependent, but there
This paper is organized as follows. Firstly, we present a is evidence that time fillers such as music,15 TV screens,16
literature review of models that have been proposed for or clock presence17 can exert a psychological influence on
waiting evaluation and the possible responses of dissatis- customers and customer satisfaction and, therefore, the
fied customers in Section 2. Secondly, we explain the con- proposed model included fillers to be evaluated.
ceptual framework of the simulation and the modeling of
the customer dissatisfaction cost in Section 3. Thirdly, we
present a case study and analyze the results in an experi- 2.2. Queuing systems with psychologically modeled
mental design in Section 4. Finally, the manuscript pre- customer behavior
sents the primary conclusions of the study and it suggests Queuing systems analysis with impatient customers has
lines for future research in Section 5. focused mainly on the phenomena of balking and rene-
ging.18–20 When modeling balking and reneging, these
behaviors are usually seen as birth–death processes.21 As a
2. Literature review consequence, Poisson or exponential random variables are
2.1. Waiting evaluation in psychology used.22,23 These birth–death processes might be affected
The relationship between waiting time and perceived time by queue length,24,25 waiting time for each customer,26 or
is a key psychological factor behind waiting evaluation. In both,27 leading to a model of waiting evaluation. More
1985, Maister11 suggested that customer satisfaction is the sophisticated approaches included an evaluation between
difference between the perceived waiting time and the the reward of the service and the cost of waiting time,
expectation of customers. In the theoretical work by Nie,4 which is a positive, increasing, and continuous function.28
and based on the theory of assimilation and contrast, it is Other simulation approaches made a comparison between
suggested that the perceived waiting time is related to the expected/tolerable and experienced time as a basic func-
waiting time estimated by the customer (expectation) and tion to define an abandonment threshold29 or to model
that the general evaluation of the service will be the differ- behaviors such as word of mouth or consumption fre-
ence between those two times. Subsequently, other studies quency change.30,31 Some of these assumptions have been
found that comparing perception and expectation is more found to be partially valid in simulated tasks,17 but have
important than the length of the line when determining not been tested in real environments. Furthermore, these
customer satisfaction.12–14 approaches did not include the possible asymmetry and
None of the previous works suggested a specific mathe- non-linearity of the relationship between perceived and
matical model for this relationship. Two mathematical actual waiting time in waiting evaluation proposed by psy-
models were found in the literature relating waited and chological approaches tested in the field.15
perceived time. Pruyn and Smidts7 developed a model to
evaluate waiting, in which the actual waiting time and
2.3. Customer behavior after satisfaction/
environmental conditions in which waiting time occurs
were the primary factors that affect customer satisfaction. dissatisfaction
The actual waiting time and the distractors directly Hirschman’s model of responses to dissatisfaction32 pro-
affected the perceived time, and the perceived attraction posed three possible types of responses from dissatisfied
Alvarado-Valencia et al. 93

customers: ending the relationship with the company or satisfaction over time in queues, with the notable excep-
reducing purchasing frequency; complaining; or maintain- tion of Ackere and Larsen.30 Adding to this work, we
ing their loyalty to the company in spite of their dissatis- included a better tested psychological model of waiting
faction. Another form of a dissatisfied response is negative evaluation, added a valuation of financial performance,
word of mouth (NWOM), which involves communicating and developed our simulation in a study case in a particu-
dissatisfaction to others, without including the service pro- lar context based on real data.
vider.33 This communication could cause the loss of cred-
ibility of an organization, affecting its reputation within
the market and significantly reducing its profits.34 In the 3. Model description
case of satisfied service, there can also be positive word of The model description follows the ODD (Overview,
mouth (PWOM).35 Design concepts, Details) protocol.43,44 The discrete-event
However, all of these effects require a business model model is integrated in the model description as input data
where customer loyalty is important, which is expressed (see Section 3.6). The following sections describe the dif-
through their repeated consumption of the service, and the ferent components of the ODD protocol.
possibility of finding an alternative for the customer to pur-
chase, that is, competitors exist that can absorb the reduced
demand due to the dissatisfaction-associated behaviors of 3.1. Purpose
the customer. For this reason, it is necessary to stress that The model is designed to evaluate the effects on company
the developed model applies particularly well in repetitive results of customer dissatisfaction in waiting lines based
and competitive environments, for example, in a restau- on several customer behaviors, which in turn are based on
rant. In this way, dissatisfaction has a cost associated with psychological models of waiting time evaluation.
negative or positive decisions that the consumer may make
in the future.
This repetitive and competitive environment has a 3.2. Entities, state variables, and scales
remarkable resemblance to the El Farol Bar problem,36 Customers (acting as agents) are the entities in the model.
where a fixed group of customers independently and repe- State variables of customers are frequency of purchase fi,
titively decide, based on a set of strategies grounded on expected waiting time ji , actual waiting time ti , and cumu-
past information, whether to go to a bar. Bar service will lative number of dissatisfactions CDi. fi is measured in
be satisfactory only if the attendance does not surpass purchases per time unit (purchases per day); expected and
some threshold. In the presented model, a group of cus- actual waiting times are in time units (minutes), and CDi
tomers decide, based on prior information, whether to pur- is a counting variable. Notice that fi and ji can be different
chase repetitively a service. However, (a) the decision for each customer, leading to a diverse population of
process is not completely independent, because there is customers.
some degree of communication between customers; (b) One time step represents one day, and each simulation
the satisfaction threshold is not the same for each cus- replication was run for 20 days. One time step implies an
tomer, due to diverse expectations; (c) customer satisfac- interaction with the queuing system.
tion is time-dependent (i.e., in the same iteration, some
customers might be satisfied and some might be dissatis-
fied); and (d) the attendance strategy is fixed. The El Farol 3.3. Process overview and scheduling
Bar problem has proved to be a situation where simple The general process scheduling of the system is depicted
individual rules lead to unexpected properties in the col- in Figure 1, which is also computationally represented in
lectivity and a rich structure of emergent and dynamic col- the following pseudocode. Initial values of parameters are
lective behavior.37 Simulations for this kind of problems indicated in Section 3.5.
are enlightening, because theoretical results are difficult to
develop for them.38 Therefore, we expect to find also a
similar rich collective behavior. In particular, we expect 3.4. Design concepts
possible stochastic fluctuations around an equilibrium 3.4.1 Basic principles. The described model allows for eval-
point, where fluctuation behavior depends on agent char- uating the consequences of two underlying basic princi-
acteristics such as the possibility of communication39 or ples. Firstly, customer waiting evaluation on a queuing
the biases of its predictions.40 system is a non-trivial psychological process that depends
Several simulation models have addressed the issue of on the comparison of expectations and perceptions. Real
customer behavior as agents and their impact in operation waiting time is just a component of this process. Secondly,
performance41 or marketing research.42 However, to the there are customer responses to dis/satisfaction in queuing
best of our knowledge, none of them have addressed this systems beyond balking and reneging, which can impact
specific issue of waiting evaluation and customer business performance through hidden costs. These
94 Simulation: Transactions of the Society for Modeling and Simulation International 93(2)

Figure 1. Simulation conceptual model. PWOM: positive word of mouth; NWOM: negative word of mouth.

Algorithm 1. Pseudocode for the general process.

1. FOR time step = 1 to Maxtime DO


2. All customers go through the queuing system with frequency fi (*see Section 3.6*)
3. SET N := Actual number of customers
4. FOR customer i=1 to N DO
5. WHEN customer i leaves the service system DO:
6. Compute perceived time of customer i (*see Section 3.7*)
7. Compute wait evaluation of customer i (*see Section 3.7*)
8. IF Wait evaluation ui < 0 (*customer dissatisfied*) THEN
9. Generate three Bernoulli random variables Ak with parameter pk (*each one representing the probability of taking
a dissatisfaction action*)
10. IF A1 =1 THEN
11. fi = fi* reduction (*the customer will reduce their future purchasing intent*)
12. IF A2 =1 THEN
13. Actual number of customers = Actual number of customers − 1 (*A customer leave the simulation from the next time
step on due to NWOM*)
14. IF A3 =1 THEN
15. Complaints = Complaints + 1 (*The customer complains, generating a paperwork cost for the service system*)
16. CDi = CDi + 1 (*Cumulative number of dissatisfaction increases*)
17. IF CDi ≥ M (* number of dissatisfactions threshold*) THEN
18. Actual number of customers = Actual number of customers − 1 (*Customer leaves the simulation from the
next time step on*)
19. ELSE
20. IF wait evaluation > = 0 (*customer satisfied*) THEN
21. Generate two Bernoulli random variables Ak with parameter pk (*each one representing the probability of taking a satisfaction
action*)
22. IF A1 = 1 THEN
23. IF fi < Maximum frequency THEN
24. fi = fi* increase (*customer increases its purchasing intent*)
25. END-IF
26. IF A2 = 1 THEN
27. Actual number of customers= Actual number of customers + 1 (*A new customer is created due to PWOM*)
28. END-IF
29. END-FOR
30. END-FOR
Alvarado-Valencia et al. 95

responses can include actions such as word of mouth, prediction of dis/satisfaction in the next queuing system
complaints, and reduction of purchase frequency. experience.

3.4.2 Emergence. System performance measures are 3.4.7 Sensing. Customers sense real waiting time through
expected to vary in complexity due to individual customer their perception filter (see Section 3.7). They can also
behaviors. In particular, satisfied customers bringing new sense PWOM/NWOM from another agent.
customers and dissatisfied customers making other cus-
tomers leave the system are expected to change service 3.4.8 Interaction. The main interaction between agents is
demand over time. In addition, customers differ in their through PWOM/NWOM on a one-to-one basis. After each
waiting time expectations, leading to different customer service experience, an agent with positive wait evaluation
actions even with the same objective waiting time. As a can bring a new agent into the system, whereas an agent
consequence, waiting time and queue length are expected with negative wait evaluation can influence another differ-
to deviate from a convergence to a steady state and will ent agent to leave the system. Therefore, agents can com-
probably show fluctuations around an equilibrium point, municate their service experience to other agents and
unpredicted by classical queuing theory but expected in influence their decisions about entering or leaving the
emergent behaviors, with an impact on company profits. system.

3.4.3 Adaptation. Customers adapt themselves to the queu- 3.4.9 Stochasticity. Customer behaviors after waiting eva-
ing system in response to their waiting evaluation. If wait- luation are taken (or not) depending on random uniform
ing evaluation is positive, they might increase their variables. Perception and waiting evaluation models have
purchase frequency or give PWOM to a potential cus- random noise (see Section 3.7)
tomer; if waiting evaluation is negative they might reduce
their purchase frequency, complain, give NWOM to actual
customers, or even stop purchasing the queuing system 3.4.10 Observation. All output data is collected and freely
service. Notice that adaptations are independent of the used.
magnitude of the positive/negative waiting evaluation.
Several studies have shown that complaining behavior 3.5. Initialization
seems to be loosely related45,46 or unrelated47 to the inten-
sity of dissatisfaction. Initial values of the model system are presented in Table
1. Table 1 also shows parameter values of submodels and
variable parameter values for the experimental design (see
3.4.4 Objectives. The goal of adaptive traits is generating Section 4.1).
customer satisfaction. Such satisfaction is generated by a
positive waiting time evaluation. Notice that word of
mouth seeks to help other individuals to predict their wait- 3.6. Input data
ing time evaluation, encouraging or discouraging them to The service system is a discrete-event-based model of a
go through the queuing system; complaints are intended to waiting line. Any service model following the Kendall
give satisfaction feedback to the queuing system (although notation48 can be coupled with the agent-based model. It
in the current model this is an unsuccessful action) and can be also extended (and it was for the current model)
purchase reduction – or even system abandonment – looks with a possible vector of fillers Xi whose only function is
for a reduction of dissatisfaction experiences, under the to modify time perception according to the submodel pre-
assumption that the service can be purchased somewhere sented in Section 3.7.1. For the case study, input data cor-
else (i.e., there is a competitive environment). respond to an M/G/1 queue (i.e., a queue with Markovian
arrivals, service times with a general distribution and a
single server) where a possible time filler can be presented
3.4.5 Learning. Each customer has a memory of the num-
to the customer. Arrival and waiting time distributions
ber of previous dissatisfactions, which helps one to decide
were extracted from Aparicio and Ortiz.49
when to stop purchasing the queuing system service. Also,
The service system receives the customers from the
frequency of purchase changes with experienced waiting
agent-based model in each time step. Depending on the
evaluation.
service system performance, the service system assigns a
current waiting time ti for each customer in each time step.
3.4.6 Prediction. There are several hidden predictions of Performance of the service system is measured through
waiting evaluation in the model. In fact, when customers classical waiting line measures, such as the average wait-
decide whether to purchase or not, there is an implicit ing time and the average queue length, along with the
96 Simulation: Transactions of the Society for Modeling and Simulation International 93(2)

Table 1. Model parameters, variables, and factors.

Entity (parameter or variable) Role (Initial)Value/probability function

Maxtime Global parameter 20 time steps


M (number of dissatisfaction threshold) Global parameter 2 dissatisfactions
Maximum frequency Global parameter 2 per time period
Reduction Global parameter ½
Increase Global parameter 2
Complaints Global parameter Starts in 0
Actual number of customers (N) Global variable Starts in 58 initial customers
fi Entity variable Starts in 1 per time period
CDi Entity variable Starts in 0
Expectation times (ξi ) Entity Random variable Normal (3.68; 1) – in minutes
Ak Global Random variable Bernoulli random variables with pk = 0.5 for all cases.
Arrival and service times Input Random variables Based on Aparicio and Ortiz49: exponential(0.392) &
lognormal(0.36;0.154) – in minutes
CQ; CBB; CDC; CSD Submodel parameter Based on the survey of Cabezas Chaparro52
a1 Submodel parameter Ln(1.2). 20% time overestimation. Based on Hornik54 and
Whiting and Donthu55
a2 Submodel parameter Ln(2). A customer is satisfied if perceived time is one
standard deviation below average waiting time found by
Aparicio and Ortiz49
b1 Submodel parameter 1
b2 Submodel parameter –1
b3 Submodel factor Low = 0.3; Medium = 0.75; High = 1.2
b4 and b5 Submodel factors Low = 1; High = 2
gi * X1 (time filler effect) Submodel factor No filler = 0; low effect = − 0.12 (1.13 min. reduction);
high effect = − 0.44 (1.55 min. reduction) Based on
Aparicio and Ortiz49
εji Submodel random variables Normal (0;0.2)

percentage of satisfied customers and the profit calculated ui = a2 + b2 lnðui + 1Þ + b3 t i + b4 t i ðlnðji + 1Þ
in the submodel explained in Section 3.7.2. ln (ui + 1)Þ + b5 ð1  t i Þðlnðji + 1Þ  lnðui + 1ÞÞ + e2i
ð2Þ
3.7. Submodels where ui refers to the evaluation of waiting time by cus-
3.7.1 Wait evaluation submodel. To calculate the perceived tomer i, ji represents the expectation of the customer
time and the evaluation of waiting time, we used the model regarding the time they want to wait, ti is a Boolean vari-
proposed in the study by Antonides et al.15 However, these able that is 1 if the expectation is higher than the percep-
equations are undefined when waiting, perceived, or tion and 0 if the opposite is true, and ui represents the
expected time are 0. Furthermore, if any of the aforemen- perceived waiting time. It can be seen that the structure of
tioned times are less than one, the original equations of the equation allows for asymmetries to affect the evalua-
Antonides et al.15 can generate negative perceived times or tion of the size of positive or negative differences between
inappropriate values in the difference between perceived perception and expectation (through parameters b4 and
and expected times, due to logarithm properties. As a con- b5). Also, b3 allows giving an additional effect for positive
sequence, we decide to add one time unit to make the differences between perception and expectation. Finally,
equations work properly in the simulation. notice that a customer might have a negative evaluation
Perceived time: even when expectation is below perception, due to time
perception effects modeled by b2. All these elements gave
lnðui Þ = a1 + b1 lnðti + 1Þ + gi Xi + e1i ð1Þ a higher flexibility for modeling wait evaluation over the
classical approaches of real time thresholds or the simple
where ui is the perceived waiting time; ti is the actual difference between perceptions and expectations.
waiting time, and Xi is a vector of possible time fillers that
influence the perceived waiting time. b1 allows for distor- 3.7.2 Profits submodel. Profit calculation is made in an
tions on time perception due to psychological factors. aggregate manner for each simulation replication, as
Evaluation of waiting time: follows:
Alvarado-Valencia et al. 97

Table 2. Effect sizes of significant factors. in accordance with several empirical studies showing per-
centages between 10% and 20% of inactivity.50,51
Factor Contrasted effects Cohen’s d The simulation was run during 20 consecutive time
b4 High vs. Low 0.23 periods. At the end of the total time period, the total profit
b3 Medium vs. Low 0.21 is calculated using Equation (3), based on the costs found
High vs. Medium 0.12 by the customer survey in Cabezas Chaparro.52 An experi-
High vs. Low 0.30 mental design of 22* 32 was proposed to evaluate the
Time filler Low vs. Control 0.17 effects of b3, b4, b5, and time fillers on profit. Table 1
High vs. Low 0.33
High vs. Control 0.45 show the parameters, random variables, and factors used
in the simulation case study, based on Equations (1)-(3)
and the simulation conceptual model, presented in Section
3. Analyses of variance (ANOVAs) were conducted for
Profit = Income  Costs = analysis, along with descriptive statistics.
ðIV + IBP Þ  (CQ + CBB + CSD + CDC Þ ð3Þ Our hypotheses were as follows.

where IV is the total income during the whole simulation H1: A larger magnitude additional effect for positive
replication due to purchases of current customers (i.e., cus- differences between perception and expectation (higher
tomers generated at simulation time = 0), IBP is the total b3) will lead to higher profits.
income during the whole simulation replication due to pur- H2: A larger magnitude effect of positive differences
chases of new customers (i.e., customers generated at between perception and expectation (higher b4) will
simulation time 6¼ 0), CQ is the total cost at the end of the lead to higher profits.
simulation replication of all complaints (complaints vari- H3: A larger magnitude effect of negative differences
able multiplied by the unitary cost of a complaint), CBB is between perception and expectation (higher b5) will
the aggregated loss over all customers on expected future lead to lower profits.
purchases (i.e., until the end of the simulation) at the initial H4: A time filler of larger magnitude effect on reducing
frequency of purchase (shown in Table 1) of any customer time perception will lead to higher profits.
that does not return to the system, CDC is the aggregated
loss over all customers on expected future purchases at the
initial frequency of purchase (shown in Table 1) of any 4.2 Experiment results
period of purchase frequency reduction of each customer, Larger magnitude effects of positive difference between
and CSD is the aggregated cost over the whole simulation perception and expectation (b4) increased profit F(1) = 64,
replication of idle service of personnel, calculated from p \ .01, supporting H2, whereas larger magnitude effects
the discrete-event simulation of the negative difference (b5) were not significant F(1) =
0.36, p . .10, not supporting H3. The additional effect of
4. Experimental design positive difference (b3) was also significant on their incre-
mental effect on profit F(2) = 41.8, p \ .01, supporting
4.1 Case study H1, along with the time filler effect F(2) = 100.67,
The simulation to be developed represents the service of a p \ .01, supporting H4. The effect sizes of profit increases
business restaurant. The description of the system under are small (below half a standard deviation) and are shown
study is based on the data collected by Aparicio and in Table 2.
Ortiz,49 and corresponds to an M/G/1 queue in a restaurant Normality assumption held for all subgroups; homoge-
where time fillers are presented to the customer. The neity of variances did not. However, ANOVA results are
model was validated based on the data from Aparicio and robust to this assumption, providing that sample sizes for
Ortiz,49 where waiting time expectations and perceptions subgroups are the same.53
were collected in a restaurant environment, and from Higher b3 and b4 implied a tendency of being satisfied
where arrival and service time distributions were extracted. even in longer waits (as long as wait does not surpass
As shown in Table 1, we decided to define the same expectation), encouraging the arrival of more customers or
probability (pk = 0.5) for any satisfaction/dissatisfaction increasing purchase frequency; on the other hand, larger
action in the case study. We were unable to find studies time filler effects made waiting time more tolerable, lead-
showing a general preference towards a given specific ing to higher satisfaction and, therefore, increasing queue
action, let alone in restaurant environments. A single study length and waiting time. As a result, the model captures
in the services sector shows pretty similar percentages of the possibility that psychological phenomena might allow
response for our modeled dissatisfaction actions.50 By poorer QoS (until a certain level) with increased QoE.
using pk = 0.5, dissatisfied customers who do not take any Given the nature of the psychological model, profits
action are 12.5% of the total customers. This percentage is increased when waiting times and queue length increased;
98 Simulation: Transactions of the Society for Modeling and Simulation International 93(2)

Table 3. Performance measures of the system.

Factor Level Waiting time (Wq)a Queue length (Lq) Satisfied Profit (G) US$

M SD M SD % M SD
b4 Low 3.54 1.10 5.64 3.30 .336 226.5 81.0
High 4.34 1.17 7.51 4.10 .340 248.7 52.9
b3 Low 3.29 1.01 5.20 2.95 .332 219.8 89.7
Medium 4.10 1.17 6.94 3.94 .340 241.8 55.4
High 4.43 1.14 7.59 4.10 .342 251.3 52.6
Time filler Control (no filler) 3.00 .73 4.78 2.45 .327 214.7 52.6
Low 3.45 .80 5.52 2.87 .336 227.1 51.7
High 4.90 1.07 8.53 4.36 .345 259.6 84.2
a
Waiting time units are minutes.
M: mean; SD: standard deviation.

however, the percentage of satisfied customers did not behavior with this emergent oscillating property, even
really change among treatments (Table 3). when key factors of the original problem are modified.
b3 and b4 can help model the important effect associ- Despite reaching a steady state, the system oscillates
ated with the type of customer of the company and his/her between sudden waves of interested customers, followed
psychological characteristics, that is, the degree of satis- by customer recessions due to long queues. This result is a
faction received by different types of customers of the reliable confirmation of the validity of the model when
companies when wait expectations surpass perception. compared to reality, because if waiting time evaluation
Varying b3 for each customer, it is possible to model the was the only deciding factor to remain in line, one would
personal or social characteristics of the company’s cus- expect that after the first satisfied customers, the line
tomers regarding the satisfactory effects of simply surpass- (queue) would begin to lengthen due to the satisfaction of
ing expectations. On the other hand, varying b4 for each the customers. This rise would increase the length of the
customer, it is possible to model the satisfactory effects of line, which in turn, would lead to customer dissatisfaction
the degree of such expectation surplus. and, subsequently, a lower demand. This lower demand
The effect of b5 was not significant, implying that dif- would in turn reduce waiting time, leading to a new cycle
ferent effect sizes when expectations are not fulfilled were of satisfaction–dissatisfaction.
not very important on changing profits. A possible expla- These cycles increase in frequency and amplitude with
nation is that expectations not fulfilled are usually associ- increases in b3 and b4. Amplitude increments show a
ated with longer waiting times, and longer waiting times higher tolerance to longer perceived waits, whereas fre-
had an additional negative effect in evaluation through b2, quency increments show sudden increases in the number
that is, passing of time already has a negative effect on of customers or their frequency of purchase due to cus-
evaluation regardless of expectations fulfillment. As a con- tomer satisfaction, followed by drops in consumption due
sequence, the negative evaluation added by differences in to these sudden longer lines. As a consequence, customer
b5 was unimportant (for the chosen levels). traits related to their valuation of shorter-than-expected
From the time filler results, one can infer that no matter waits can have an effect on system predictability, making
the nature of the filler, as long as it has a strong effect on operation management and planning more difficult.
perception, it will help maximize profits. As a result, cer- All the results show that the simulation in the case study
tain time fillers can be useless in terms of satisfying the exhibits the expected behaviors of the model and, therefore,
objectives of the company, and it is necessary to evaluate the simulation model is useful in evaluating the psychologi-
the effects of time fillers before applying them in a waiting cal phenomenon of satisfaction. Moreover, the model
line. becomes useful in discussing at least three aspects in the
Figure 2 shows a cyclical behavior in perceived time design of a waiting line: the cyclic nature expected of cus-
and waiting evaluation over the time with ascending and tomer satisfaction with waiting time when the services are
descending cycles. Given that perceived time is closely no different from the competition; how time fillers in the
related to actual waiting time, performance measures such waiting line affect profits; and the consequences of psycho-
as queue length and waiting time also show this cyclical logical and cultural aspects of customers that may affect the
behavior. These system performance behaviors resemble performance of the waiting line, which is represented by the
the behavior of attendance in the El Farol Bar problem, satisfaction bonus and the possible satisfaction asymmetries
suggesting that there might be a larger kind of system in the differences between perception and expectation.
Alvarado-Valencia et al. 99

Figure 2. Behavior of wait evaluation, perceptions, and expectations in time.

5. Conclusions and perspectives


At least three limitations of the work should be
The present research work proposes a simulation model pointed out. Firstly, customers might differ in their
applicable to competitive and repetitive business environ- response due to diverse personal and societal factors not
ments where the queue is an important part of customer included, not only by different expectations about their
satisfaction. The model incorporates one of the most waiting time. However, we believe that our work is a
recent and influential psychological models for the evalua- good first step towards personalized modeling of cus-
tion of waiting times using a customer response model that tomers. Secondly, we are evaluating only the waiting
includes several customer behaviors that affect business time, but the service experience might in fact completely
performance over time. This model could provide busi- modify this evaluation. In fact, we are bringing light
nesses with a valuable simulation tool that allows them to only to repetitive and competitive services where service
model the costs of customer dissatisfaction associated with experience is less important, but improved models
waiting lines. should include service experience as an important factor.
The simulation model evaluates the effects of different Thirdly, time filler effects are not necessarily linear and
values of psychological and time filler parameters on the additive, and their specific effects are still unclear in the
evaluation of waiting time in an actual environment. For literature. Despite that, the model might allow the inclu-
example, it is possible to simulate different types of inter- sion of more sophisticated relationships of time fillers
ventions and different types of customers. In the applied and perception, where they become clearer from psycho-
case, in fact, one can see the effect of different levels of logical research.
time fillers with more or less effectiveness; the results also Several lines of research arise based on the results
compare different psychological types of customers and obtained in the present article. The first research perspec-
their effect on the cost. Similarly, it was possible to tive consists of taking into account the value of the service
observe the cyclic behavior that would be exhibited by as a mediating factor in user satisfaction. In addition, the
waiting time evaluation if one could isolate it from service high and low levels of the psychological factors that affect
evaluation, and compare it with similar problems of emer- waiting time evaluation, which are represented in the
gent behavior present in the literature. equations by coefficients b4 and b5 , were selected by the
100 Simulation: Transactions of the Society for Modeling and Simulation International 93(2)

relationship between them when expectation surpasses 14. Chebat JC, Salem NH, Poirier JF, et al. Reaction to waiting
perception and vice versa. For future research, a focus on online by men and women. Psychol Rep 2010; 106: 851–869.
determining the real values of these psychological factors 15. Antonides G, Verhoef PC and van Aalst M. Consumer per-
and how they affect customer satisfaction and costs in ception and evaluation of waiting time: a field experiment.
actual cases is suggested, in order to increase external J Consumer Psychol 2002; 12: 193–202.
16. Borges A, Herter MM and Chebat JC. "It was not that long!":
validity of the model.
the effects of the in-store TV screen content and consumers
emotions on consumer waiting perception. J Retailing
Funding Consumer Serv 2015; 22: 96–106.
17. Pazgal A and Radas S. Comparison of customer balking and
This research received no specific grant from any funding agency
in the public, commercial, or not-for-profit sectors. reneging behavior to queueing theory predictions: an experi-
mental study. Comput Oper Res 2007; 35: 2537–2548.
18. Ou J and Rao BM. Benefits of providing amenities to
6. References impatient waiting customers. Comput Oper Res 2003; 30:
1. Akpmar MG, Ozkan B, Sayin C, et al. Consumer risk percep- 2211–2225.
tions towards food supply chain preferences: the case of the 19. Kawanishi Ki. QBD approximations of a call center queue-
supermarket. J Food Agric Environ 2010; 8: 256–260. ing model with general patience distribution. Comput Oper
2. Bjertnaes OA, Garratt A, Iversen H, et al. The association Res 2008; 35: 2463–2481.
between GP and patient ratings of quality of care at outpati- 20. Kangzhou W, Na L and Zhibin J (eds) Queueing system
ent clinics. Family Pract 2009; 26: 384–390. with impatient customers: a review. In: 2010 IEEE interna-
3. Hightower R, Brady M and Baker T. Investigating the role tional conference on service operations and logistics and
of the physical environment in hedonic service consumption: informatics (SOLI) (ed Y Gao), QingDao, China, 15–17 July
an exploratory study of sporting events. J Bus Res 2002; 55: 2010, pp.82–87. Shandong: IEEE.
697–707. 21. Economopoulos AA, Kouikoglou VS and Grigoroudis E.
4. Nie W. Waiting: integrating social and psychological per- The base stock/base backlog control policy for a make-to-
spectives in operations management. Omega 2000; 28: stock system with impatient customers. IEEE Trans Autom
611–629. Sci Eng 2011; 8: 243–249.
5. Lu Y, Musalem A, Olivares M, et al. Measuring the effect of 22. Dequan Y, Haiying L, Wenjie X, et al. (eds) Analysis of
queues on customer purchases. Manag Sci 2013; 59: 1743– two-server queueing system with impatient customers and
1763. a synchronous vacation policy. In: 2012 24th Chinese con-
6. Molina A, Martı́n VJ, Santos J, et al. Consumer service and trol and decision conference (CCDC) (ed GH Yang),
loyalty in Spanish grocery store retailing: an empirical study. Taiyuan, China, 23–25 May 2012, pp.4047–4051. IEEE.
Int J Consumer Stud 2009; 33: 477–485. 23. Gustafsson L. Poisson simulation as an extension of continu-
7. Pruyn A and Smidts A. Effects of waiting on the satisfaction ous system simulation for the modeling of queuing systems.
with the service: beyond objective time measures. Int J Res Simulation 2003; 79: 528–541.
Market 1998; 15: 321–334. 24. Wang Y, Guo J, Ceder A, et al. Waiting for public transport
8. Schatz R, Hoßfeld T, Janowski L, et al. From packets to peo- services: queueing analysis with balking and reneging beha-
ple: quality of experience as a new measurement challenge viors of impatient passengers. Transp Res B Methodolog
(Lecture Notes in Computer Science (including subseries 2014; 63: 53–76.
Lecture Notes in Artificial Intelligence and Lecture Notes in 25. Pen-Yuan L (ed.) Optimal pricing strategy for queuing sys-
Bioinformatics)). 2013, pp.219–263. Berlin: Springer. tems with capacity constraint problem. In: 2007 third inter-
9. Montreuil B, Labarthe O and Cloutier C. Modeling client national conference on intelligent information hiding and
profiles for order promising and delivery. Simulat Model multimedia signal processing (IIHMSP) (ed BY Liao),
Pract Theor 2013; 35: 1–25. Kaohsiung, Taiwan, 26–28 November 2007, pp. 561–564.
10. Axtell R. Why agents? On the varied motivations for agent IEEE.
computing in the social sciences. Washington, DC: The 26. Miao Y, Jun G, Jiafu T, et al. (eds) The method of staffing a
Brookings Institution, Center on Social and Economic call center with delay information considering the customers’
Dynamics, Working Paper 17. behavior. In: 2013 25th Chinese control and decision confer-
11. Maister DH. The psychology of waiting lines. In: Czepiel ence (CCDC) (ed GH Yang), Guiyang, China, 25–27 May
JA, Solomon MR and Surprenant CF (eds) The service 2013, pp.4723–4727. IEEE.
encounter: managing employee/customer interaction in busi- 27. Chiang YJ, Ouyang YC and Hsu CH. Performance and cost-
ness services. Lexington, MA: Lexington Books, 1985, effectiveness analyses for cloud services based on rejected
pp.113–123. and impatient users. IEEE Trans Serv Comput 2014; 9: 446–455.
12. Cameron M. The effects of music, wait-length evaluation, 28. Guo P and Zipkin P. The effects of the availability of
and mood on a low-cost wait experiences. J Bus Res 2003; waiting-time information on a balking queue. Eur J Oper
56: 421–430. Res 2009; 198: 199–209.
13. Katz KL, Larson BM and Larson RC. Prescription for the 29. Zohar E, Mandelbaum A and Shimkin N. Adaptive behavior
waiting-in-line blues: entertain, enlighten, and engage. Oper of impatient customers in tele-queues: theory and empirical
Manag 1991; 2: 160. support. Manag Sci 2002; 48: 566–583.
Alvarado-Valencia et al. 101

30. Ackere AV and Larsen ER. Long-term and short-term cus- 50. Day RL and Bodur M. Consumer response to dissatisfaction
tomer reaction: a two-stage queueing approach. Syst Dynam with services and intangibles. Adv Consumer Res 1978; 5:
Rev 2006; 22: 349–369. 263–272.
31. Jin Yan S, Ming X, Li X, et al. (eds) Customer-centric opti- 51. Sing J. A typology of consumer dissatisfaction response
mal resource reconfiguration for service outlets. In: IEEE/ styles. J Retail 1990; 66: 57–99.
INFORMS international conference on service operations, 52. Cabezas Chaparro MP. Modelamiento de los costos de
logistics and informatics (SOLI ’09) (ed D Lui), Chicago, espera por insatisfacción de clientes en una lı́nea de espera
USA, 22–24 July 2009, pp.754–759. IEEE. con una aplicación basada en restaurantes: Pontificia
32. Hollender N, Hofmann C, Deneke M, et al. Integrating cog- Universidad Javeriana, http://hdl.handle.net/10554/17609
nitive load theory and concepts of human-computer interac- (2010, accessed 24 July 2016).
tion. Comput Hum Behav 2010; 26: 1278–1288. 53. Harwell M. Summarizing Monte Carlo results in methodolo-
33. Chan SH. The roles of user motivation to perform a task and gical research: the single-factor, fixed-effects ANCOVA
case. J Educ Behav Stat 2003; 28: 45–70.
decision support system (DSS) effectiveness and efficiency
54. Hornik J. and Subjective vs. objective time measures: a note
in DSS use. Comput Hum Behav 2009; 25: 217–228.
on the perception of time in consumer behavior. J Consumer
34. Rieh SY and Danielson DR. Credibility: a multidisciplinary
Res 1984; 11: 615.
framework. Ann Rev Inform Sci Technol 2007; 41: 307–364.
55. Whiting A and Donthu N. Closing the gap between perceived
35. Singh J. A typology on consume dissatisfaction response and actual waiting times in a call center: results from a field
styles. J Retail 1990; 66: 57–99. study. J Serv Market 2009; 23: 279–288.
36. Arthur WB. Inductive reasoning and bounded rationality. Am
Econ Rev 1994; 84: 406–411.
37. de Cara MAR, Pla O and Guinea F. Competition, efficiency Author biographies
and collective behavior in the ‘‘El Farol’’ bar model. Eur Jorge A Alvarado Valencia is Associate Professor of
Phys J B Condens Matt Complex Syst 1999; 10: 187–191. Analytics at the Industrial Engineering Department at
38. Casti JL. Seeing the light at El Farol: a look at the most
Pontificia Universidad Javeriana. He has an MSc in
important problem in complex systems theory. Complexity
Analytics from North Carolina State University and a PhD
1996; 1: 7–10.
in Engineering from Pontificia Universidad Javeriana. His
39. Edmonds B. Modeling socially intelligent agents. Appl Artif
Intell 1998; 12: 677–699.
research interests are analytics of databases of human
40. Challet D, Marsili M and Ottino G. Shedding light on El behavior and human–computer interaction with business
Farol. Phys A Stat Mech Appl 2004; 332: 469–482. implications.
41. Siebers PO, Aickelin U, Celia H, et al. Simulating customer
experience and word-of-mouth in retail - a Case Study. Gabriela C Tueti Silva is a Production Engineering and
Simulation 2010; 86: 5–30. graduated from Universidad Simon Bolivar – Venezuela.
42. Negahban A and Yilmaz L. Agent-based simulation applica- She holds a master’s degree in Industrial Engineering
tions in marketing research: an integrated review. J Simulat from Pontificia Universidad Javeriana – Colombia. Since
2014; 8: 129–142. 2011 she has worked as project leader at Avianca airlines.
43. Grimm V, Berger U, Bastiansen F, et al. A standard protocol
for describing individual-based and agent-based models. Jairo R Montoya-Torres is Professor within the School
Ecol Model 2006; 198: 115–126. of Management at Universidad de los Andes, Bogotá,
44. Grimm V, Berger U, DeAngelis DL, et al. The ODD pro- Colombia. Since 2014 he has been a member of the
tocol: a review and first update. Ecol Model 2010; 221: Executive Board of the Colombia Association of
2760–2768. Operational Research (ASOCIO), and Vice-president for
45. Oliver RL. An investigation of the interrationship between the period 2016–2018. He holds the postdoctoral diploma
customer consumer (dis)satisfaction and complaint reports. ‘‘Habilitation for Research Direction’’ (HDR) in logistics
Adv Consumer Res 1987; 14: 218–222.
and operations research from the National Institute of
46. Prakash V. Intensity of dissatisfaction and consumer com-
Applied Sciences (INSA) of Lyon, France, and a PhD
plaint behaviors. J Consumer Satisfaction Dissatisfaction
degree in Industrial Engineering from École National
Complain Behav 1991; 4: 110–122.
47. Day RL. Modeling choices among alternative responses to
Supérieure des Mines de Saint-Etienne, France. His
dissatisfaction. Adv Consumer Res 1984; 11: 496–499.
research interests lie broadly in simulation and optimiza-
48. Kendall DG. Stochastic processes occurring in the theory of tion of logistics and production systems, scheduling, and
queues and their analysis by the method of the imbedded supply chain management under collaborative and sustain-
Markov Chain. Ann Math Statist 1953; 24: 338–354. able environments. He was co-editor of the Proceedings of
49. Aparicio R and Ortiz N. Modelamiento del tiempo de impa- the Winter Simulation Conference in 2010. His email
ciencia de los usuarios en una lı́nea de espera aplicado a un address is jr.montoya@uniandes.edu.co and his web page
casio empresarial: Pontificia Universidad Javeriana, http:// is http://jrmontoya.wordpress.com.
www.javeriana.edu.co/biblos/tesis/ingenieria/Tesis241.pdf
(2009, accessed 24 July 2016).

You might also like