- Email: [email protected]
A combined AHP–GP model for quality control systems
Int. J. Production Economics 72 (2001) 27}40
A combined AHP}GP model for quality control systems Masood A. Badri* Department of Management, College of Business and Economics, United Arab Emirates University, Post Box 17555, Al-Ain, United Arab Emirates Accepted 20 June 2000
Abstract Using the results of previous studies of service quality attributes, "ve sets of quality measures are identi"ed. These indicators or measures, through the analytic hierarchy process (AHP), are then accurately and consistently weighted. The priority weights are, in turn, incorporated in a goal-programming model to help select the `besta set of quality control instruments for customer data collection purposes. The paper proposes a decision aid that will allow weighting (prioritizing) of a "rm's unique service quality measures, consider the real world resource limitations (i.e., budget, hour, labor, etc.), and select the optimal set of service quality control instruments. The paper addresses two important issues: how to incorporate and decide upon quality control measures in a service industry, and how to incorporate the AHP into the model. A real world case study illustrates the application of this combined analytic hierarchy process and goal-programming (AHP}GP) model. 2001 Elsevier Science B.V. All rights reserved. Keywords: Analytic hierarchy process; Goal programming; Service quality
1. Introduction A problem encountered in designing quality control systems for service organizations is the measurement of the quality construct. Schniederjans and Karuppan [1] developed a goal-programming model to aid in selecting the `besta set of quality control instruments in designing a quality control system. Goal programming is a procedure for handling multiple-objective situations within the general framework of linear programming. Each objective is viewed as a goal. Then, given the usual resource limitations or constraints, the decision-maker attempts to develop decisions that pro* Corresponding author. Tel.: 971-3-7665644; fax: 971-37515538. E-mail address: [email protected] (M.A. Badri).
vide the `besta solution in terms of coming as close as possible to reaching all goals. Through a review of the literature they developed a list of indicators de"ning the quality construct. These indicators were, in turn, incorporated into a goal-programming model for the design of a quality control system in service organizations. A zero}one goalprogramming model was developed to help select the best set of quality control instruments. They presented a small business application to implement the model. The model employed a scoring method to rate the instruments (on a scale from 1 to 10). A simple scoring method was used to establish the priorities for the quality measures with regard to each instrument. One of the problems in any multi-objective method is the bias introduced by the initial solution provided by the selection process. The simple
0925-5273/01/$ - see front matter 2001 Elsevier Science B.V. All rights reserved. PII: S 0 9 2 5 - 5 2 7 3 ( 0 0 ) 0 0 0 7 7 - 3
28
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
scoring method employed by Schniederjans and Karuppan is highly susceptible to such bias where there is a tendency to adjust biases and anchor upon initial points. Moreover, Tversky and Kahneman [2], and Steuer [3] raise the issue that when simple scoring methods are used, the decisionmaker's consistency is not veri"ed. This method does not provide consistency feedback to the decision-maker. Saaty [4] points out that while decision-makers theoretically are paragons of consistency, in practice humans have been known to change their mind, either through reconsideration, or through the process of learning. The analytic hierarchy process (AHP) has been proposed as a means of reconciling initial decision-maker's expression of preference, as well as means of identifying the consistency of that expression. It provides an estimate of additive utility weight that best matches the initial information provided by the decision-maker. Moreover, when the AHP is used to obtain an initial estimate of the priorities, the initial points are selected on the bases of pairwise comparison of alternatives. The e!ect of introducing bias is lessened [5]. Even though Schniederjans and Karuppan did not use the AHP, they recommended its use to more accurately weight the importance of the quality measures. The purpose of this paper is to extend the model presented by Schniederjans and Karuppan [1] by using the AHP method to aid in accurately representing the goal-programming (GP) model's objective function and quality measure goals. The paper will demonstrate how current limitations in decision-making involving selecting quality control instruments can be overcome by combining the AHP and GP. The use of the proposed model is illustrated in a real world case study. 1.1. Service quality attributes in literature Early conceptualizations suggested several general service attributes that might be used to assess service quality. Sasser et al. [6] proposed three di!erent dimensions of service performance: levels of material, facilities, and personnel. Gronroos [7] proposed two types of service quality: technical quality (what customers actually receive from the service provider), and functional quality (the
manner in which customers receive the service). Lehtinen and Lehtinen [8] discussed three kinds of quality: physical, corporate, and interactive quality. Although there has been an avalanche of publications on service quality attributes, only few provided attributes that are developed and tested scienti"cally. Garvin [9,10] proposed eight dimensions to measure quality. He did not discriminate between goods producing or service providing "rms. The eight dimensions included performance, features, reliability, conformance, durability, serviceability, aesthetics, and perceived quality. DeSouza [11] reported service quality attributes from the Pro"t Impact of Market Strategy (PIMS) database. There were 12 attributes: delivery, warranty, repair and maintenance, sales services, corporate viability, advertising and promotional material, customization, technical support, location, complaint handling, ordering and billing simplicity, and communications. Parasuraman et al. [12] proposed 10 determinants of service quality that included reliability, responsiveness, competence, access, courtesy of personnel, communications, credibility or trustworthiness of the organization, security or protection from risk, understanding of customers' needs, and tangibles or physical elements attesting to the service. In a later study, these determinants were factor analyzed and generated "ve principal quality dimensions: tangibles, reliability, responsiveness, assurance, and empathy [13]. A consistent theme emerging from these dimensions is that customers might use more than just service outcome or `corea in assessing service quality. Customer assessment may also be in#uenced by the service process and the `peripheralsa associated with the service. The research conducted by Parasuraman et al. [12] con"rmed that both outcome and process dimensions in#uence customers' evaluation of service quality regardless of service sector. The 10 determinants of their study, and identi"ed earlier, constitute a more comprehensive set of service quality dimensions. However, the researchers acknowledged the possibility of overlapping dimensions. Through extensive empirical research, using statistical and psychometric tests, they developed and re"ned the SERVQUAL instrument to focus on "ve principal quality dimensions, which
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
29
Table 1 Summary of quality dimensions and attributes for service organizations Quality dimensions
De"nition of the dimension
Tangibles
Appearance of physical facilities, equipment, personnel, and communication personnel (modern-looking equipment, visually appealing physical facilities, neat-appearing employees, and visually appealing materials associated with the service)
Reliability
Ability to perform the promised service dependably and accurately (doing things by the promised time, showing sincere interest in solving problems, performing the service right the "rst time, and insisting on error-free records)
Responsiveness
Willingness to help customers and provide prompt service (employees telling exactly when services will be performed, employees giving prompt service, employees always willing to help, and employees never being too busy to respond to requests)
Assurance
Knowledge and courtesy of employees and their ability to inspire trust and con"dence (behavior of employees instilling con"dence in customers, customers feeling safe in their transaction with the "rm, employees constantly being courteous, and employees having the knowledge to answer questions)
Empathy
Caring, individualized attention the "rm provides its customers (giving individual attention, having convenient operating hours, giving personal attention, having the best interest at heart, and understanding the speci"c needs of customers)
has become the most widely used methodology for measuring service quality (see Table 1). This paper will utilize the dimensions of quality as derived by Parasuraman et al. [14]. The SERVQUAL instrument has generated considerable interest in service quality measurement among academic researchers. It has gone through several enhancement stages as a direct reaction to some concerns raised by other researchers with regard to its expectations component [15], the interpretation and operationalization of expectations [16], its reliability and validity [17], and its dimensionality [18]. In response to these questions, the three colleagues have presented counter arguments, clari"cations, and additional evidence to rea$rm the instrument's psychometric soundness and practical value [13,19,20]. The instrument has served as the basis for measurement approaches used in published studies examining service quality in a variety of contexts } e.g., real estate brokers [21]; physicians in private practice [22]; public recreation programs [23]; motor carrier companies [24]; a business school placement center, and a tire shop [25]; an accounting "rm [26]; discount and department stores [16]; a gas and electric utility company [27]; hospi-
tals [25]; banking, pest control, dry cleaning, and fast food [28]; and higher education [29]. For all these reasons, the basic dimensions of SERVQUAL will serve as attributes of service quality and will be used in this paper.
2. Selecting the service quality control instruments 2.1. Weighting the quality measures: The AHP approach The AHP, introduced by Saaty [4], addresses how to determine the relative importance of a set of activities in a multi-criteria decision problem. The process makes it possible to incorporate judgments on intangible qualitative criteria alongside tangible quantitative criteria. The method utilizes pairwise comparisons of alternatives (quality control instruments) as well as pairwise comparisons of the multiple criteria (the "ve attributes of service quality). The use of such pairwise comparisons to collect data from the decision-maker o!ers signi"cant advantages [30]. It allows the decision-maker to focus on the comparison of just two objects, which makes the observation as free as possible from
30
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
extraneous in#uences. Additionally, pairwise comparisons generate meaningful information about the decision problem, improving consistency in the decision-making process, especially if the process involves group decision-making. The `quality control systema model is given in the form of a hierarchy (Fig. 1). The decision problem is structured in a hierarchy of three basic levels: the goal of the decision is at the top, followed by a second level of criteria (or objectives) and a third level of alternatives. This arrangement makes it possible for decision-makers to focus on each and every part of such a complex problem, and to derive priorities from simple pairwise comparisons. In this study, the goal of the AHP model will be to decide on prioritizing the set of quality control instruments given some quality of service criteria. The criteria or objectives of the quality system decision consist of "ve elements: reliability, assurance, responsiveness, empathy, and tangibles. The decision alternatives, the options from which a choice is made, are the quality instruments for customer data collection purposes. Once the model is built, the decision-makers evaluate the elements by making pairwise comparisons. A pairwise comparison is the process of comparing the relative importance, preference, or likelihood of two elements with respect to an element in the level above. When all the comparisons are completed, we calculate the priorities and
a measure of consistency of our judgment. Generally the consistency ratio should be less than 0.10 (10%) [31]. As a result, we derive priorities and an inconsistency ratio with respect to each of the "ve quality-of-service criteria. The next step is to make judgments about the "ve criteria. A comparison is made with respect to each pair [the number of comparisons will be u(u!1)/2, where u is the number of criteria in the model]. The next step will be to synthesize the derived priorities that were based on the decision-makers' judgments. Synthesis means adding up the global weights of the common nodes at the bottom level of the hierarchy so as to generate a composite priority for an alternative across all criteria. Synthesizing shows the results of the entire work, the overall priorities of the alternatives. These alternative priorities will be used later in the combined AHP}GP model as weights in the objective function. The derived priorities with respect to each of the "ve quality-of-service criteria will be used in the combined model to serve as the contribution that each criterion makes to each alternative. 2.2. The goal-programming model In the goal-programming model, the decision variable is x (0 or 1). The objective function, given G by Eq. (1), seeks to minimize deviation from desired targets for limited resources (costs, available
Fig. 1. Structure of the AHP model.
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
management hours, and hours)
available
employee
Min Z"P (d\, d>)#P (d\, d>)#P (d\, d>). (1) A A A @ @ @ C C C The goal constraints in Eqs. (2), (3) and (4) represent the availability of limited resources. The right-hand side of each equation re#ects the targeted or desired level of the resource utilization, where C denotes cost, B available management hours, and E available employee hours. We could also express these limitations of available resources as `system constraintsa by removing the deviation variables from the constraints and the objective function and by changing the equality signs of the constraints to less than or equal signs. For convenience, this will be done through the combined model: K a x #d\!d>"C, (2) AG G A A G K a x #d\!d>"B, (3) @G G @ @ G K a x #d\!d>"E. (4) CG G C C G The ordering of these goal constraints depends on the nature of the problem situation. Either preemptive or non-preemptive goals could be used depending on the order of importance, if any, of the goals. 2.3. The combined model There are several studies that used the AHP methodology in combination with goal programming [30,32}35]. In the combined model, the objective function also includes deviation variables associated with the quality measure goals. It will seek to minimize such deviations from desired levels. The revised objective function is given in Eq. (5). Moreover, a set of constraints, as shown in Eq. (6), will be added to re#ect the quality target of Q in each of the `goal constraintsa. An equation I associated with the AHP weights for the quality control instruments will be added to re#ect the preferences for the di!erent instruments. This is
31
given in Eq. (7):
) Min Z" P (w d\, w d>) #P (d\, d>) I I I I I ? ? ? I #P (d\, d>)#P (d\, d>)#P (d\, d>), A A A @ @ @ C C C (5) K a x #d\!d>"Q (for k"1,2, 2, K), (6) IG G I I I G K w x #d\!d>"1. (7) G G ? ? G Eq. (6) shows that there will be one equation for each of the quality measures. The ordering of goal constraints in Eq. (6) depends on the nature of the problem situation. The classic lexicographic ordering of preemptive priorities may be appropriate if the highest ranked quality measures are preemptively important. To better utilize the results of the AHP method, the a IG will denote the contribution that each criterion makes to each alternative. Compared to previous research, and as shown in Eq. (1), more appropriate weighting of quality measures is desired and used. By utilizing the obtained AHP weightings we have established the ranks of the individual quality control instrument, by establishing a relative weighting w for each I of them. The term P is some k priority I lexicographic rankings where P 'P '2'P . I In this model, Eq. (1) is the objective function, which seeks to minimize deviation from the desired goals consistent with the AHP ranking of the quality control instruments. The lexicographic nature of goal programming treats the AHP weights given by w as a sub-ranking I of the quality control instruments within their speci"c P . The greater the w , the more desirable I I the selection of quality instrument in the decision process. As mentioned before, the constraints for limited resources of total cost, available management hour, and available employee hour could also be expressed as `system constrainta by removing the deviation variables and replacing the equality signs with inequality signs in Eqs. (2)}(4).
32
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
3. Application of the combined model
3.1. Deriving the AHP weights
Based on a real life problem involving a large department store, the proposed AHP}GP methodology is illustrated to show how it is applied. The department store is evaluating seven potential alternatives to better design a quality control system for customer data collection purposes. Unfortunately, each instrument captures the principal quality measures to a di!erent degree and requires varying levels of resources. The problem for a large department store with limited resources is therefore to select the best set of instruments to obtain customer feedback and perception on key quality measures. Our model requires a group process rather than aggregation of individual inputs. In this process, the group establishes a single set of weights for the decision criteria, and then rates the decision alternatives. The group process has several advantages over aggregation of individual ratings [36]. It facilitates a common understanding of the meaning and signi"cance of each criterion. This commonality of understanding is not achieved through aggregating the inputs of individual evaluations. The group is often able to clarify misunderstanding and di!erences in interpretation of the data so that there is a more uniform understanding of the facts. In addition, a group process utilizes the dynamics of powerful in#uence within the decision-making group. To provide expert judgment and consultations in a group process, the department store management formed a team, which consisted of "ve members to work side by side with the research team. The "ve members consisted of the general managers of marketing, "nance and personnel, and the two heads of advertising and public relations. The research was given full support by the top management. To derive the AHP weights, a preliminary data collection stage attempted to get some reaction and feedback with regard to the "ve quality attributes. A survey of known customers was carried out to identify the relative importance of each criterion. The results obtained, along with the management-team's input were used to later structure and synthesize the AHP model.
To facilitate the use of AHP, the problem is decomposed into a multilevel hierarchy showing the overall goal of the decision process, each decision criterion to be used, and the decision alternatives to be considered. The "ve decision criteria are: reliability, empathy, responsiveness, tangibles, and assurance. The decision alternatives to be considered are: large-scale mail survey of known customers, large-scale telephone survey of known customers, open-ended comment cards, personal interviews of customers while shopping, point of purchase survey to measure employee service delivery, point of purchase survey to measure product line service delivery, and small-scale telephone survey of known customers. The multilevel hierarchy is shown in Fig. 2. A relative preference matrix is formulated for each subordinate level of hierarchy. The decisionmaking team was given the chance to express its preferential expert judgment (the pairwise comparisons) in the selection process. Each decision alternative is compared against another decision alternative with respect to one particular decision criterion at a time. These within-pairwise comparisons result in priority values of each decision alternative by each decision criterion. The result of elicitation at this level of the hierarchy is a matrix showing the values expressed by the decision-making team. The `right eigenvector methoda is applied to each of the matrices in turn using Expert Choice Pro software (version 9.0) [37]. From this application, the relative priority for each alternative decision by criterion is computed. The results are shown in Table 2. During the elicitation process, an appropriate level of consistency is necessary to achieve meaningful results. The measure of consistency provided by the AHP is given by Saaty [38]: "( !n)/(n!1), where is the maximum
eigenvalue of the positive reciprocal matrix, and n is the number of objectives. Maximum consistency occurs as approaches zero. Researchers use a consistency ratio of 0.10 or less as guidelines in evaluating consistencies [31]. As can be seen in Table 2, all of the ratios are below the maximum 0.10 level. Via `synthesizinga [31, p. 159] the sets of priority
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
33
Fig. 2. The AHP model for a department store. Table 2 Resulting priority values for each decision alternative by each criterion Decision alternatives
x x x x x x x Total Inconsistency ratio
Service quality criterion decisions Responsiveness
Assurance
Reliability
Empathy
Tangibles
0.031 0.335 0.036 0.099 0.099 0.099 0.302
0.219 0.160 0.145 0.068 0.070 0.130 0.208
0.028 0.397 0.041 0.067 0.071 0.105 0.291
0.026 0.376 0.040 0.165 0.078 0.075 0.240
0.415 0.055 0.097 0.154 0.034 0.036 0.208
1.000 0.050
1.000 0.010
1.000 0.020
1.000 0.050
1.000 0.030
values (from Table 2) are combined into a matrix and multiplied by the overall criteria relative priorities as presented in Table 3. The result is an overall prioritization of the decision alternatives. The results of this step, presented in Table 4, are the
overall rankings (in terms of weights) of the seven quality instrument candidates. Before we discuss the combined model application, and for simpli"cation, Fig. 3 provides an illustration of the combined model development steps.
34
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
Table 3 Resulting priority values for each criterion Criterion decisions
Priority
Responsiveness Assurance Reliability Empathy Tangibles Total
0.479 0.139 0.281 0.025 0.076 1.000
Table 4 Overall AHP weights of the decision alternatives Decision alternatives
AHP weighting
Decision preference
X x x x x x x Total
0.086 0.308 0.057 0.091 0.081 0.099 0.277
Fifth preference First preference Seventh preference Fourth preference Sixth preference Third preference Second preference
1.000 Fig. 3. Illustrative development.
diagram
of
the
combined
model
3.2. The combined AHP}GP application Goal programming permits resource limitations such as budgetary limitations or limited hours of labor, and other selection limitations (such as system constraints) that must be observed in establishing and maintaining a service quality control system. Trying to decide which of the seven service control instruments to use in designing the department store's service quality control system requires the recognition of several resource limitations relative to the nature of each of the instruments. Table 5 provides a summary of yearly cost and labor information on the seven service quality control instruments. The decision team felt that a minimum of three of the seven instruments would be needed to justify the expense of the budget. Also, the nature of some of the quality instruments created duplication that would be wasteful and unnecessary. For example, the minor telephone survey was just a scaled down
version of the large-scale telephone survey. As a result, system constraints would be needed to prevent such duplications. The decision team also wanted to ensure that the selection of quality control instruments would measure those quality attributes that were most important to their customers. To determine the ability of each of the instruments to measure each of the "ve quality measures presented in Table 1, the AHP scores (priority values) for each decision alternative by each criterion will be used. The priority scores are presented in Table 2. We recall that the derived priority scores are the relative contribution that each of the "ve criteria makes to each of the seven alternatives. Since it is our objective to select quality instruments with the highest weights, a goal constraint could be added to maximize the overall weights of
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
35
Table 5 Yearly cost and labor information on service quality control instruments Resource item
Estimated cost Management hours Employee hours
Service quality control instruments (decision alternatives)
Total budgeted yearly resources
x
x
x
x
x
x
x
$9760 186
$9760 186
$1207 54
$640 50
$852 98
$2500 98
$4738 42
$15,000 530
47
47
57
93
278
493
600
1070
the instruments selected. We recall that the AHP solution provided us with overall weights for each of the seven quality instruments (alternatives). These weights are presented in Table 4. 3.3. Formulation of the department store's problem System constraints are needed to reduce resource wastage by not selecting similar quality control instruments. In our case, the minor telephone survey was just a scaled down version of the major telephone survey. The management team also requested that only one form of large-scale survey of known customers be performed, a telephone or mail survey. We should recall also, that at least three instruments are to be selected. Mathematically, these `system constraintsa are treated as simple linear programming constraints. Typically, they create resource boundaries where an interval of possible solutions is usually narrowed down to the optimal choice by subsequently considering the priority of goal constraints: x #x )1 and x #x )1, (8) x #x #x #x #x #x #x *3, (9) 9760x #9760x #1207x #640x #852x #2500x #4738x #d\!d>"15,000. (10) A A The three resource limitation constraints will have deviation variables associated with them, and will attempt to minimize the positive deviations by adding deviation variables to the overall objective function. The department store allocated a maximum yearly operating budget limitation of $15,000 for the cost of materials, postage, phone calls, printing, etc., for all of the quality control instruments
selected. The equation shows that point of purchase surveys to measure employee service delivery and product line service delivery are most expensive with respect to annual cost. Moreover, the management of the department store limited the yearly number of hours that management could devote to the service of quality control instruments to 530 hours, and the number of employee labor hours for data collection, interviews and clerical activities to a maximum of 1070 hours: 186x #186x #54x #50x #98x #98x #42x #d\!d>"530, (11) @ @ 47x #47x #57x #93x #278x #493x #600x #d\!d>"1070. (12) C C The three equations associated with resource limitations could also be formulated as system constraints if desired by decision-makers. However, the management team asked to keep them as goal constraints to better understand their contribution to the overall structure of the model. The management of the department store wanted to also ensure that the selection of the quality control instruments would really measure those quality measures that were important to their customers. One goal constraint is needed to ensure that the instruments with the highest weights obtained from the AHP analysis will be selected. Such goal constraint will attempt to maximize the weights by selecting the quality control instruments with the highest priorities: 0.086x #0.308x #0.057x #0.091x #0.081x #0.099x #0.277x #d\!d>"1. (13) N N
36
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
To determine the ability of the instruments to measure each of the "ve measures presented in Table 1, the AHP priority values (or weights) for each quality control instrument by each quality measure will be used. The individual weights represent the a parameters in the combined model. IG The development of the `goal constraintsa in the situation originates chie#y from the a data in IG Table 2. The right-hand-side values for each goal constraint seek to pressure a selection of service quality control instruments or x with the highest G scores (i.e., the most useful collection of instruments to measure quality). In other words, the best set of three quality instruments was chosen for each of the "ve quality measures. Of course, the right-hand side could be set at any level; however for our case, each of the right-hand-side Q was simply derived, I by summing the best three a , for each of the "ve IG quality measures. The model seeks to select the best or perfect set of quality instruments, for each of the quality measures: 0.031x #0.335x #0.036x #0.099x #0.099x #0.099x #0.302x #d\ !d> "0.736, (14) I I 0.219x #0.160x #0.145x #0.068x #0.070x #0.130x #0.208x #d\ #d> "0.587, (15) I I 0.028x #0.397x #0.041x #0.067x #0.071x #0.105x #0.291x #d\ !d> "0.793, (16) I I 0.026x #0.376x #0.040x #0.165x #0.078x #0.075x #0.240x #d\ !d> "0.784, (17) I I 0.415x #0.055x #0.097x #0.154x #0.034x #0.036x #0.208x #d\ #d> "0.777. (18) I I The objective function will attempt to minimize the overall deviations in each of the goal constraints. The goal constraints include resource limitation constraints as well as desired quality measure goals: Min Z"P (d>#d>#d>)#P d\ A @ C N #P (0.479d\ #0.139d\ #0.281d\ I I I #0.025d\ #0.076d\ ). (19) I I
The ordering of deviation variables in the objective function associated with goal constraints (d }d ) will rely on the data obtained via the I I AHP process and reported in Table 3. We recall that the order of the "ve measures of service quality were responsiveness, reliability, assurance, tangibles, and empathy, respectively. The three measures of responsiveness, reliability, and assurance accounted for 89.9% of the overall weights, while the remaining two measures of empathy and tangibles accounted for only 10.1% of the total weight. The results also re#ect the importance of the top three measures as the quality control instruments that are selected portray these facets of quality. For example, the selection of x (point of purchase survey to measure employee service delivery) re#ects the importance of responsiveness (0.335), reliability (0.397), and assurance (0.160). We also notice that the selection of x which assigns its largest weights to empathy (0.165) and tangibles (0.154) attempts to help in the creation of an overall quality control system that does not neglect the contribution of the two measures of empathy and tangibles. Management was satis"ed with the solution since the selected instruments, in general, tried to re#ect the department store's image to its customers through the "ve measures of service quality. We could simply use the order of importance as reported in the table. Meanwhile, we could make better use of the AHP process results by utilizing the speci"c weights provided in the table. In this case, as mentioned earlier, the AHP weights provide internal relative rankings (or desirability scores) for each of the quality measures. The objective function will seek to minimize the overall deviations. 3.4. Solution of the combined model The solution to this problem was generated using LINDO, Version 6.0 [39]. Table 6 presents the optimal selection of service quality control instruments. The optimal selection turned out in this case study to be where the department store will use point of purchase survey to measure employee service delivery (x ), open-ended comment cards (x ), small-scale telephone survey of known customers
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
37
Table 6 The combined AHP}GP model solution (a) Decision variables X "1 X "1 X "1 X "1 X "0 X "0 X "0
Select point of purchase survey to measure employee service delivery Select open-ended comment cards Select small-scale telephone survey of known customers Select large-scale mail survey of known customers Do not select point of purchase to measure product line service delivery Do not select large-scale telephone survey of known customers Do not select personal interviews of customers while shopping
(b) Slack in resource system constraints Constraints
Usage
Total available
Slack
Budget Management hours Employee hours
$14,107 388 690
$15,000 530 1070
$893 142 380
(c) Deviation in quality measure constraints Constraint
Target
Obtained
Underachievement
Responsiveness Assurance Reliability Empathy Tangibles
0.736 0.587 0.793 0.784 0.777
0.569 0.503 0.610 0.656 0.342
0.167 0.084 0.183 0.128 0.435
(x ), and large-scale mail survey of known cus tomers (x ). The selected instruments will act as a collective system designed to monitor performance on the "ve quality measures de"ned in Table 1. The other three instruments were not chosen because of the limitations posed by the system constraints, goal constraints, as well as the assigned priorities. The slack in the goal constraints associated with resource limitations that were used to model cost and hours usage also provides some summary information on those resources. The amount of deviation actually provides a crude measure of deviation of the resulting quality control system in measuring the "ve quality attributes. The smaller the resulting deviation, the better the system measures the respective quality attributes. The high-quality attributes like responsiveness, reliability, and assurance should be more comprehensively measured under the proposed quality control sys-
tem than low-priority attributes such as empathy and tangibles. The management of the department store was delighted to see that the resulting quality control system would be achievable under-budget (a saving of $893). With regard to the "ve quality measures, we note that the resulting solution adequately addresses the attributes with the highest priorities. For example, with regard to the measures of responsiveness, reliability, and assurance, the three attributes with the highest priorities, they are adequately represented by the selected instruments. Being happy with cost "gures of the model solution, management requested a scenario to be checked where the priorities for the "ve quality measures are placed ahead of the resource limitation goals in the objective function. The scenario called for a solution that overshoots the budget by $11,105, which was totally unacceptable. Other scenarios were also looked at but most of them
38
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
a!ected the budget cost and therefore were not considered further. Due to the importance of the three quality measures of responsiveness, reliability, and assurance as the top three highest priorities, a righthand-side sensitivity analysis was undertaken on these three goals (mainly Eqs. (14)}(16)). Results show that substantial changes to the right-handside goals must occur before altering the existing solution. In other words, the management team that assisted in this research would have had erred in the assessment of the weights assigned to the "ve quality measures. The department store management implemented the suggested results obtained from the combined AHP}GP model in the summer of 1998. After almost one year of its implementation, the top management of the department store has conveyed its complete satisfaction with the results. The implementation of the study was reported several times in the daily-newspapers of the city. Management feels that many customers have noticed increased awareness in service quality, re#ecting to some degree the comprehensiveness of the quality control system to reach the customers. The top management has also promised more money to be available for future implementations of similar projects.
4. Concluding remarks The quality constructs suggested in this paper for designing quality control systems in service organizations were measured via "ve criteria (or service attributes): responsiveness, reliability, assurance, empathy, and tangibles. The combined AHP}GP model presented has helped a large department store select from a set of seven survey instruments those four that will collectively measure the ranked quality measures de"ned by the management team. The AHP is "rst used to prioritize the set of quality control instruments along the "ve measures in a consistent manner. The selected set of four survey instruments collectively represents the department store's service quality control systems. The AHP process was used to derive weights for each of these quality measures using solicitations from both cus-
tomers and management. The derived weights were used to establish preemptive priorities in the objective function, and utility weights for the "ve quality measures in the goal constraints. The proposed combined model has saved resources by recognizing necessary constrained resources. It also presented a more thorough decision process. For example, the completeness and non-bias nature of the pairwise comparisons in the AHP method was a more thorough means of weighting quality control instruments in the decision process relative to ranking methods employed in previous studies dealing with quality control instrument selection. The most obvious advantage of using the AHP model for deriving weights for the quality control instruments as well as measures is that it provides for consistent decision-making. All instruments are evaluated on a single set of weighted criteria (quality control measures). This should help to reduce the subjectivity of the process. In addition, using a mathematical programming method like goal programming allowed for additional trade-o! information that permitted decision-makers to see these tradeo!s in terms of cost and hours. The combined AHP}GP method o!ers a systematic, easy-to-use approach to the service quality control instruments selection decision problem. It extends previous research in the area by incorporating a comprehensive prioritization system within an optimization resource allocation process. The method accurately represents the model's objective function and quality measure goals, while avoiding some pitfalls of goal-programming modeling encountered in previous studies. The management team formally presented the model to top management of the department store. The top management viewed the model's structure and mechanism favorably, believing that it is a tool that can be used in the development of a quality control system and ordered the immediate implementation of its results. Top management has suggested that the research team should continuously re"ne the hierarchy, modify the judgments, and generate new set of weights, since customer taste is dynamically sensitive. In addition, they have suggested that the model should also expand on the resource limitation part to include others such as
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
advertisement costs, employee training costs for better implementation of some of the instruments, and other costs associated with providing customers with promotion gifts to better solicit their participation in data collection. They also recommended that future e!orts should also examine the viability of other forms of quality control instruments such as home visits of some loyal customers, personal interviews with suppliers and personal interviews with service providing sta! members. A revised model should be ready for implementation in the fall of 2000.
References [1] M. Schniederjans, C. Karuppan, Designing a quality control system in a service organization: A goal programming case study, European Journal of Operational Research 81 (1995) 249}258. [2] A. Tversky, D. Kahneman, Judgment under uncertainty: Heuristics and biases, Science 185 (1974) 1124}1131. [3] R. Steuer, An interactive multiple objective linear programming procedure, TIMS Studies on Management Science 6 (1977) 225}239. [4] T. Saaty, Ascaling method for priorities in hierarchical structures, Journal of Mathematical Psychology 15 (1977) 234}281. [5] D. Olson, M. Venkataraman, J. Mote, Atechnique using the analytic hierarchy process in multiple objective planning models, Socio-Economic Planning Science 20 (6) (1986) 361}368. [6] W. Sasser, P. Olsen, D. Wycko!, Management of Service Operations: Text and Cases, Allyn and Bacon, Boston, MA, 1978. [7] C. Gronroos, Strategic management and marketing in the service sector, Swedish School of Economics and Business Administration, Helsinki, Finland, 1982. [8] U. Lehtinen, J. Lehtinen, Service quality: A study of quality dimensions, Unpublished Research Report, Service Management Group OY, Finland, 1982. [9] D. Garvin, What does `product qualitya really mean? Sloan Management Review 26 (1) (1984) 25}43. [10] D. Garvin, Competing on the eight dimensions of quality, Harvard Business Review 65 (5) (1987) 101}109. [11] G. DeSouza, Now service business must manage quality, Journal of Business Strategy 10 (3) (1989) 21}25. [12] A. Parasuraman, V. Zeithaml, L. Berry, A conceptual model of service quality and its implications for future research, Journal of Marketing 49 (4) (1985) 41}50. [13] A. Parasuraman, V. Zeithaml, L. Berry, Reassessment of expectations as a comparison standard in measuring service quality: Implications for further research, Journal of Marketing 58 (1) (1994) 111}124.
39
[14] A. Parasuraman, V. Zeithaml, L. Berry, SEVQUAL: a multiple-item scale for measuring consumer perceptions of service quality, Journal of Retailing 64 (1) (1988) 12}40. [15] S. Taylor, J. Nicholson, J. Milan, R. Martinez, Assessing the roles of service quality and customer satisfaction in the formation of the purchase intentions of Mexican consumers, Journal of Marketing Theory and Practice 5 (1) (1997) 78}90. [16] R. Teas, Expectations, performance evaluation and consumer's perceptions of quality, Journal of Marketing 57 (4) (1993) 18}34. [17] T. Brown, G. Churchill, P. Peter, Improving the measure of service quality, Journal of Retailing 69 (1) (1993) 127}139. [18] D. Finn, C. Lamb, An evaluation of the SERVQUAL scales in retail setting, in: R.H. Holman, M.R. Solomon (Eds.), Advances in Consumer Research, Vol. 18, Association for Consumer Research, Provo, UT, 1991. [19] A. Parasuraman, L. Berry, V. Zeithaml, Re"nement and reassessment of the SERVQUAL scale, Journal of Retailing 67 (1991) 420}450. [20] A. Parasuraman, L. Berry, V. Zeithaml, More on improving service quality measurement, Journal of Retailing 69 (1993) 40}147. [21] L. Johnson, M. Dotson, B. Dunlop, Service quality determinants and e!ectiveness in the real estate brokerage industry, The Journal of Real Estate Research 3 (1988) 21}36. [22] S. Brown, T. Swartz, Agap analysis of professional service quality, Journal of Marketing 53 (1989) 92}98. [23] J. Crompton, K. Mackay, User's perceptions of the relative importance of service quality dimensions in selected public recreation programs, Leisure Sciences 11 (1989) 367}375. [24] R. Brensinger, D. Lambert, Can the SERVQUAL scale be generalized to business-to-business services? Knowledge Development in Marketing, 1990 AMA's Summer Educators' Conference Proceedings, 1990. [25] J. Carman, Consumer perceptions of service quality: An assessment of the SERVQUAL dimensions, Journal of Marketing Research 66 (1) (1990) 33}55. [26] D. Bojanic, Quality measurement in professional service "rms, Journal of Professional Services Marketing 7 (2) (1991) 27}36. [27] E. Babakus, G. Boller, An empirical assessment of the SERVQUAL scale, Journal of Business Research 24 (1992) 253}268. [28] J. Cronin, S. Taylor, Measuring service quality: A reexamination and extension, Journal of Marketing 56 (1992) 55}68. [29] W. Boulding, A. Kalra, R. Staelin, V. Zeithaml, A dynamic process model of service quality: From expectations to behavioral intentions, Journal of Marketing Research 30 (1993) 7}27. [30] M. Schniederjans, R. Wilson, Using the analytic hierarchy process and goal programming for the information system project selection, Information and Management 20 (1991) 333}342.
40
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
[31] T. Saaty, Decision Making for Leaders, RWS Publications, Pittsburgh, PA, 1995. [32] N. Bryson, A goal-programming method for generating priority vectors, Journal of the Operational Research Society 46 (5) (1995) 641}648. [33] D. Despotis, Fractional minimax goal-programming: A uni"ed approach to priority estimation and preference analysis in MCDM, Journal of the Operational Research Society 47 (8) (1996) 989}999. [34] S. Myint, M. Tabucanon, Amultiple criteria approach to machine selection for #exible manufacturing systems, International Journal of Production Economics 33 (1}3) (1994) 121}131. [35] M. Schniederjans, T. Garvin, Using the analytic hierarchy process and multi-objective programming for the selection
[36]
[37] [38] [39]
of cost drivers in activity-based costing, European Journal of Operational Research 100 (1997) 72}80. E. Erkut, S. Moran, Locating obnoxious facilities in the public sector: An application of the analytic hierarchy process to the municipal land"ll sitting decisions, Socio-Economic Planning Science 25 (2) (1991) 89}102. Expert Choice, Decision Support Software, Pittsburgh, PA, 1995. T. Saaty, The Analytic Hierarchy Process, McGraw-Hill, New York, 1980. L. Schrage, Optimization Modeling with LINDO, 5th Edition, Brooks/Cole Publishing Co., Paci"c Grove, CA, 1997.
A combined AHP}GP model for quality control systems Masood A. Badri* Department of Management, College of Business and Economics, United Arab Emirates University, Post Box 17555, Al-Ain, United Arab Emirates Accepted 20 June 2000
Abstract Using the results of previous studies of service quality attributes, "ve sets of quality measures are identi"ed. These indicators or measures, through the analytic hierarchy process (AHP), are then accurately and consistently weighted. The priority weights are, in turn, incorporated in a goal-programming model to help select the `besta set of quality control instruments for customer data collection purposes. The paper proposes a decision aid that will allow weighting (prioritizing) of a "rm's unique service quality measures, consider the real world resource limitations (i.e., budget, hour, labor, etc.), and select the optimal set of service quality control instruments. The paper addresses two important issues: how to incorporate and decide upon quality control measures in a service industry, and how to incorporate the AHP into the model. A real world case study illustrates the application of this combined analytic hierarchy process and goal-programming (AHP}GP) model. 2001 Elsevier Science B.V. All rights reserved. Keywords: Analytic hierarchy process; Goal programming; Service quality
1. Introduction A problem encountered in designing quality control systems for service organizations is the measurement of the quality construct. Schniederjans and Karuppan [1] developed a goal-programming model to aid in selecting the `besta set of quality control instruments in designing a quality control system. Goal programming is a procedure for handling multiple-objective situations within the general framework of linear programming. Each objective is viewed as a goal. Then, given the usual resource limitations or constraints, the decision-maker attempts to develop decisions that pro* Corresponding author. Tel.: 971-3-7665644; fax: 971-37515538. E-mail address: [email protected] (M.A. Badri).
vide the `besta solution in terms of coming as close as possible to reaching all goals. Through a review of the literature they developed a list of indicators de"ning the quality construct. These indicators were, in turn, incorporated into a goal-programming model for the design of a quality control system in service organizations. A zero}one goalprogramming model was developed to help select the best set of quality control instruments. They presented a small business application to implement the model. The model employed a scoring method to rate the instruments (on a scale from 1 to 10). A simple scoring method was used to establish the priorities for the quality measures with regard to each instrument. One of the problems in any multi-objective method is the bias introduced by the initial solution provided by the selection process. The simple
0925-5273/01/$ - see front matter 2001 Elsevier Science B.V. All rights reserved. PII: S 0 9 2 5 - 5 2 7 3 ( 0 0 ) 0 0 0 7 7 - 3
28
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
scoring method employed by Schniederjans and Karuppan is highly susceptible to such bias where there is a tendency to adjust biases and anchor upon initial points. Moreover, Tversky and Kahneman [2], and Steuer [3] raise the issue that when simple scoring methods are used, the decisionmaker's consistency is not veri"ed. This method does not provide consistency feedback to the decision-maker. Saaty [4] points out that while decision-makers theoretically are paragons of consistency, in practice humans have been known to change their mind, either through reconsideration, or through the process of learning. The analytic hierarchy process (AHP) has been proposed as a means of reconciling initial decision-maker's expression of preference, as well as means of identifying the consistency of that expression. It provides an estimate of additive utility weight that best matches the initial information provided by the decision-maker. Moreover, when the AHP is used to obtain an initial estimate of the priorities, the initial points are selected on the bases of pairwise comparison of alternatives. The e!ect of introducing bias is lessened [5]. Even though Schniederjans and Karuppan did not use the AHP, they recommended its use to more accurately weight the importance of the quality measures. The purpose of this paper is to extend the model presented by Schniederjans and Karuppan [1] by using the AHP method to aid in accurately representing the goal-programming (GP) model's objective function and quality measure goals. The paper will demonstrate how current limitations in decision-making involving selecting quality control instruments can be overcome by combining the AHP and GP. The use of the proposed model is illustrated in a real world case study. 1.1. Service quality attributes in literature Early conceptualizations suggested several general service attributes that might be used to assess service quality. Sasser et al. [6] proposed three di!erent dimensions of service performance: levels of material, facilities, and personnel. Gronroos [7] proposed two types of service quality: technical quality (what customers actually receive from the service provider), and functional quality (the
manner in which customers receive the service). Lehtinen and Lehtinen [8] discussed three kinds of quality: physical, corporate, and interactive quality. Although there has been an avalanche of publications on service quality attributes, only few provided attributes that are developed and tested scienti"cally. Garvin [9,10] proposed eight dimensions to measure quality. He did not discriminate between goods producing or service providing "rms. The eight dimensions included performance, features, reliability, conformance, durability, serviceability, aesthetics, and perceived quality. DeSouza [11] reported service quality attributes from the Pro"t Impact of Market Strategy (PIMS) database. There were 12 attributes: delivery, warranty, repair and maintenance, sales services, corporate viability, advertising and promotional material, customization, technical support, location, complaint handling, ordering and billing simplicity, and communications. Parasuraman et al. [12] proposed 10 determinants of service quality that included reliability, responsiveness, competence, access, courtesy of personnel, communications, credibility or trustworthiness of the organization, security or protection from risk, understanding of customers' needs, and tangibles or physical elements attesting to the service. In a later study, these determinants were factor analyzed and generated "ve principal quality dimensions: tangibles, reliability, responsiveness, assurance, and empathy [13]. A consistent theme emerging from these dimensions is that customers might use more than just service outcome or `corea in assessing service quality. Customer assessment may also be in#uenced by the service process and the `peripheralsa associated with the service. The research conducted by Parasuraman et al. [12] con"rmed that both outcome and process dimensions in#uence customers' evaluation of service quality regardless of service sector. The 10 determinants of their study, and identi"ed earlier, constitute a more comprehensive set of service quality dimensions. However, the researchers acknowledged the possibility of overlapping dimensions. Through extensive empirical research, using statistical and psychometric tests, they developed and re"ned the SERVQUAL instrument to focus on "ve principal quality dimensions, which
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
29
Table 1 Summary of quality dimensions and attributes for service organizations Quality dimensions
De"nition of the dimension
Tangibles
Appearance of physical facilities, equipment, personnel, and communication personnel (modern-looking equipment, visually appealing physical facilities, neat-appearing employees, and visually appealing materials associated with the service)
Reliability
Ability to perform the promised service dependably and accurately (doing things by the promised time, showing sincere interest in solving problems, performing the service right the "rst time, and insisting on error-free records)
Responsiveness
Willingness to help customers and provide prompt service (employees telling exactly when services will be performed, employees giving prompt service, employees always willing to help, and employees never being too busy to respond to requests)
Assurance
Knowledge and courtesy of employees and their ability to inspire trust and con"dence (behavior of employees instilling con"dence in customers, customers feeling safe in their transaction with the "rm, employees constantly being courteous, and employees having the knowledge to answer questions)
Empathy
Caring, individualized attention the "rm provides its customers (giving individual attention, having convenient operating hours, giving personal attention, having the best interest at heart, and understanding the speci"c needs of customers)
has become the most widely used methodology for measuring service quality (see Table 1). This paper will utilize the dimensions of quality as derived by Parasuraman et al. [14]. The SERVQUAL instrument has generated considerable interest in service quality measurement among academic researchers. It has gone through several enhancement stages as a direct reaction to some concerns raised by other researchers with regard to its expectations component [15], the interpretation and operationalization of expectations [16], its reliability and validity [17], and its dimensionality [18]. In response to these questions, the three colleagues have presented counter arguments, clari"cations, and additional evidence to rea$rm the instrument's psychometric soundness and practical value [13,19,20]. The instrument has served as the basis for measurement approaches used in published studies examining service quality in a variety of contexts } e.g., real estate brokers [21]; physicians in private practice [22]; public recreation programs [23]; motor carrier companies [24]; a business school placement center, and a tire shop [25]; an accounting "rm [26]; discount and department stores [16]; a gas and electric utility company [27]; hospi-
tals [25]; banking, pest control, dry cleaning, and fast food [28]; and higher education [29]. For all these reasons, the basic dimensions of SERVQUAL will serve as attributes of service quality and will be used in this paper.
2. Selecting the service quality control instruments 2.1. Weighting the quality measures: The AHP approach The AHP, introduced by Saaty [4], addresses how to determine the relative importance of a set of activities in a multi-criteria decision problem. The process makes it possible to incorporate judgments on intangible qualitative criteria alongside tangible quantitative criteria. The method utilizes pairwise comparisons of alternatives (quality control instruments) as well as pairwise comparisons of the multiple criteria (the "ve attributes of service quality). The use of such pairwise comparisons to collect data from the decision-maker o!ers signi"cant advantages [30]. It allows the decision-maker to focus on the comparison of just two objects, which makes the observation as free as possible from
30
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
extraneous in#uences. Additionally, pairwise comparisons generate meaningful information about the decision problem, improving consistency in the decision-making process, especially if the process involves group decision-making. The `quality control systema model is given in the form of a hierarchy (Fig. 1). The decision problem is structured in a hierarchy of three basic levels: the goal of the decision is at the top, followed by a second level of criteria (or objectives) and a third level of alternatives. This arrangement makes it possible for decision-makers to focus on each and every part of such a complex problem, and to derive priorities from simple pairwise comparisons. In this study, the goal of the AHP model will be to decide on prioritizing the set of quality control instruments given some quality of service criteria. The criteria or objectives of the quality system decision consist of "ve elements: reliability, assurance, responsiveness, empathy, and tangibles. The decision alternatives, the options from which a choice is made, are the quality instruments for customer data collection purposes. Once the model is built, the decision-makers evaluate the elements by making pairwise comparisons. A pairwise comparison is the process of comparing the relative importance, preference, or likelihood of two elements with respect to an element in the level above. When all the comparisons are completed, we calculate the priorities and
a measure of consistency of our judgment. Generally the consistency ratio should be less than 0.10 (10%) [31]. As a result, we derive priorities and an inconsistency ratio with respect to each of the "ve quality-of-service criteria. The next step is to make judgments about the "ve criteria. A comparison is made with respect to each pair [the number of comparisons will be u(u!1)/2, where u is the number of criteria in the model]. The next step will be to synthesize the derived priorities that were based on the decision-makers' judgments. Synthesis means adding up the global weights of the common nodes at the bottom level of the hierarchy so as to generate a composite priority for an alternative across all criteria. Synthesizing shows the results of the entire work, the overall priorities of the alternatives. These alternative priorities will be used later in the combined AHP}GP model as weights in the objective function. The derived priorities with respect to each of the "ve quality-of-service criteria will be used in the combined model to serve as the contribution that each criterion makes to each alternative. 2.2. The goal-programming model In the goal-programming model, the decision variable is x (0 or 1). The objective function, given G by Eq. (1), seeks to minimize deviation from desired targets for limited resources (costs, available
Fig. 1. Structure of the AHP model.
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
management hours, and hours)
available
employee
Min Z"P (d\, d>)#P (d\, d>)#P (d\, d>). (1) A A A @ @ @ C C C The goal constraints in Eqs. (2), (3) and (4) represent the availability of limited resources. The right-hand side of each equation re#ects the targeted or desired level of the resource utilization, where C denotes cost, B available management hours, and E available employee hours. We could also express these limitations of available resources as `system constraintsa by removing the deviation variables from the constraints and the objective function and by changing the equality signs of the constraints to less than or equal signs. For convenience, this will be done through the combined model: K a x #d\!d>"C, (2) AG G A A G K a x #d\!d>"B, (3) @G G @ @ G K a x #d\!d>"E. (4) CG G C C G The ordering of these goal constraints depends on the nature of the problem situation. Either preemptive or non-preemptive goals could be used depending on the order of importance, if any, of the goals. 2.3. The combined model There are several studies that used the AHP methodology in combination with goal programming [30,32}35]. In the combined model, the objective function also includes deviation variables associated with the quality measure goals. It will seek to minimize such deviations from desired levels. The revised objective function is given in Eq. (5). Moreover, a set of constraints, as shown in Eq. (6), will be added to re#ect the quality target of Q in each of the `goal constraintsa. An equation I associated with the AHP weights for the quality control instruments will be added to re#ect the preferences for the di!erent instruments. This is
31
given in Eq. (7):
) Min Z" P (w d\, w d>) #P (d\, d>) I I I I I ? ? ? I #P (d\, d>)#P (d\, d>)#P (d\, d>), A A A @ @ @ C C C (5) K a x #d\!d>"Q (for k"1,2, 2, K), (6) IG G I I I G K w x #d\!d>"1. (7) G G ? ? G Eq. (6) shows that there will be one equation for each of the quality measures. The ordering of goal constraints in Eq. (6) depends on the nature of the problem situation. The classic lexicographic ordering of preemptive priorities may be appropriate if the highest ranked quality measures are preemptively important. To better utilize the results of the AHP method, the a IG will denote the contribution that each criterion makes to each alternative. Compared to previous research, and as shown in Eq. (1), more appropriate weighting of quality measures is desired and used. By utilizing the obtained AHP weightings we have established the ranks of the individual quality control instrument, by establishing a relative weighting w for each I of them. The term P is some k priority I lexicographic rankings where P 'P '2'P . I In this model, Eq. (1) is the objective function, which seeks to minimize deviation from the desired goals consistent with the AHP ranking of the quality control instruments. The lexicographic nature of goal programming treats the AHP weights given by w as a sub-ranking I of the quality control instruments within their speci"c P . The greater the w , the more desirable I I the selection of quality instrument in the decision process. As mentioned before, the constraints for limited resources of total cost, available management hour, and available employee hour could also be expressed as `system constrainta by removing the deviation variables and replacing the equality signs with inequality signs in Eqs. (2)}(4).
32
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
3. Application of the combined model
3.1. Deriving the AHP weights
Based on a real life problem involving a large department store, the proposed AHP}GP methodology is illustrated to show how it is applied. The department store is evaluating seven potential alternatives to better design a quality control system for customer data collection purposes. Unfortunately, each instrument captures the principal quality measures to a di!erent degree and requires varying levels of resources. The problem for a large department store with limited resources is therefore to select the best set of instruments to obtain customer feedback and perception on key quality measures. Our model requires a group process rather than aggregation of individual inputs. In this process, the group establishes a single set of weights for the decision criteria, and then rates the decision alternatives. The group process has several advantages over aggregation of individual ratings [36]. It facilitates a common understanding of the meaning and signi"cance of each criterion. This commonality of understanding is not achieved through aggregating the inputs of individual evaluations. The group is often able to clarify misunderstanding and di!erences in interpretation of the data so that there is a more uniform understanding of the facts. In addition, a group process utilizes the dynamics of powerful in#uence within the decision-making group. To provide expert judgment and consultations in a group process, the department store management formed a team, which consisted of "ve members to work side by side with the research team. The "ve members consisted of the general managers of marketing, "nance and personnel, and the two heads of advertising and public relations. The research was given full support by the top management. To derive the AHP weights, a preliminary data collection stage attempted to get some reaction and feedback with regard to the "ve quality attributes. A survey of known customers was carried out to identify the relative importance of each criterion. The results obtained, along with the management-team's input were used to later structure and synthesize the AHP model.
To facilitate the use of AHP, the problem is decomposed into a multilevel hierarchy showing the overall goal of the decision process, each decision criterion to be used, and the decision alternatives to be considered. The "ve decision criteria are: reliability, empathy, responsiveness, tangibles, and assurance. The decision alternatives to be considered are: large-scale mail survey of known customers, large-scale telephone survey of known customers, open-ended comment cards, personal interviews of customers while shopping, point of purchase survey to measure employee service delivery, point of purchase survey to measure product line service delivery, and small-scale telephone survey of known customers. The multilevel hierarchy is shown in Fig. 2. A relative preference matrix is formulated for each subordinate level of hierarchy. The decisionmaking team was given the chance to express its preferential expert judgment (the pairwise comparisons) in the selection process. Each decision alternative is compared against another decision alternative with respect to one particular decision criterion at a time. These within-pairwise comparisons result in priority values of each decision alternative by each decision criterion. The result of elicitation at this level of the hierarchy is a matrix showing the values expressed by the decision-making team. The `right eigenvector methoda is applied to each of the matrices in turn using Expert Choice Pro software (version 9.0) [37]. From this application, the relative priority for each alternative decision by criterion is computed. The results are shown in Table 2. During the elicitation process, an appropriate level of consistency is necessary to achieve meaningful results. The measure of consistency provided by the AHP is given by Saaty [38]: "( !n)/(n!1), where is the maximum
eigenvalue of the positive reciprocal matrix, and n is the number of objectives. Maximum consistency occurs as approaches zero. Researchers use a consistency ratio of 0.10 or less as guidelines in evaluating consistencies [31]. As can be seen in Table 2, all of the ratios are below the maximum 0.10 level. Via `synthesizinga [31, p. 159] the sets of priority
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
33
Fig. 2. The AHP model for a department store. Table 2 Resulting priority values for each decision alternative by each criterion Decision alternatives
x x x x x x x Total Inconsistency ratio
Service quality criterion decisions Responsiveness
Assurance
Reliability
Empathy
Tangibles
0.031 0.335 0.036 0.099 0.099 0.099 0.302
0.219 0.160 0.145 0.068 0.070 0.130 0.208
0.028 0.397 0.041 0.067 0.071 0.105 0.291
0.026 0.376 0.040 0.165 0.078 0.075 0.240
0.415 0.055 0.097 0.154 0.034 0.036 0.208
1.000 0.050
1.000 0.010
1.000 0.020
1.000 0.050
1.000 0.030
values (from Table 2) are combined into a matrix and multiplied by the overall criteria relative priorities as presented in Table 3. The result is an overall prioritization of the decision alternatives. The results of this step, presented in Table 4, are the
overall rankings (in terms of weights) of the seven quality instrument candidates. Before we discuss the combined model application, and for simpli"cation, Fig. 3 provides an illustration of the combined model development steps.
34
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
Table 3 Resulting priority values for each criterion Criterion decisions
Priority
Responsiveness Assurance Reliability Empathy Tangibles Total
0.479 0.139 0.281 0.025 0.076 1.000
Table 4 Overall AHP weights of the decision alternatives Decision alternatives
AHP weighting
Decision preference
X x x x x x x Total
0.086 0.308 0.057 0.091 0.081 0.099 0.277
Fifth preference First preference Seventh preference Fourth preference Sixth preference Third preference Second preference
1.000 Fig. 3. Illustrative development.
diagram
of
the
combined
model
3.2. The combined AHP}GP application Goal programming permits resource limitations such as budgetary limitations or limited hours of labor, and other selection limitations (such as system constraints) that must be observed in establishing and maintaining a service quality control system. Trying to decide which of the seven service control instruments to use in designing the department store's service quality control system requires the recognition of several resource limitations relative to the nature of each of the instruments. Table 5 provides a summary of yearly cost and labor information on the seven service quality control instruments. The decision team felt that a minimum of three of the seven instruments would be needed to justify the expense of the budget. Also, the nature of some of the quality instruments created duplication that would be wasteful and unnecessary. For example, the minor telephone survey was just a scaled down
version of the large-scale telephone survey. As a result, system constraints would be needed to prevent such duplications. The decision team also wanted to ensure that the selection of quality control instruments would measure those quality attributes that were most important to their customers. To determine the ability of each of the instruments to measure each of the "ve quality measures presented in Table 1, the AHP scores (priority values) for each decision alternative by each criterion will be used. The priority scores are presented in Table 2. We recall that the derived priority scores are the relative contribution that each of the "ve criteria makes to each of the seven alternatives. Since it is our objective to select quality instruments with the highest weights, a goal constraint could be added to maximize the overall weights of
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
35
Table 5 Yearly cost and labor information on service quality control instruments Resource item
Estimated cost Management hours Employee hours
Service quality control instruments (decision alternatives)
Total budgeted yearly resources
x
x
x
x
x
x
x
$9760 186
$9760 186
$1207 54
$640 50
$852 98
$2500 98
$4738 42
$15,000 530
47
47
57
93
278
493
600
1070
the instruments selected. We recall that the AHP solution provided us with overall weights for each of the seven quality instruments (alternatives). These weights are presented in Table 4. 3.3. Formulation of the department store's problem System constraints are needed to reduce resource wastage by not selecting similar quality control instruments. In our case, the minor telephone survey was just a scaled down version of the major telephone survey. The management team also requested that only one form of large-scale survey of known customers be performed, a telephone or mail survey. We should recall also, that at least three instruments are to be selected. Mathematically, these `system constraintsa are treated as simple linear programming constraints. Typically, they create resource boundaries where an interval of possible solutions is usually narrowed down to the optimal choice by subsequently considering the priority of goal constraints: x #x )1 and x #x )1, (8) x #x #x #x #x #x #x *3, (9) 9760x #9760x #1207x #640x #852x #2500x #4738x #d\!d>"15,000. (10) A A The three resource limitation constraints will have deviation variables associated with them, and will attempt to minimize the positive deviations by adding deviation variables to the overall objective function. The department store allocated a maximum yearly operating budget limitation of $15,000 for the cost of materials, postage, phone calls, printing, etc., for all of the quality control instruments
selected. The equation shows that point of purchase surveys to measure employee service delivery and product line service delivery are most expensive with respect to annual cost. Moreover, the management of the department store limited the yearly number of hours that management could devote to the service of quality control instruments to 530 hours, and the number of employee labor hours for data collection, interviews and clerical activities to a maximum of 1070 hours: 186x #186x #54x #50x #98x #98x #42x #d\!d>"530, (11) @ @ 47x #47x #57x #93x #278x #493x #600x #d\!d>"1070. (12) C C The three equations associated with resource limitations could also be formulated as system constraints if desired by decision-makers. However, the management team asked to keep them as goal constraints to better understand their contribution to the overall structure of the model. The management of the department store wanted to also ensure that the selection of the quality control instruments would really measure those quality measures that were important to their customers. One goal constraint is needed to ensure that the instruments with the highest weights obtained from the AHP analysis will be selected. Such goal constraint will attempt to maximize the weights by selecting the quality control instruments with the highest priorities: 0.086x #0.308x #0.057x #0.091x #0.081x #0.099x #0.277x #d\!d>"1. (13) N N
36
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
To determine the ability of the instruments to measure each of the "ve measures presented in Table 1, the AHP priority values (or weights) for each quality control instrument by each quality measure will be used. The individual weights represent the a parameters in the combined model. IG The development of the `goal constraintsa in the situation originates chie#y from the a data in IG Table 2. The right-hand-side values for each goal constraint seek to pressure a selection of service quality control instruments or x with the highest G scores (i.e., the most useful collection of instruments to measure quality). In other words, the best set of three quality instruments was chosen for each of the "ve quality measures. Of course, the right-hand side could be set at any level; however for our case, each of the right-hand-side Q was simply derived, I by summing the best three a , for each of the "ve IG quality measures. The model seeks to select the best or perfect set of quality instruments, for each of the quality measures: 0.031x #0.335x #0.036x #0.099x #0.099x #0.099x #0.302x #d\ !d> "0.736, (14) I I 0.219x #0.160x #0.145x #0.068x #0.070x #0.130x #0.208x #d\ #d> "0.587, (15) I I 0.028x #0.397x #0.041x #0.067x #0.071x #0.105x #0.291x #d\ !d> "0.793, (16) I I 0.026x #0.376x #0.040x #0.165x #0.078x #0.075x #0.240x #d\ !d> "0.784, (17) I I 0.415x #0.055x #0.097x #0.154x #0.034x #0.036x #0.208x #d\ #d> "0.777. (18) I I The objective function will attempt to minimize the overall deviations in each of the goal constraints. The goal constraints include resource limitation constraints as well as desired quality measure goals: Min Z"P (d>#d>#d>)#P d\ A @ C N #P (0.479d\ #0.139d\ #0.281d\ I I I #0.025d\ #0.076d\ ). (19) I I
The ordering of deviation variables in the objective function associated with goal constraints (d }d ) will rely on the data obtained via the I I AHP process and reported in Table 3. We recall that the order of the "ve measures of service quality were responsiveness, reliability, assurance, tangibles, and empathy, respectively. The three measures of responsiveness, reliability, and assurance accounted for 89.9% of the overall weights, while the remaining two measures of empathy and tangibles accounted for only 10.1% of the total weight. The results also re#ect the importance of the top three measures as the quality control instruments that are selected portray these facets of quality. For example, the selection of x (point of purchase survey to measure employee service delivery) re#ects the importance of responsiveness (0.335), reliability (0.397), and assurance (0.160). We also notice that the selection of x which assigns its largest weights to empathy (0.165) and tangibles (0.154) attempts to help in the creation of an overall quality control system that does not neglect the contribution of the two measures of empathy and tangibles. Management was satis"ed with the solution since the selected instruments, in general, tried to re#ect the department store's image to its customers through the "ve measures of service quality. We could simply use the order of importance as reported in the table. Meanwhile, we could make better use of the AHP process results by utilizing the speci"c weights provided in the table. In this case, as mentioned earlier, the AHP weights provide internal relative rankings (or desirability scores) for each of the quality measures. The objective function will seek to minimize the overall deviations. 3.4. Solution of the combined model The solution to this problem was generated using LINDO, Version 6.0 [39]. Table 6 presents the optimal selection of service quality control instruments. The optimal selection turned out in this case study to be where the department store will use point of purchase survey to measure employee service delivery (x ), open-ended comment cards (x ), small-scale telephone survey of known customers
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
37
Table 6 The combined AHP}GP model solution (a) Decision variables X "1 X "1 X "1 X "1 X "0 X "0 X "0
Select point of purchase survey to measure employee service delivery Select open-ended comment cards Select small-scale telephone survey of known customers Select large-scale mail survey of known customers Do not select point of purchase to measure product line service delivery Do not select large-scale telephone survey of known customers Do not select personal interviews of customers while shopping
(b) Slack in resource system constraints Constraints
Usage
Total available
Slack
Budget Management hours Employee hours
$14,107 388 690
$15,000 530 1070
$893 142 380
(c) Deviation in quality measure constraints Constraint
Target
Obtained
Underachievement
Responsiveness Assurance Reliability Empathy Tangibles
0.736 0.587 0.793 0.784 0.777
0.569 0.503 0.610 0.656 0.342
0.167 0.084 0.183 0.128 0.435
(x ), and large-scale mail survey of known cus tomers (x ). The selected instruments will act as a collective system designed to monitor performance on the "ve quality measures de"ned in Table 1. The other three instruments were not chosen because of the limitations posed by the system constraints, goal constraints, as well as the assigned priorities. The slack in the goal constraints associated with resource limitations that were used to model cost and hours usage also provides some summary information on those resources. The amount of deviation actually provides a crude measure of deviation of the resulting quality control system in measuring the "ve quality attributes. The smaller the resulting deviation, the better the system measures the respective quality attributes. The high-quality attributes like responsiveness, reliability, and assurance should be more comprehensively measured under the proposed quality control sys-
tem than low-priority attributes such as empathy and tangibles. The management of the department store was delighted to see that the resulting quality control system would be achievable under-budget (a saving of $893). With regard to the "ve quality measures, we note that the resulting solution adequately addresses the attributes with the highest priorities. For example, with regard to the measures of responsiveness, reliability, and assurance, the three attributes with the highest priorities, they are adequately represented by the selected instruments. Being happy with cost "gures of the model solution, management requested a scenario to be checked where the priorities for the "ve quality measures are placed ahead of the resource limitation goals in the objective function. The scenario called for a solution that overshoots the budget by $11,105, which was totally unacceptable. Other scenarios were also looked at but most of them
38
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
a!ected the budget cost and therefore were not considered further. Due to the importance of the three quality measures of responsiveness, reliability, and assurance as the top three highest priorities, a righthand-side sensitivity analysis was undertaken on these three goals (mainly Eqs. (14)}(16)). Results show that substantial changes to the right-handside goals must occur before altering the existing solution. In other words, the management team that assisted in this research would have had erred in the assessment of the weights assigned to the "ve quality measures. The department store management implemented the suggested results obtained from the combined AHP}GP model in the summer of 1998. After almost one year of its implementation, the top management of the department store has conveyed its complete satisfaction with the results. The implementation of the study was reported several times in the daily-newspapers of the city. Management feels that many customers have noticed increased awareness in service quality, re#ecting to some degree the comprehensiveness of the quality control system to reach the customers. The top management has also promised more money to be available for future implementations of similar projects.
4. Concluding remarks The quality constructs suggested in this paper for designing quality control systems in service organizations were measured via "ve criteria (or service attributes): responsiveness, reliability, assurance, empathy, and tangibles. The combined AHP}GP model presented has helped a large department store select from a set of seven survey instruments those four that will collectively measure the ranked quality measures de"ned by the management team. The AHP is "rst used to prioritize the set of quality control instruments along the "ve measures in a consistent manner. The selected set of four survey instruments collectively represents the department store's service quality control systems. The AHP process was used to derive weights for each of these quality measures using solicitations from both cus-
tomers and management. The derived weights were used to establish preemptive priorities in the objective function, and utility weights for the "ve quality measures in the goal constraints. The proposed combined model has saved resources by recognizing necessary constrained resources. It also presented a more thorough decision process. For example, the completeness and non-bias nature of the pairwise comparisons in the AHP method was a more thorough means of weighting quality control instruments in the decision process relative to ranking methods employed in previous studies dealing with quality control instrument selection. The most obvious advantage of using the AHP model for deriving weights for the quality control instruments as well as measures is that it provides for consistent decision-making. All instruments are evaluated on a single set of weighted criteria (quality control measures). This should help to reduce the subjectivity of the process. In addition, using a mathematical programming method like goal programming allowed for additional trade-o! information that permitted decision-makers to see these tradeo!s in terms of cost and hours. The combined AHP}GP method o!ers a systematic, easy-to-use approach to the service quality control instruments selection decision problem. It extends previous research in the area by incorporating a comprehensive prioritization system within an optimization resource allocation process. The method accurately represents the model's objective function and quality measure goals, while avoiding some pitfalls of goal-programming modeling encountered in previous studies. The management team formally presented the model to top management of the department store. The top management viewed the model's structure and mechanism favorably, believing that it is a tool that can be used in the development of a quality control system and ordered the immediate implementation of its results. Top management has suggested that the research team should continuously re"ne the hierarchy, modify the judgments, and generate new set of weights, since customer taste is dynamically sensitive. In addition, they have suggested that the model should also expand on the resource limitation part to include others such as
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
advertisement costs, employee training costs for better implementation of some of the instruments, and other costs associated with providing customers with promotion gifts to better solicit their participation in data collection. They also recommended that future e!orts should also examine the viability of other forms of quality control instruments such as home visits of some loyal customers, personal interviews with suppliers and personal interviews with service providing sta! members. A revised model should be ready for implementation in the fall of 2000.
References [1] M. Schniederjans, C. Karuppan, Designing a quality control system in a service organization: A goal programming case study, European Journal of Operational Research 81 (1995) 249}258. [2] A. Tversky, D. Kahneman, Judgment under uncertainty: Heuristics and biases, Science 185 (1974) 1124}1131. [3] R. Steuer, An interactive multiple objective linear programming procedure, TIMS Studies on Management Science 6 (1977) 225}239. [4] T. Saaty, Ascaling method for priorities in hierarchical structures, Journal of Mathematical Psychology 15 (1977) 234}281. [5] D. Olson, M. Venkataraman, J. Mote, Atechnique using the analytic hierarchy process in multiple objective planning models, Socio-Economic Planning Science 20 (6) (1986) 361}368. [6] W. Sasser, P. Olsen, D. Wycko!, Management of Service Operations: Text and Cases, Allyn and Bacon, Boston, MA, 1978. [7] C. Gronroos, Strategic management and marketing in the service sector, Swedish School of Economics and Business Administration, Helsinki, Finland, 1982. [8] U. Lehtinen, J. Lehtinen, Service quality: A study of quality dimensions, Unpublished Research Report, Service Management Group OY, Finland, 1982. [9] D. Garvin, What does `product qualitya really mean? Sloan Management Review 26 (1) (1984) 25}43. [10] D. Garvin, Competing on the eight dimensions of quality, Harvard Business Review 65 (5) (1987) 101}109. [11] G. DeSouza, Now service business must manage quality, Journal of Business Strategy 10 (3) (1989) 21}25. [12] A. Parasuraman, V. Zeithaml, L. Berry, A conceptual model of service quality and its implications for future research, Journal of Marketing 49 (4) (1985) 41}50. [13] A. Parasuraman, V. Zeithaml, L. Berry, Reassessment of expectations as a comparison standard in measuring service quality: Implications for further research, Journal of Marketing 58 (1) (1994) 111}124.
39
[14] A. Parasuraman, V. Zeithaml, L. Berry, SEVQUAL: a multiple-item scale for measuring consumer perceptions of service quality, Journal of Retailing 64 (1) (1988) 12}40. [15] S. Taylor, J. Nicholson, J. Milan, R. Martinez, Assessing the roles of service quality and customer satisfaction in the formation of the purchase intentions of Mexican consumers, Journal of Marketing Theory and Practice 5 (1) (1997) 78}90. [16] R. Teas, Expectations, performance evaluation and consumer's perceptions of quality, Journal of Marketing 57 (4) (1993) 18}34. [17] T. Brown, G. Churchill, P. Peter, Improving the measure of service quality, Journal of Retailing 69 (1) (1993) 127}139. [18] D. Finn, C. Lamb, An evaluation of the SERVQUAL scales in retail setting, in: R.H. Holman, M.R. Solomon (Eds.), Advances in Consumer Research, Vol. 18, Association for Consumer Research, Provo, UT, 1991. [19] A. Parasuraman, L. Berry, V. Zeithaml, Re"nement and reassessment of the SERVQUAL scale, Journal of Retailing 67 (1991) 420}450. [20] A. Parasuraman, L. Berry, V. Zeithaml, More on improving service quality measurement, Journal of Retailing 69 (1993) 40}147. [21] L. Johnson, M. Dotson, B. Dunlop, Service quality determinants and e!ectiveness in the real estate brokerage industry, The Journal of Real Estate Research 3 (1988) 21}36. [22] S. Brown, T. Swartz, Agap analysis of professional service quality, Journal of Marketing 53 (1989) 92}98. [23] J. Crompton, K. Mackay, User's perceptions of the relative importance of service quality dimensions in selected public recreation programs, Leisure Sciences 11 (1989) 367}375. [24] R. Brensinger, D. Lambert, Can the SERVQUAL scale be generalized to business-to-business services? Knowledge Development in Marketing, 1990 AMA's Summer Educators' Conference Proceedings, 1990. [25] J. Carman, Consumer perceptions of service quality: An assessment of the SERVQUAL dimensions, Journal of Marketing Research 66 (1) (1990) 33}55. [26] D. Bojanic, Quality measurement in professional service "rms, Journal of Professional Services Marketing 7 (2) (1991) 27}36. [27] E. Babakus, G. Boller, An empirical assessment of the SERVQUAL scale, Journal of Business Research 24 (1992) 253}268. [28] J. Cronin, S. Taylor, Measuring service quality: A reexamination and extension, Journal of Marketing 56 (1992) 55}68. [29] W. Boulding, A. Kalra, R. Staelin, V. Zeithaml, A dynamic process model of service quality: From expectations to behavioral intentions, Journal of Marketing Research 30 (1993) 7}27. [30] M. Schniederjans, R. Wilson, Using the analytic hierarchy process and goal programming for the information system project selection, Information and Management 20 (1991) 333}342.
40
M.A. Badri / Int. J. Production Economics 72 (2001) 27}40
[31] T. Saaty, Decision Making for Leaders, RWS Publications, Pittsburgh, PA, 1995. [32] N. Bryson, A goal-programming method for generating priority vectors, Journal of the Operational Research Society 46 (5) (1995) 641}648. [33] D. Despotis, Fractional minimax goal-programming: A uni"ed approach to priority estimation and preference analysis in MCDM, Journal of the Operational Research Society 47 (8) (1996) 989}999. [34] S. Myint, M. Tabucanon, Amultiple criteria approach to machine selection for #exible manufacturing systems, International Journal of Production Economics 33 (1}3) (1994) 121}131. [35] M. Schniederjans, T. Garvin, Using the analytic hierarchy process and multi-objective programming for the selection
[36]
[37] [38] [39]
of cost drivers in activity-based costing, European Journal of Operational Research 100 (1997) 72}80. E. Erkut, S. Moran, Locating obnoxious facilities in the public sector: An application of the analytic hierarchy process to the municipal land"ll sitting decisions, Socio-Economic Planning Science 25 (2) (1991) 89}102. Expert Choice, Decision Support Software, Pittsburgh, PA, 1995. T. Saaty, The Analytic Hierarchy Process, McGraw-Hill, New York, 1980. L. Schrage, Optimization Modeling with LINDO, 5th Edition, Brooks/Cole Publishing Co., Paci"c Grove, CA, 1997.