Group Decis Negot DOI 10.1007/s10726-015-9464-4
Developing a Group Decision Support System for Advertising Media Evaluation: A Case in the Middle East Parham Fami Tafreshi1 · Mohammad Hasan Aghdaie1 · Majid Behzadian2 · Mahdieh Ghani Abadi3
© Springer Science+Business Media Dordrecht 2015
Abstract The importance of advertising media evaluation as a multifaceted problem is well known by both academics and practitioners. Although previous studies tried to optimize media evaluation, there still are some gaps and problems to address, particularly in areas of flexibility of models/frameworks, decision making quality, tension management, and agility of the evaluation process. Most of previous studies are based on inflexible models/frameworks that have limitations on number of criteria/alternatives they can consider and type of data they can process. A great volume of the work used arbitrary decision making; arbitrary decision making regarding criteria and media importance may reduce effectiveness of advertising campaigns. Furthermore, the academic literature offers little guidance on group decision aggregation, and tension management during decision making is neglected. Media evaluation is a time taking process and any acceleration will reduce pre-campaign costs. The main aim of this paper is to illustrate how a group decision support system (GDSS) can assist media planners to overcome mentioned problems more systematically. For this purpose, we developed a GDSS that is an integration of three well-known multi-criteria decision making techniques. With a real world case study, we illustrate the performance of the proposed GDSS. Results of our quantitative assessments indicate that the GDSS is flexible, allows decision makers to express their opinions, reduces tension among decision makers, and saves time.
B
Parham Fami Tafreshi
[email protected]
1
Marketing and Sales, Solico Group, 433 N Camden Drive, Beverly Hills, CA 90210, USA
2
Department of Industrial Engineering, Mehr Alborz University, 15 Second St., Gisha Ave., Tehran, Iran
3
Department of Management, Faculty of Management, Adib Mazandaran University, Khazar Rd., Sari, Iran
123
P. F. Tafreshi et al.
Keywords GDSS · Advertising media evaluation · MCDM · ANP · VIKOR · DEMATEL
1 Introduction Advertising media evaluation is of paramount importance as one of the most significant problems facing media planners is deciding in which medium to place their advertisements. When millions of dollars are at stake, managers want a reliable procedure to achieve marketing objectives. Unfortunately, conventional knowledge and experience is remained as common used criterion in advertising media evaluation (Hackley 2005). Arbitrary media evaluation, which heavily draw on experience and knowledge of individuals, may prevent firms to reach maximum effectiveness of their advertising campaigns. Poor performance in advertising places financial burden on the firm and may threaten its survival. Media evaluation is a very complex and multifaceted decision making problem. The growing number of advertising media (e.g. TV, magazine, radio, the Internet) and vehicles (for the Internet e.g. Facebook, YouTube, Yahoo) perplexes media planners. In addition to numerous media and vehicles, the problem encompasses a considerable number of criteria, and this makes the evaluation process more complex. As different expertise are required due to diverse aspects of the problem, a group of experts participates in the media evaluation. Therefore, another element of complexity comes from the nature of group decision making when several experts, sometimes with conflict of interest, have to reach to a mutual and satisfying solution. Group decision meetings are usually unproductive in terms of efficiently utilizing decision makers’ time, and tension among decision makers prevents them to efficiently achieve objectives (Adla et al. 2011). Furthermore, tension among decision makers slows down media evaluation progress, causes more pre-campaign costs, and may deteriorate effectiveness of the advertising campaign. Media evaluation is a large-scale decision problem as it includes numerous criteria and many alternatives. One should consider both criteria and alternatives to make judgment and decide where to advertise. For instance, assume that we have five criteria to consider and five media to advertise; this makes one to consider 5×5 = 25 things in his/her mind while judging. According to psychological findings, generally humans, such as chess players, can cope with information containing only a few facts, seven plus or minus two (Saaty and Ozdemir 2003). With more facts to consider, people become confused, and this weaken their ability to process the information. Large-scale decision problems have often been based on multi-criteria decision making (MCDM). Since the 1990s, MCDM has been adopted in many studies in various fields and its advantages make it one of the most suitable approaches to be applied whenever the decision making problem is complex (Zlofani et al. 2013). Complex decision making problems are often consisted of a myriad of elements such as criteria and alternatives. MCDM concentrates chiefly on distinguishing the criteria in order to evaluate the alternatives. It can deal with qualitative or quantitative, continuous, Boolean, probabilistic or deterministic criteria (Milani et al. 2013). Moreover,
123
Developing a Group Decision Support System...
MCDM includes methods and approaches that can evaluate many alternatives at the same time. Decision makers and managers can benefit from an MCDM based GDSS to overcome existing gaps in media evaluation process as a group decision problem. A GDSS with a systematic procedure can reduce role of arbitrary decisions by making pairwise comparisons, manage tension among decision makers using rational aggregation of group decisions, and accelerate the evaluation process. Researchers have suggested various types of solutions to optimize advertising media evaluation/selection. Unfortunately, preceding studies neglected the fact that media evaluation/selection is a group task that involves decision makers form different divisions and usually with conflicts of interest. Therefore, the opportunity to address some gaps and problems, which are related to group decision making and negotiation, such as aggregation of judgments, tension management, and group decision acceleration is missed. The main purpose of this study is to represent how a model-driven GDSS can assist media planners and managers to address mentioned problems and gaps. The proposed GDSS of this study is comprised of three MCDM techniques. First, decision making trial and evaluation laboratory (DEMATEL) method is applied, which was developed in Geneva Research Center (Gabus and Fontela 1973; Fontela and Gabus 1976). We used this technique to reveal relationships, dependence, and feedbacks among criteria. It transforms intangible linkages to weighted tangible linkages. Then, analytic network process (ANP) is used to measure relative importance of the criteria. It measures relative weight of each criterion with respect to its relations, given influence and taken effect, with other factors (Saaty 2013). Third, to prioritize advertising vehicles VIKOR method, which is an abbreviation for Serbian VIseKriterijumska Optimizacija I Kompromisno Resenje, is applied (Opricovic 1998). The VIKOR method concentrates on ranking the given set of alternatives based on a specific measure of “closeness to the idea/aspired level” in the presence of conflicting criteria (You et al. 2015). The reminder of the article is organized as follows. In Sect. 2, relative studies are reviewed. Then the methodology of the GDSS is elaborated in Sect. 3. Section 4 with a real world case study indicates performance of the GDSS. Conclusions and future remarks are presented in Sect. 5.
2 Previous Studies 2.1 Decision Support Systems During the last five decades, decision support systems realm changed drastically. In 60s, academics began to work on quantitative models computerization for decision making assistance (Raymond 1966; Turban 1967; Holt and Huber 1969). In 1967, Scott Morton built, implemented, and tested a model-driven DSS in his field research, which became a milestone in DSS literature. In 1969, with the help of a computer-aided decision system Ferguson and Jones conducted an experimental study on production scheduling. Gorry and Morton (1971) coined the term “decision support system” as systems that assist decision makers in semi-structured and unstructured decision prob-
123
P. F. Tafreshi et al.
lems. In the 1970s, Little suggested four criteria (robustness, ease of control, simplicity, and completeness of relevant detail) for evaluation of DSSs. Later, Alter (1975, 1977) suggested a thinking framework for both management and business DSSs. Based on a field research on 56 DSSs, Alter (1980) suggest seven categories: (a) file drawer systems, (b) data analysis systems, (c) accounting and financial systems, (d) representational systems, (e) optimization systems, and (f) suggestion systems. Robert Bonczek et al. (1981) introduced a framework to understand four major design aspects that affect all DSSs: (a) language system, (b) presentation system, (c) knowledge system, (d) problem processing system. In 1982, Sprague and Carlson defned DSSs as “a class of information system that draws on transaction processing systems and interacts with the other parts of the overall information system to support the decisionmaking activities of managers and other knowledge workers in organizations”. Based on type of assistance a decision support system can provide, Power (2002) proposed several domains for classifying decision support systems: (a) model-driven, (b) knowledge-driven, (c) communication-driven, (d) data-driven, document-driven A model-driven DSS emphasizes on data manipulation and analyzing using mathematical and logical methods for optimization, simulation, etc. (Zhang et al. 2015; Lei and Moon 2015). A knowledge-driven DSS emphasizes on solving a decision making problem using facts, cased-based reasoning, rules, procedures, and similar structures (Koo et al. 2014). A communication-driven DSS facilitates working on a share task by allowing sending and receiving data among a group of decision makers (Bose 2015). A data-driven DSS assist decision makers by providing access to data and sometimes manipulation option (Sharma et al. 2015). Most of model-driven DSSs are one individual user only; on the other hand, data-driven DSSs are used by multiple users across organizations. A document-driven DSS concentrates on managing, retrivising, and manipulation of data in various electronic formats (Baumeister and Striffler 2015). The literature indicates that most of applications focused on data manipulation using quantitative models, big data analysis, and group decision making supporting (Alavi and Joachimsthaler 1992; Eom 2002; Arnott and Pervan 2005). The latest momentous expansion is in pervasive (anywhere-anytime) DSSs, which are embedded in or connected to various electronic tools and gadgets (Yea et al. 2015). Pervasive decision support systems are coupled with opportunities; higher efficiency, greater agility, more accuracy, more effectiveness, are of opportunities that pervasive decision supporting topic may bring decision makers.
2.2 Group Decision Support Systems Group decision making processes share a set of characteristics that caused group decision support systems formation; communication barriers, group consensus, process productivity and efficiency, information exchange, content management, and so on are of characteristics for almost any group decision making process. A GDSS is an interactive, computer based/cooperative work system that assist decision makers to find the best possible solution for a complex decision making problem using decision methods and technologies (Behzadian et al. 2013). GDSSs have been researched deeply, as they affect important tasks in organizations. GDSSs were applied for identifying/analyzing
123
Developing a Group Decision Support System...
a problem, group decision making, various type of planning, conflict management, and enhancing meeting/negotiation productivity (DeSanctis et al. 2008). As the nature of decision problems are getting more complex, scientists introduced and studied GDSS to face these complex decision problems. GDSSs have proven their ability to increase decision makers’ participation and quality of decision making (Adla et al. 2011). Previous studies on effectiveness of GDSSs shown a wide range of inconsistent results. Gallupe et al. (1988) indicated that GDSS helped groups in generating more ideas than unsupported groups. Watson et al. (1988) stated that group members equipped with GDSS understand each other’s notion and point of view better. Throughout the history of GDSSs scientists and practitioners have developed various models and frameworks to identify and analyze nature of decision making problems, to improve efficiency and effectiveness of decision making process, and to manage conflicts in negotiation and decisive meeting (Ishizaka and Nemery 2013). Based on purpose of usage, GDSSs are of two main stems: • Systems that deal with human communication aspects such as structuring, storing, facilitating, and distributing. • Systems that are developed to process and then aggregate different perceptions and views of members of a decision making group. Most often, these type of GDSSs are integration of several techniques to model decision problem (Ackermann and Eden 2001). A number of web-based share point systems, cooperative work computer-supported systems, and computer-mediated communication have been developed to facilitate human communication and negotiation (Limayem et al. 2006; Penichet et al. 2009; Fan and Shen 2011; Duque et al. 2012; Efremov and Av 2014). GDSSs of which the second stem is formed are widely studied and applied to process and aggregate experts’ views and perceptions in complex decision making problems (Matsatsinis et al. 2005; Rios and Rios-Insua 2008; Behzadian et al. 2013; Tavanaa et al. 2014). 2.3 Advertising Media Optimization History of the topic go back to the late 1950s; Riorden did optimization studies on media planning (1958). Lee and Burkart (1960), Day (1962), Nerlove and Arrow (1962), Engel and Warshaw (1964), Stasch (1965), Brown and Warshaw (1965), Bass and Lonsdale (1966), and De Kluyver (1978) developed Linear programing models for advertising media selection/planning. Dynamic programing was adopted by Little and Lodish (1966); they approached media selection problem with a discrete time stochastic model of multiple media selection. Another operation research method that has been widely applied in advertising media optimization is goal programing (Charnes et al. 1968; Lee 1973; Jha et al. 2011). Keown and Duncan (1979) apply an integer goal programing model to solve this particular problem. Zufryden (1973) approached media scheduling in a probabilistic frame. Fruchter and Kalish (1998) has presented a differential game model for media budgeting and allocation. Lee and Kwak (1999) have developed an analytic hierarchy process based goal programing model. Mesak and Zhang (2001) have presented an optimal advertising pulsation policies using dynamic programing method. Later, Ching et al. (2006) based their
123
P. F. Tafreshi et al.
research on Mesak and Zhang study and reached to an advertising model that could capture the advertising wear out phenomenon using the linear programing. To find the best possible combination of placements of commercial (which channel, what time, how often) with the aim of highest rating subjected to limited advertisement budgets, Mihiotis and Tsakiris (2004) developed an integer programing model. With presenting a case study, Kwak et al. (2005) developed a mixed integer goal programing model for helping experts to make a strategic decision about selecting advertising media and to facilitate the advertising media selection process in a dual (industrial/consumer) market. Kumar et al. (2006) schedule the advertising on web pages to maximize the revenue have used hybrid genetic algorithm. Bhattacharya (2009) provides us with a chance constraints goal programing model for the advertising planning problem. His model seeks for appropriate number of advertisement in different advertising media and optimal allocation of the budget assigned to each medium. Locander et al. (1978) presented a media allocation using a nonlinear benefit curves model. Farzipoor-Saen (2011) applied an imprecise data envelope analysis for media selection in the presence of flexible factors and imprecise data. Danaher et al. (2010) optimized media selection using multivariate negative binomial distribution. Azadi et al. (2013) developed an imprecise neutral data envelopment analysis model for advertising media selection. Existing literature indicates that only few papers deal with advertising media evaluation/optimization. Most of the studies used MCDM approach to address media evaluation, selection, and planning issues. Most of the preceding studies applied operation research techniques and approaches, in particular goal programing, dynamic programing, and linear programing, to optimize advertising media evaluation. Mentioned methods and approaches have several drawbacks to name. They develop single-use mathematical models; once the model is developed, changing any value (of variables) or adding/removing any variable/constraint need advanced mathematical knowledge. Since some constraints of the model might go false, and the model must be checked for feasibility. As a result, adoption of developed models in similar problems take decision makers a lot of time and cause waste of money. Preceding studies used methods and approaches that could only consider one decision maker’s notions despite the real business environment in which several experts work together to evaluate advertising media and group decision making and judgment aggregation is neglected. As preceding studies ignored group decision making and negotiation, the opportunity to address other aspects of media evaluation such as tension management and process acceleration is missed.
3 The Proposed GDSS In this section, we concentrate on the proposed GDSS and its integrated conceptual framework. The media evaluation process is structured in four phases (Fig. 1). In the first phase, decision-making team is formed; criteria for the evaluation problem are selected. The next three phases are building blocks of the GDSS. In the second phase, decision makers with the use of the DEMATEL method reveal and illustrate relations and linkages among the criteria. In the next phase, decision makers make pairwise comparisons to obtain importance of each criterion. The GDSS combines outputs of
123
Developing a Group Decision Support System...
Fig. 1 Schematic representation of the proposed GDSS
the DEMATEL and the ANP to obtain final weights of the criteria. In the last phase, with respect to the profile of alternatives, the GDSS ranks all the alternatives using the VIKOR method. The propose GDSS is flexible so that users can add/delete any criterion or alternative, or edit values without any interruption in the evaluation process. Therefore, it can be used as a dynamic (online) advertising vehicle evaluation system. Online functionality is of considerable importance for any media evaluation system, as today’s information providers companies are collecting and combining multiplatform performance data and customer purchases online. The GDSS handles pre-campaign planning, in flight optimization, and post-campaign evaluation. Using proper data, such as time series, the GDSS can optimize pre-campaign planning. Its online functionality allows in-flight optimization. The GDSS encompasses three stages: (1) impact-relations assessment among criteria, (2) criteria weights calculation, and (3) ranking alternatives.
123
P. F. Tafreshi et al.
3.1 Architecture of the GDSS Software are designed because of their benefits. Those benefits must be provided by software functions, and needed functions are translated into software features. Features are programed based on stakeholders’ concerns and the required quality. In our case, stakeholders includes decision makers, advertising manager, marketing manager, sales analysts, media planers, and financial coordinator. The stakeholders have different concerns that affect usability, reliability, availability, and the security of the GDSS. With a layered approach, we indicated three separate, yet interconnected layers of the proposed GDSS (Fig. 2): • User interface layer bridges users and model of the GDSS and provides the decision makers with graphical representation of the outputs. • Model layer includes integrated mathematical modules for implementation of the proposed GDSS. • Data layer stores all necessary data for advertising media evaluation.
Fig. 2 Architecture of the proposed GDSS
123
Developing a Group Decision Support System...
3.2 Implementation Decision makers attend a meeting, in one place and the same time, in which there is a LAN-enabled computer for each individual and a facilitator to manage the meeting. Our GDSS architecture uses non-anonymous communication; therefore, physical setting of chairs and monitors allows decision makers to see one another and to speak with facilitator during the meeting. Decision makers can interact during the meeting; therefore, they can assess for what they stand and where they stand. In-depth understanding of other decision makers’ preferences and their reasons for those will improve group consensus (Bose 2015). First, facilitator asks decision makers to fulfill a square matrix using a 0–4 integer scale to reveal intangible relations and transform them into tangible linkages among criteria and to obtain magnitude of impact between any two of them. As all the decision makers finished fulfilling and editing the matrix, the facilitator manages a session in which decision makers debate and explain their reasons behind their judgments. Decision makers can edit entered data during the session. Facilitator decides when to finish the session. Then facilitator obtains aggregated judgments that indicates magnitude of interactions among criteria. She also uses the GDSS to draw an impact-relations map (IRM) that indicates noticeable given and received influences among criteria. After a break, decision makers are asked to fulfill a squared matrix using Saaty’s fundamental scale. A fulfilled matrix of each decision maker indicates importance of all criteria. Again, facilitator manages a session in which decision makers debate and explain their reasons for their judgments and edit their entered data if necessary. When the facilitator finishes the session, she obtains aggregation impacts of linkages among criteria and overall criteria weights. Then using VIKOR method, the GDSS obtains ranks of all advertising vehicles as alternatives of the problems. At the end, facilitator asks decision makers to answer a questionnaire about the evaluation process, methods, and the GDSS. In the following, we focus on mathematical modelling of the GDSS, as it is a modelbased one. Integrated model of the GDSS is of great importance, for it must be reliable and effectively improve efficiency of the group decision making process.
3.3 The DEMATEL Method Media evaluation is consisted of criteria; these criteria interact with each other and make mutual directly or indirectly linkages. The DEMATEL method reveals dependences or interdependences of relationships amongst criteria. This method exposes cause and effect relationships among criteria and aims to transform the intangible relationships among criteria into tangible and intelligible relationships (Hashemian et al. 2014). We describe the DEMATEL method in four steps, as follows (for more mathematical details see Tzeng et al. 2007): Step 1: The average matrix Each decision maker is asked to measure influence of criterion i on criterion j with five digits 0, 1, 2, 3, and 4, representing “no influence”, “low impact”, “medium influence”, “high impact”, and ‘very high influence’, respectively. Thus, each expert
123
P. F. Tafreshi et al.
forms a n × n non-negative integer matrix as X k = [xikj ], where k is the number of a respondent and 1 ≤ k ≤ H; note that n is the number of criteria and H is number of decisoim makers. To incorporate all perceptions the average matrix A can be computed using Eq. 1: 1 k xi j H K =1 ⎡ ⎤ a11 · · · a1 j · · · a1n ⎢ .. .. ⎥ .. ⎢. . ⎥ . ⎢ ⎥ ⎢ A = ⎢ ai1 · · · ai j · · · ain ⎥ ⎥ ⎢ .. .. .. ⎥ ⎣. . . ⎦ an1 · · · an j · · · ann H
ai j =
(1)
(2)
Step 2: The normalized initial direct-relation matrix The Normalized initial direct-relation matrix D (i.e. D = [di j ]n×n ) can be obtained by normalizing the average matrix A. We obtain matrix D through Eqs. 3 and 4. ⎫ ⎧ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎬ ⎨ 1 1 (3) S = min , n n
⎪ ⎪ ⎪ ⎪ ⎪ ⎪ max a max a ij ij ⎭ ⎩ 1≤i≤n 1≤ j≤n j=1
i=1
D = A×S
(4)
Each value in matrix D falls between zero and one. Step 3: The total relation matrix The total relation matrix T is defined as: T = D(I − D)−1
(5)
where I is the identity matrix, and T = [ti j ]n×n for i, j = 1, 2, 3, . . . , n. The DEMATEL method describes each row sum, indicated in Eq. 6, and column sum, which is shown in Eq. 7, of matrix T. ⎡ ⎤ n ti j ⎦ (6) r = (ri )n×1 = ⎣ j=1
n×1
c = (c j )n×1 = (c j )1×n =
n i=1
ti j
(7) 1×n
For total relation matrix T, r and c will be n × 1 and 1 × n vectors representing sum of arrow and sum of columns. Therefore, ri is the sum of the ith row in matrix T, so both
123
Developing a Group Decision Support System...
indirect effects and direct effects of i factor summarize in ri . c j is the jth column sum in matrix T, hence ci represents direct and indirect effects received by factor j from the other factors. When j = i, the sum (ci + ri ) indicates both direct effects and indirect effects given and taken by factor i. The difference (ri − c j ) shows the net effect of factor i in the system. When this difference is positive, it means factor i is a net cause, and it is a net receiver when the difference (ri − c j ) is negative (Altuntasa and Dereli 2015). Step 4: Setting up a threshold value to obtain the IRM Matrix T provides us with effects of each criterion on the others. In other words, each factor ti j of matrix T provides network information on how factor i affects factor j. In practice, if all the relation values of matrix T move to the IRM, they will form a cluttered map. This map would be too complex to reveal the necessary network information. Therefore, we can set up a threshold value α to filter minor impacts of the factors in matrix T. In order to increase simplicity of the IRM, only values that are greater than the threshold value α would be shown in the IRM. Consequently, relation values smaller than the threshold value α have to be reset to zero. Generated matrix with α-cut is called the α-cut total relation matrix Tα . The IRM can be drawn by coordinating the dataset of (r + c, r − c). 3.4 The ANP Method ANP measures importance of the criteria with respect to feedbacks, direct linkages and indirect relationships within or among them (for more details see Tsai et al. 2013). To obtain relative weights of the criteria, each decision maker pairwisely compares any two criteria. For this purpose, we form three types of matrices: unweighted supermatrix, weighted supermatrix, and limited matrix. The Formations of these three supermatrices are elaborated in the next three steps. Step 5: Making pairwise comparisons to form the unweighted supermatrix. The ANP method uses Chains of pairwise comparisons to obtain weights of criteria with considering relationships and feedback within or among them. This step is done through pairwise comparisons by asking “How much importance does a criterion have compared to another criterion with respect to your interests or preferences?” The relative importance value can be determine based on Saaty 1–9 scale (see Table 1). Where wi j represents the importance of ith element compared with the jth element, w ji is equal to w1i j . In addition, in the pairwise matrix W, wii is equal to one. To calculate the local priority vectors of the pairwise comparison matrix, the eigenvector approach is applied. The general form of supermatrix is illustrated as follows:
123
P. F. Tafreshi et al. Table 1 The fundamental scale of absolute number Intensity of importance
Definition
Explanation
1
Equal importance
Two criteria contribute equally to the objective
2
Weak importance
Experiment and judgment slightly favor one criterion over another
3
Moderate importance
4
Moderate plus
5
Strong importance
6
Strong plus
7
Very strong or demonstrated importance
8
Very, very strong
9
Extreme importance
Reciprocals of above
If criterion i has one of the above nonzero numbers assign to it when compared with criterion j, then j has the reciprocal value when compared with i
Experiment and judgment strongly favor one criterion over another A criterion is favored very strongly over another; its dominance demonstrated in practice
C1 C2 .. Wi = . .. . Cn
The evidence favoring one criterion over another is of the highest possible order of affirmation
C1 ⎡w
11
⎢w21 ⎢ ⎢ ⎢ ⎢ . ⎣ . . wn1
C2 w12 w22 .. . wn2
... ... ... .. . ...
A reasonable assumption
Cn w1n ⎤ w2n ⎥ ⎥ ⎥ ⎥ .. ⎥ ⎦ . wnn
where Wi denotes ith decision maker’s judgments. i = 1, 2, 3, . . . , k Then the total pairwise comparison matrix W T is formed based on Eq. 8.
WiTj =
k
k1 Wi j
(8)
i=1
Step 6: Weighted supermatrix construction In Eq. 9, the normalized matrix T s and unweighted supermatrix WT are used to obtain the weighted supermatrix W.
123
Developing a Group Decision Support System...
⎡
s × wt t11 11 ⎢ .. ⎢. ⎢ s t W =⎢ ⎢ t1 j × w 1 j ⎢. ⎣ ..
··· ···
s × wt tn1 n1
t1s j × w1t j .. . tisj × wit j .. . tns j × wnt j
··· ···
s × wt ⎤ t1n 1n .. ⎥ ⎥ . ⎥ s t tin × win ⎥ ⎥ ⎥ .. ⎦ . s × wt tnn nn
(9)
where, Ts represents the normalized α-cut total relation matrix based on Eq. 10. The α-cut relation matrix T α is normalized by dividing the values in row i by di =
n total α. t j=1 i j
⎡
α /d t11 1 ··· ⎢ .. ⎢. ⎢ α Ts = ⎢ ⎢ ti1 /di · · · ⎢. ⎣ .. α /d tn1 n
t1αj /d1 · · · .. . tiαj /di · · · .. . · · · tnαj /dn · · ·
⎡ s α /d ⎤ t11 · · · t1n 1 .. ⎥ ⎢ .. ⎥ ⎢. . ⎥ ⎢ s α ⎢ tin /di ⎥ ⎥ = ⎢ ti1 · · · ⎢ ⎥ .. ⎦ ⎣ ... . α /d tnn n
s tn1
t1s j · · · .. . tisj · · · .. . · · · tns j · · ·
s ⎤ t1n .. ⎥ . ⎥ ⎥ s ⎥ tin ⎥ .. ⎥ . ⎦
(10)
s tnn
After calculating weighted supermatrix W, we normalize it to derive the weighted supermatrix.
n The weighted supermatrix is derived by dividing the values in column j by d j = i=1 wi j . Then, each column sums exactly to unity and ensures us that sum of values in any column equals 1, which is need to obtain the limited supermatrix in next step. In traditional ANP, normalization is done based on the idea that criteria of a model have equal effect on each other. However, effects of different criteria on each other are not equal in a real world decision-making problem. Therefore, the assumption of equal effect and influence for all criteria is not a reasonable one to obtain the weighted supermatrix, as it is in traditional ANP. As stated, the DEMATEL method is applied to handle this wrong assumption by influential weights. Step 7: Calculation of the overall priorities with the limited supermatrix Criteria weights are obtained by multiplying the supermatrix W by itself in g = 2k + 1 power where k ∈ {0, 1, 2, . . . } and until the rows are stabilized. The limited supermatrix represents global priority-influential vectors, also called the ANP weights. W ∗ = lim (W )g g→∞
(11)
123
P. F. Tafreshi et al.
3.5 VIKOR The VIKOR method helps decision makers to rank a set of alternatives by considering performance scores of the criteria. This method is based on the concept of compromised programing of MCDM by comparing the measure of “closeness” to the “ideal” value (for mathematical details see Opricovic 1998). The VIKOR is described in four steps. Step 8: L p -metrics Let k = 1, 2, 3, . . . , m and A1 , A2 , A3 , . . . , Am represent the m alternatives facing a decision maker; Let j = 1, 2, 3, . . . , n, when n represents number of the criteria. Then f k j denotes the performance score for the alternative Ak in presence of jth criterion. Here, the weight of jth criterion, w j , is calculated by using integration of the DEMATEL and the ANP methods as described above. The VIKOR method uses the following L p -metrics: ⎧ ⎫1/ p n ⎨ p⎬ p Lk = w j (| f j∗ − f k j |)/(| f j∗ − f j− |) (12) ⎩ ⎭ j=1
where 1 ≤ p ≤ ∞; k = 1, 2, . . . , m. Note that f* is the most desired value among values for a criterion. For instance, if the criterion is number of advertising exposure per campaign then f* is the greatest number, as the greater is the better. On the other hand, when less is better, f* is smallest value; therefore each criterion has one f*. Step 9: The aspired level and the tolerable level The highest performance score (aspired level) with respect to the jth criterion among all alternatives that represents the positive ideal point of that criterion is defined as f j+ = maxk f k j . Similarly, the negative ideal point (tolerable level) of the jth criterion which reflects the lowest performance score with respect to the jth criterion between all alternatives is represented as f j− = mink f k j . Step 10: Mean group utility and maximal regret p=1
p=∞
(as Rk ) to formulate the ranking measure. This method uses Lk (as Sk ) and L k Where, Sk denotes the rations of distance to the aspired level (positive-ideal) based on Eq. 13. In addition, it provides the synthesized gap of all criteria from the DEMATEL and ANP methods results. n p=1 w j ( f j+ − f k j )/( f j+ − f j− ) (13) Sk = L k = j=1
The VIKOR method uses rk j = ( f j+ − f k j )/( f j+ − f j− ) to represent the normalized gap of k alternative in j criterion. Rk represents the maximal gap in j criterion of k alternative for improvement priority. Based on Eq. 14, this value can be calculated as follows: p=∞ Rk = L k f k+ − f j− j = 1, 2, 3, . . . , n (14) = max w j f j+ − f k j j
123
Developing a Group Decision Support System...
Step 11: The comprehensive indicator Q k In this step, we calculate comprehensive indicator Q k to rank model’s alternatives with respect to the chosen criteria based on Eq. 15. Rk − R + Sk − S + + (1 − v) (15) Qk = v − S − S+ R− − R+ where S + = mink S K , S − = maxk Sk , Q + = mink Q k , Q − = maxk Q k , and v as the weight of gap falls between 0 and 1 represents the weight on the strategy of maximum group utility (average gap in scale normalization). 1−ν indicates the weight on individual regret (maximum gap in special criterion for priority improvement). The VIKOR ranks the alternatives by sorting the values of Sk , Rk , and Q k , for k = 1, 2, . . . , m, in a decreasing order. Opricovic propose as a compromise the alternative A1 which is ranked first by the measure min = {Q k |k = 1, 2, . . . , m} if the following two conditions are satisfied (1998): H1 Acceptable advantage: Q(A2 )−Q(A1 ≥ 1)/(m − 1), where A2 is the alternative in the second position of the ranking list by R; where m shows the number of alternatives. H2 Acceptable stability in decision-making: The alternative A1 must also be the best when ranked bySk and/orRk , where k = 1, 2, . . . , m. A set of compromise solutions is proposed if one of the above conditions is not satisfied. The set of compromise solutions consists of: (1) Alternatives A1 and A2 , if H1 is satisfied and H2 is not satisfied. (2) Alternatives A1 , A2 , . . . , Am , if H1 is not satisfied; Am is determined by the relation Q(Am ) − Q(A1 ≥ 1)/(m − 1) for maximum M (the positions of these alternatives are close).
4 Empirical Case In this section, we begin with a case study to indicate performance of the purposed GDSS. In the following, first, we clarify our case profile and describe data collection. Next, we indicate background information of five decision makers. Then the components, criteria, alternatives of the evaluation process and results of the GDSSs are depicted. 4.1 Background and Data Collection A Middle Eastern dairy company developed a GDSS to reduce its pre-campaign media evaluation time and to improve effectiveness of its group decision making process. The GDSS uses data of actual performance of advertising vehicles, provided by a marketing research company. The data were collected during the tracking studies of advertising campaigns for an ice cream brand and derived from in-home surveys. Families with number of two households or more, ages 25–49, upscale, and welleducated audiences are as the target market in entire country. The target audience
123
P. F. Tafreshi et al. Table 2 The characteristics of the six decision-making experts Gender
Age
Educational level
Experience (years)
Job title
Expert 1
Male
51
Master of business administration
28
Advertising consultant
Expert 2
Male
46
Bachelor in management
21
Advertising specialist
Expert 3
Female
38
Ph.D. candidate in economics
11
Digital media planner
Expert 4
Female
52
Masters in marketing
23
Media director
Expert 5
Male
39
Bachelor in industrial engineering
15
Media research analyst
Table 3 The criteria and their descriptions Notation
Criteria
Description
C1
Reach
The unduplicated number of individuals potentially receiving at least one advertising exposure from a media vehicle or media plan
C2
Effective reach
Reach achieved among individuals who are exposed to an advertisement with a frequency greater than or equal to the effective frequency
C3
CPM
A common measure of individual vehicle efficiency and stand for the cost per thousand exposures from a single advertisement in that vehicle
C4
GRPs
Impressions divided by the number of people in the audience for an advertisement
C5
Composition
The percent of all exposed people who are in the target
C6
Brand penetration
Purchasers of a brand as a percentage of total population
C7
Brand awareness
Percentage of total population that is aware of a brand’s name
C8
Advertising awareness
Percentage of total population that is aware of a brand’s advertising
C9
Top of mind
First brand to consider
defined by a special interest “healthy food” and most of them participate in at least one outdoor sport. The company’s advertising department wants to evaluate available advertising vehicles with respect to post-campaign results and data. Criteria that are involved in advertising media selection have been selected according to the professionals’ knowledge and the experience of experts. To conduct the study, a decision making group of five experts is formed; Background information is illustrated in Table 2. Experts identified nine determinant criteria (see Table 3). The first five criteria, C1 to C5 , data set comprises total value for each of them during a 26-week campaign period. For example, the value number for the “C1 reach”
123
Developing a Group Decision Support System...
is the total sum of all reached people during 26 weeks of advertising campaign. For criteria C6 to C9 , the data were obtained from in-home surveys. Weekly, on average 167 households were participating in personal interviews. In our case, criteria C6 to C9 are advertising effectiveness measures. Note that all nine criteria are measured separately for each vehicle. 4.2 Results and Discussion We developed the DEMATEL based upon five experts’ perceptions and knowledge. The average matrix of A can be calculated based on Eq. 1 (Fig. 3). In Fig. 4, normalized initial direct-relation matrix D is illustrated. In Table 4, the total relation matrix T is calculated based on Eq. 5. Direct impacts and indirect influences of each criterion and total impact of them are illustrated in Table 5. As indicated in Fig. 5, “C1 reach” has the greatest influence on other criteria, and after that, “C8 brand awareness” with the value of 2.542 has the highest impact.
Fig. 3 Average matrix for experts’ opinions
Fig. 4 Normalized initial direct-relation matrix D
123
P. F. Tafreshi et al. Table 4 The total relation matrix T Criteria
C1
C2
C3
C4
C5
C6
C7
C8
C9
C1
0.09826
0.33065
0.34195
0.33815
0.15579
0.44596
0.39243
0.36566
0.39341
C2
0.06010
0.12138
0.17434
0.11904
0.08691
0.30351
0.27681
0.25030
0.27320
C3
0.09538
0.15550
0.14205
0.25955
0.18595
0.33426
0.19110
0.22648
0.28127
C4
0.15275
0.21547
0.29771
0.17618
0.18923
0.40034
0.35678
0.30234
0.31898
C5
0.07391
0.27109
0.19628
0.24914
0.07819
0.39107
0.32001
0.27818
0.34167
C6
0.04545
0.10093
0.20246
0.10665
0.07240
0.12064
0.13107
0.11615
0.12909
C7
0.07813
0.21227
0.18614
0.11344
0.06072
0.31088
0.14677
0.20620
0.26833
C8
0.17766
0.28868
0.28137
0.31074
0.11812
0.41448
0.37492
0.21285
0.36329
C9
0.07316
0.19328
0.13620
0.21107
0.06155
0.32666
0.28592
0.21987
0.15582
Table 5 The sum of influences given and received
Criteria
r
c
r+c
r−c
C1
2.862
0.855
3.717
2.007
C2
1.666
1.889
3.554
−0.223
C3
1.872
1.959
3.830
−0.086
C4
2.410
1.884
4.293
0.525
C5
2.200
1.009
3.208
1.190
C6
1.025
3.048
4.072
−2.022
C7
1.583
2.476
4.058
−0.892
C8
2.542
2.178
4.720
0.364
C9
1.664
2.525
4.188
−0.861
From received influences point, the least affected criterion is “C1 reach”. The second criterion in the term of received influences is “C5 composition” and the most affected criterion is “C6 Brand penetration”. According to the values of ri + ci , “C8 advertising awareness” is the first in the index of strength of influence given to and received by amongst all criteria, with the value of 4.720. And after C8 , “C4 GRPs” with the value of 4.293 and next three criteria are “C9 -Top of mind”, “C6 brand penetration”, and “C7 brand awareness”. With respecting the values of the column ri − ci in Fig. 5, as we expected the values for “C1 reach” is positive, since it affects other criteria more than the other ones affect it. The second criterion in terms of value is “C5 composition”, we interpret that experts have seen “C1 reach” and “C5 composition” as chief influencers on campaign result. On the other hand, we have C2, C3, C6, C7, and C9 with negative values of ri − ci denoting that these criteria are significantly affected by the other criteria, more than their influences on the other criteria. To obtain the IRM through the α-cut total relation matrix T α , we applied threshold value α = 0.15 (see Table 6). To draw the IRM, we idealized sum of direct and indirect effects. Then we used α > t α , then single headed arrows to exhibit the impact direction between criteria. If t21 12
123
Developing a Group Decision Support System...
Fig. 5 Influences given and received Table 6 The α-cut total relation matrix T α Criteria
C1
C2
C3
C4
C5
C6
C7
C8
C9
C1
0.0000
0.3307
0.3420
0.3382
0.1558
0.4460
0.3924
0.3657
0.3934
C2
0.0000
0.0000
0.1743
0.0000
0.0000
0.3035
0.2768
0.2503
0.2732
C3
0.0000
0.1555
0.0000
0.2595
0.1859
0.3343
0.1911
0.2265
0.2813
C4
0.1527
0.2155
0.2977
0.1762
0.1892
0.4003
0.3568
0.3023
0.3190
C5
0.0000
0.2711
0.1963
0.2491
0.0000
0.3911
0.3200
0.2782
0.3417
C6
0.0000
0.0000
0.2025
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
C7
0.0000
0.2123
0.1861
0.0000
0.0000
0.3109
0.0000
0.2062
0.2683
C8
0.1777
0.2887
0.2814
0.3107
0.0000
0.4145
0.3749
0.2129
0.3633
C9
0.0000
0.1933
0.0000
0.2111
0.0000
0.3267
0.2859
0.2199
0.1558
α > t α , then the flow is drawn from the flow is drawn from C2 to C1 , C2 → C1 ; If t12 21 C1 to C2 , C1 → C2 . Figure 6 indicates The IRM. Then experts fill out a matrix to give their opinion about importance of each criterion with respect to the other criteria. In Table 7, the total pairwise comparison matrix W T is represented based on Eq. 8. Now, we calculate the normalized α-cut total relation matrix T s based on Eq. 10 (see Table 8). After calculating T s and W T , we calculate the weighted supermatrix W based on Eq. 9 (Table 9). Table 10 illustrates the limited supermatrix W ∗ . The bar chart of the final weights is illustrated in Fig. 7. The normalized profiles of the alternatives are depicted in Table 11. As stated before, alternatives are advertising vehicles (e.g. Facebook, New York Times, etc).
123
P. F. Tafreshi et al.
Fig. 6 The IRM of the problem
Table 7 The total pairwise comparison matrix WT Criteria
C1
C2
C3
C4
C5
C6
C7
C8
C9
C1
0.0000
0.1940
0.2189
0.2379
0.2379
0.2027
0.2060
0.1986
0.2604
C2
0.0824
0.0000
0.1675
0.0844
0.0844
0.1098
0.1346
0.1033
0.1587
C3
0.1637
0.2238
0.0000
0.1626
0.1626
0.1284
0.1721
0.1676
0.1070
C4
0.1116
0.1594
0.1508
0.0000
0.0864
0.1194
0.1249
0.0765
0.0827
C5
0.0958
0.1058
0.1052
0.0864
0.0000
0.1296
0.1506
0.1038
0.1051
C6
0.0639
0.0726
0.0623
0.0611
0.0611
0.0000
0.0749
0.0412
0.0978
C7
0.1185
0.0939
0.1137
0.1196
0.1196
0.0944
0.0000
0.1618
0.1190
C8
0.2734
0.0974
0.0873
0.1723
0.1723
0.0999
0.0814
0.0000
0.0692
C9
0.0908
0.0530
0.0941
0.0758
0.0758
0.1159
0.0556
0.1472
0.0000
Table 8 The normalized α-cut total relation matrix T s Criteria
C1
C2
C3
C4
C5
C6
C7
C8
C9
C1
0.0000
1.0766
1.1134
1.1011
0.5073
1.4521
1.2778
1.1906
1.2810
C2
0.0000
0.0000
1.2276
0.0000
0.0000
2.1371
1.9491
1.7625
1.9237
C3
0.0000
0.8565
0.0000
1.4295
1.0241
1.8409
1.0525
1.2474
1.5491
C4
0.5705
0.8047
1.1119
0.6580
0.7067
1.4952
1.3325
1.1292
1.1913
C5
0.0000
1.1916
0.8628
1.0952
0.0000
1.7190
1.4067
1.2228
1.5019
C6
0.0000
0.0000
9.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
C7
0.0000
1.6138
1.4151
0.0000
0.0000
2.3635
0.0000
1.5676
2.0400
C8
0.6596
1.0718
1.0447
1.1538
0.0000
1.5389
1.3920
0.7903
1.3488
C9
0.0000
1.2491
0.0000
1.3641
0.0000
2.1111
1.8478
1.4209
1.0070
123
Developing a Group Decision Support System... Table 9 The weighted supermatrix W Criteria
C1
C2
C3
C4
C5
C6
C7
C8
C9
C1
0.0000
0.2089
0.2438
0.2620
0.1207
0.2944
0.2632
0.2364
0.3336
C2
0.0000
0.0000
0.2057
0.0000
0.0000
0.2346
0.2623
0.1821
0.3054
C3
0.0000
0.1917
0.0000
0.2324
0.1665
0.2364
0.1811
0.2090
0.1658
C4
0.0637
0.1283
0.1677
0.0000
0.0611
0.1785
0.1664
0.0864
0.0985
C5
0.0000
0.1261
0.0908
0.0947
0.0000
0.2228
0.2118
0.1269
0.1578
C6
0.0000
0.0000
0.5611
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
C7
0.0000
0.1516
0.1610
0.0000
0.0000
0.2230
0.0000
0.2537
0.2428
C8
0.1803
0.1044
0.0912
0.1987
0.0000
0.1538
0.1133
0.0000
0.0933
C9
0.0000
0.0663
0.0000
0.1034
0.0000
0.2446
0.1027
0.2092
0.0000
Table 10 The limited supermatrix W ∗ Criteria
C1
C2
C3
C4
C5
C6
C7
C8
C9
C1
0.1796
0.1796
0.1796
0.1796
0.1796
0.1796
0.1796
0.1796
0.1796
C2
0.0842
0.0842
0.0842
0.0842
0.0842
0.0842
0.0842
0.0842
0.0842
C3
0.1428
0.1428
0.1428
0.1428
0.1428
0.1428
0.1428
0.1428
0.1428
C4
0.1207
0.1207
0.1207
0.1207
0.1207
0.1207
0.1207
0.1207
0.1207
C5
0.0785
0.0785
0.0785
0.0785
0.0785
0.0785
0.0785
0.0785
0.0785
C6
0.0527
0.0527
0.0527
0.0527
0.0527
0.0527
0.0527
0.0527
0.0527
C7
0.0836
0.0836
0.0836
0.0836
0.0836
0.0836
0.0836
0.0836
0.0836
C8
0.1934
0.1934
0.1934
0.1934
0.1934
0.1934
0.1934
0.1934
0.1934
C9
0.0645
0.0645
0.0645
0.0645
0.0645
0.0645
0.0645
0.0645
0.0645
Fig. 7 Weights of the criteria
123
P. F. Tafreshi et al. Table 11 The profile of alternatives C1
C2
C3
C4
C5
C6
C7
C8
C9
A1
0.066
0.081
0.122
0.078
0.111
0.119
0.115
0.132
0.082
A2
0.171
0.168
0.103
0.146
0.069
0.030
0.049
0.074
0.098
A3
0.264
0.290
0.095
0.208
0.056
0.030
0.033
0.074
0.082
A4
0.011
0.006
0.049
0.025
0.056
0.090
0.066
0.044
0.049
A5
0.003
0.000
0.049
0.004
0.042
0.030
0.049
0.029
0.049
A6
0.092
0.097
0.076
0.073
0.083
0.075
0.082
0.088
0.098
A7
0.053
0.048
0.042
0.042
0.069
0.075
0.082
0.088
0.098
A8
0.119
0.081
0.086
0.106
0.069
0.060
0.066
0.044
0.049
A9
0.105
0.097
0.078
0.166
0.056
0.075
0.115
0.088
0.082
A10
0.053
0.032
0.061
0.042
0.097
0.090
0.066
0.074
0.066
A11
0.021
0.065
0.132
0.025
0.083
0.104
0.066
0.059
0.049
A12
0.006
0.034
0.064
0.047
0.111
0.119
0.098
0.088
0.066
A13
0.011
0.001
0.024
0.008
0.056
0.045
0.049
0.059
0.049
A14
0.026
0.001
0.020
0.031
0.042
0.060
0.066
0.059
0.082
Table 12 Aspired and tolerable levels C1
C2
C3
C4
C5
C6
C7
C8
C9
f j+
0.2637
0.2903
0.0196
0.2079
0.1111
0.1194
0.1148
0.1324
0.0984
f j−
0.0026
0.0003
0.1320
0.0042
0.0417
0.0299
0.0328
0.0294
0.0492
wj
0.1796
0.0842
0.1428
0.1207
0.0785
0.0527
0.0836
0.1934
0.0645
Table 13 Mean group utility and maximal regret of the alternatives A1
A2
A3
A4
A5
A6
A7
A8
A9
A10
A11
A12
A13
A14
Si 0.426 0.519 0.427 0.763 0.878 0.500 0.532 0.668 0.458 0.608 0.777 0.546 0.759 0.676 Ri 0.136 0.111 0.111 0.174 0.193 0.118 0.145 0.166 0.109 0.145 0.167 0.178 0.174 0.163
Aspired level and the tolerable level are calculated for each criterion (see Table 12). Oppositely, for a negative criterion (e.g. cost) aspired level defines as f j+ = mink f k j and tolerable level represents as f j− = maxk f k j . In this case, “C3 CPM” is a negative criterion. Then we calculate mean group utility and maximal regret of each vehicle (Table 13). The comprehensive indicator Q k is obtained based on Eq. 15. Figure 8 provides the values of Q k in an increasing order. Since the H1 hypothesis is not satisfied, Alternative A3, A9, A2, A6, A1, A7 are considered as a set of compromise solutions. As stated before, the less the Q k , the better the rank of the alternative.
123
Developing a Group Decision Support System...
Fig. 8 Final rankings of the alternatives
4.3 Assessment of Experts’ Perceptions Two quantitative assessments were made in order to collect experts’ opinions regarding pairwise comparisons and the GDSS. First, after making comparisons, we asked all five decision makers about how much they agree or disagree with the stated questions. The average values of experts’ answers are illustrated in Table 14, standard deviations are indicated in Parentheses. Next, after combining judgments that were made in both the DEMATEL and the ANP results were shown to each expert. In the second qualitative assessment, each expert is asked about how much he/she agree or disagree with the stated questions regarding outcomes of the GDSS. See Table 15 for the average values of second assessment, standard deviations are indicated in Parentheses. Both the DEMATEL and the ANP appeared enjoyable and easy to experts, yet the DEMATEL is a bit more enjoyable than the ANP. Experts did not rate either of techniques as very confusing. Comparisons of both techniques are made in pairs; yet in
Table 14 Perceptions on procedures DEMATEL
ANP
Using a scale where a 1 means “strongly disagree” and a 7 means “strongly agree,” how much do you agree or disagree that the previous series of comparisons you just made… …was enjoyable
5.6 (1.356)
…was confusing
2.2 (1.166)
2.8 (0.4)
5 (1.414)
…was easy
5.8 (1.67)
5.2 (1.166)
…allowed you to express your opinions
6 (1.264)
6 (1.414)
…for the first time took you a lot of time to complete
2 (1.159)
4.6 (1.854)
123
P. F. Tafreshi et al. Table 15 Experts opinions about results of the GDSS Average point (SD) Using a scale where a 1 means “strongly disagree” and a 7 means “strongly agree,” how much do you agree or disagree that the GDSS… …reduced tension in the group …helped in saving time during decision making process …could consider all aspects that you wanted
6.2 (1.720) 4 (1.414) 6.2 (1.720)
…gave you rational results on weights and rankings
5.8 (2.561)
…can be used in different media evaluation/selection problems
5.6 (2.576)
…results satisfyingly contained your opinions
6.4 (1.019)
…results are easy to interpret
4.8 (1.469)
the ANP, the decision maker should keep rational ratios between comparisons and this increased the complexity of comparisons of the ANP. From values of the “the allowed you to express your opinions” we can perceive that either techniques are equal in terms of the ability to capture experts’ opinions and perceptions (6 out of 7-pt scale). From experts’ point of view, the DEMATEL took less time to complete than the ANP for the first time. By putting together the value of confusion, the value of easiness, and the required time for completion, we can perceive that these items potentially affects the level of enjoyment for each technique. The DEMATEL was less confusing, easier, and took less time to complete, so we expect it to be more enjoyable than the ANP. Not only Suggested GDSS lowered tension among decision makers but also it brought time efficiency. They also rated the GDSS as a capable tool that can consider all needed aspects of media evaluation. By giving 5.8 out of 7, they expressed that the GDSS gives rational results on both weights of criteria and rankings of alternatives. The average point 5.6 out of 7 indicates that the proposed GDSS has the potential to be applied in a different media evaluation/selection problem. Although the GDSS lowered the tension during the decision making process, but tension is in prospect if decision makers did not find results of the GDSS satisfying. The assessment indicates that decision makers satisfyingly saw effects of their opinion in results, weights of criteria and rankings of alternatives, and this eliminates the possibility of further tension. Moreover, decision makers believe that the results are relatively easy to interpret.
5 Conclusions and Future Research Directions Media evaluation is of paramount importance since it directly affects results of advertising campaigns. Researchers developed various models and frameworks to optimize this decision-making problem. Although they tried to optimize media evaluation as a multifaceted problem, there still are several gaps to address. Previous studies ignored the fact that media evaluation is a group task. Therefore, some aspects of media evaluation such as decision making quality, tension management, and agility of the evaluation process that pertain to group decision making were neglected. To address mentioned gaps, we developed a flexible MCDM based GDSS that is able to deal
123
Developing a Group Decision Support System...
with pre-campaign planning, in flight optimization, and post-campaign evaluation. The GDSS allows media planners to add/remove any criterion/alternative and edit any value without any interruption in evaluation process and further mathematical modeling. The proposed GDSS consisted of three well-known MCDM methods, DEMATEL, ANP, and the VIKOR method. It uses DEMATEL method to capture and aggregate decision makers’ opinions about relations and linkages amongst criteria. Then decision makers’ judgments for criteria importance are aggregated using ANP method. Combination of these two methods eliminated the possibility of tension occurrence among decision makers; consequently, required amount of both time and money is reduced and the evaluation process became agile. At the end, the GDSS ranks the media using VIKOR method. With a case study in the Middle East, we demonstrated that the proposed GDSS helped decision makers to make rational decisions, reduced the evaluation time, precampaign costs, and tension among decision makers. Contributions of the paper is fivefold: First, a flexible and integrated MCDM model is developed to deal with advertising media evaluation that can lend itself to pre-campaign planning, in flight optimization, and post-campaign evaluation. The flexibility of the model allows online analytical processing (OLAP) as changing values of criteria does not lead to reworking on the model for possible corrections. Second, for the first time in the literature of media evaluation, group aspect of the process is taken into account and a realistic mathematical model is formed to capture decision makers’ judgments. Fourth, evaluation criteria for multiple platforms advertising campaigns are introduced. Third, the model of the GDSS can consider qualitative and quantitative criteria simultaneously, which was absent from the literature. Fifth, an alternative mathematical approach is proposed to correct wrong holistic assumption in ANP that all criteria affect one another equally. Although the GDSS addressed mentioned problems and gaps, it has some drawbacks. Pairwise comparisons using the fundamental scale turned out to be a time consuming process, especially for the first time. Many aspects affect the performance of advertising vehicles; the suggested GDSS did not consider several aspects. First, our GDSS did not include media planning details such as scheduling strategies. Second, effects of advertising placement and contents were not taken into account. Third, the GDSS only considered values of the criteria, while in many situations values not only depend on vehicles performance but also depend on competitors’ actions, noise level, sales patterns, product life cycle, etc. The developed GDSS can potentially lend itself to many practical applications; however, there are possibilities for future research. For example, instead of crisp sets researchers can use fuzzy sets or grey relation analysis to improve perception of decision makers’ vague opinions. Furthermore, simulation methods can be integrated to the GDSS to forecast decision outcomes and advertising campaign results based on past data. Other MCDM methods and approaches can be utilized instead of current methods and techniques to see if methodological changes enhance performance of the GDSS and effectiveness of campaigns.
123
P. F. Tafreshi et al. Acknowledgments The authors would appreciate Paul W. Ferris for contributing to development of the paper with his useful suggestions and constructive comments. We wish to thank Jaleh Hoseinzadeh for sharing unpublished results and participating in discussions.
References Ackermann F, Eden C (2001) Contrasting single user and networked group decision support systems for strategy making. Group Decis Negot 10:47–66 Adla A, Zarate P, Soubie JL (2011) A proposal of toolkit for GDSS facilitators. Group Decis Negot 20:57–77 Alavi M, Joachimsthaler EA (1992) Revisiting DSS implementation research: a meta-analysis of the literature and suggestions for researchers. MIS Q 16:95–116 Alter SL (1975) A study of computer aided decision making in organizations. Ph.D. dissertation, M.I.T Alter SL (1977) Why is man–computer interaction important for decision support systems? Interfaces 7:109–115 Alter SL (1980) Decision support systems: current practice and continuing challenge. Addison-Wesley, Reading, Boston Altuntasa S, Dereli T (2015) A novel approach based on DEMATEL method and patent citation analysis for prioritizing a portfolio of investment projects. Expert Syst Appl 42:1003–1012 Arnott D, Pervan G (2005) A critical analysis of decision support systems research. J Inf Technol 20:67–87 Azadi M, Farzipour-Saen R, Momeni E (2013) Developing a new neutral DEA model for media selection in the existence of imprecise data. Int J Oper Res 18:16–34 Bass FM, Lonsdale RT (1966) An exploration of linear programing in media selection. J Mark Res 2: 179–188 Baumeister J, Striffler A (2015) Knowledge-driven systems for episodic decision support. Knowl Based Syst 88:45–56 Behzadian M, Hosseini-Motlagh S, Ignatius J, Goh M, Sepehri M (2013) PROMETHEE group decision support system and the house of quality. Group Decis Negot 22:189–205 Bhattacharya UK (2009) A chance constraints goal programing model for the advertising planning problem. Eur J Oper Res 192:382–395 Bonczek RH, Holsapple CW, Whinston AB (1981) Foundations of decision support systems. Academic Press, New York Bose U (2015) Design and evaluation of a group support system supported process to resolve cognitive conflicts. Comput Hum Behav 49:303–312 Brown DB, Warshaw M (1965) Media selection by linear programing. J Mark Res 2:83–88 Charnes A, Cooper WW, DeVoe JK, Learner DB, Reinecke W (1968) A goal programing model for media planning. Manag Sci 14:422–430 Ching W, Yuen W, Michael NG, Zhang S (2006) A linear programing approach for determining optimal advertising policy. IMA J Manag Math 17:83–96 Danaher PJ, Lee J, Kerbache L (2010) Optimal internet media selection. Mark Sci 29:336–347 Day RL (1962) Linear programing in media selection. J Advert Res 2:40–44 DeSanctis G, Poole M, Zigurs I, Associates O (2008) The Minnesota GDSS Research Project: group support systems, group processes, and outcomes. J Assoc Inf Syst 9:551–608 Duque R, Rodríguez M, Hurtado M, Bravo C, Rodríguez-Domínguez C (2012) Integration of collaboration and interaction analysis mechanisms in a concern-based architecture for groupware systems. Sci Comput Program 77:29–45 Efremov RV, Av Lotov (2014) Multi-criteria remote asynchronous group decision screening: an experimental study. Group Decis Negot 23:31–48 Engel JF, Warshaw MR (1964) Allocating advertising dollars by linear programing. J Advert Res 4:42–48 Eom SB (2002) Decision support systems research (1970–1999). Lewiston, New York Fan S, Shen Q (2011) The effect of using group decision support systems in value management studies: an experimental study in Hong Kong. Int J Proj Manag 29:13–25 Farzipoor-Saen R (2011) Media selection in the presence of flexible factors and imprecise data. J Oper Res Soc 62:1695–1703 Ferguson RL, Jones CH (1969) A computer aided decision system. Manag Sci 15:550–562 Fontela E, Gabus A (1976) The DEMATEL observer. Battelle Institute, Geneva Research Center
123
Developing a Group Decision Support System... Fruchter GE, Kalish S (1998) Dynamic promotional budgeting and media allocation. Eur J Oper Res 111:15–27 Gabus A, Fontela E, (1973) Perceptions of the world problematique: communication procedure, communicating with those bearing collective responsibility. DEMATEL Report No.1, Battelle Institute, Geneva Research Centre Gallupe RB, DeSanctis G, Dickson GW (1988) Computer-based support for group problem-finding: an experimental investigation. MIS Q 12:277–298 Gorry A, Morton MSS (1971) A framework for information systems. Sloan Manag Rev 13:56–79 Hackley C (2005) Advertising and promotion: communicating brands. Sage, London Hashemian SM, Behzadian M, Samizadeh R, Ignatius J (2014) A fuzzy hybrid group decision support system approach for the supplier evaluation process. Int J Adv Manuf Technol 73:1105–1117 Holt CC, Huber GP (1969) A computer aided approach to employment service placement and counselling. Manag Sci 15:573–594 Ishizaka A, Nemery P (2013) A multi-criteria group decision GDSS for partner grouping when sharing facilities. Group Decis Negot 22:773–799 Jha PC, Aggarwal R, Gupta A (2011) Optimal media planning for multi-products in segmented market. Appl Math Comput 217:6802–6818 Keown AJ, Duncan CP (1979) Integer goal programing in advertising media selection. Decis Sci 10:577–592 Kluyver-De CA (1978) Hard and soft constraints in media scheduling. J Advert Res 18:27–31 Koo C, Hong T, Kim J (2014) A decision support system for determining the optimal size of a new expressway service area: focused on the profitability. Decis Support Syst 67:9–20 Kumar S, Jacob V, Sriskandarajah C (2006) Scheduling advertisements on a web page to maximize revenue. Eur J Oper Res 173:1067–1089 Kwak NK, Lee CW, Kim JH (2005) An MCDM model for media selection in the dual consumer/industrial market. Eur J Oper Res 166:255–265 Lee AM, Burkart AJ (1960) Some optimization problems in advertising media planning. Oper Res Q 11:113–122 Lee CW, Kwak NK (1999) Information resource planning using AHP-based goal programing model. J Oper Res Soc 51:1–8 Lee SM (1973) Goal programing for decision analysis of multiple objectives. Sloan Manag Rev 14:11–24 Lei N, Moon SK (2015) A decision support system for market-driven product positioning and design. Decis Support Syst 69:82–91 Limayem M, Banerjee P, Ma L (2006) Impact of GDSS: opening the black box. Decis Support Syst 42: 945–957 Little JDC, Lodish LM (1966) A media selection model and its optimization by dynamic programing. Ind Manag Rev 8:15–23 Locander WB, Scamell RW, Sparkman RM, Burton JP (1978) Media allocation model using nonlinear benefit curves. J Bus Res 6:273–293 Matsatsinis NF, Grigoroudis E, Samaras A (2005) Aggregation and disaggregation of preferences for collective decision-making. Group Decis Negot 14:217–232 Mesak H, Zhang H (2001) Optimal advertising pulsation policies: a dynamic programing approach. J Oper Res Soc 11:1244–1255 Mihiotis A, Tsakiris I (2004) A mathematical programing study of advertising allocation problem. Appl Math Comput 148:373–379 Milani AS, Shanian A, Lynam C, Scarinci T (2013) An application of the analytic network process in multiple criteria material selection. Mater Des 42:622–632 Nerlove M, Arrow KJ (1962) Optimal advertising policy under dynamic conditions. Economica 29:129–142 Opricovic S (1998) Multicriteria optimization of civil engineering systems. Faculty of Civil Engineering, Belgrade Penichet V, Lozano M, Gallud J, Tesoriero R (2009) User interface analysis for groupware applications in the TOUCHE process model. Adv Eng Softw 40:1212–1222 Power DJ (2002) Decision support systems: concepts and resources for managers. Quorum Books, Westport Raymond RC (1966) Use of the time-sharing computer in business planning and budgeting. Manag Sci 12:363–381 Riorden J (1958) An introduction to combinatorial analysis. Wiley, New York Rios J, Rios-Insua D (2008) A framework for participatory budget elaboration support. J Oper Res Soc 59:203–212
123
P. F. Tafreshi et al. Saaty TL (2013) The modern science of multicriteria decision making and its practical applications: the AHP/ANP approach. Oper Res 61:1101–11018 Saaty TL, Ozdemir MS (2003) Why the magic number seven plus or minus two. Math Comput Model 38:233–244 Sharma RN, Chand N, Sharma V, Yadav D (2015) Decision support system for operation, scheduling and optimization of hydro power plant in Jammu and Kashmir region. Renew Sustain Energy Rev 43:1099–1113 Sprague RH, Carlson JR, Carlson ED (1982) Building effective decision support systems. Prentice-Hall, Upper Saddle River Stasch SF (1965) Linear programing and space–time considerations in media selection. J Advert Res 5: 40–46 Tavanaa M, Behzadian M, Pirdashti M, Pirdashti H (2014) A PROMETHEE-GDSS for oil and gas pipeline planning in the Caspian Sea basin. Energy Econ 36:716–728 Tsai WH, Yang CC, Leu JD, Lee YF, Yang CH (2013) An integrated group decision making support model for corporate financing decisions. Group Decis Negot 22:1103–1127 Turban E (1967) The use of mathematical models in plant maintenance decision making. Manag Sci 13: 342–359 Tzeng GH, Chiang CH, Li CW (2007) Evaluating intertwined effects in e-learning programs: a novel hybrid MCDM model based on factor analysis and DEMATEL. Expert Syst Appl 32:1028–1044 Watson RT, DeSanctis G, Poole MS (1988) Using a GDSS to facilitate group consensus: some intended and unintended consequences. MIS Q 12:463–478 Yea J, Dasiopoulou S, Stevenson G, Meditskos G, Kontopoulos E, Kompatsiaris I, Dobsona S (2015) Semantic web technologies in pervasive computing: a survey and research roadmap. Pervasive Mobile Comput 23:1–25 You X, You J, Liu H, Zhen L (2015) Group multi-criteria supplier selection using an extended VIKOR method with interval 2-tuple linguistic information. Expert Syst Appl 42:1906–1916 Zhang S, Li Y, Zhang T, Peng Y (2015) an integrated environmental decision support system for water pollution control based on TMDL—a case study in the Beiyun River watershed. J Environ Manag 156:31–40 Zlofani SH, Aghdaie MH, Derakhti A, Zavadskas EK, Varzandeh MHM (2013) Decision making on business issues with foresight perspective; an application of new hybrid MCDM model in shopping mall locating. Expert Syst Appl 40:7111–7121 Zufryden FS (1973) Media scheduling a stochastic dynamic model approach. Manag Sci 19:1395–1406
123