You are on page 1of 7

Expert Systems With Applications 89 (2017) 266–272

Contents lists available at ScienceDirect

Expert Systems With Applications


journal homepage: www.elsevier.com/locate/eswa

The behavioral TOPSIS


K. Paul Yoon a, Won Kyung Kim b,∗
a
Department of Information Systems and Decision Sciences, Silberman College of Business, Fairleigh Dickinson University, 1000 River Road, Teaneck, NJ
07666, USA
b
Department of Computer Engineering, Kyungnam University, Wolyundong 449, Masan, Kyungnam 631-702, Republic of Korea

a r t i c l e i n f o a b s t r a c t

Article history: This paper updates and expands the TOPSIS that was introduced four decades ago. First, we re-present
Received 3 March 2017 the logic of TOPSIS according to the traditional decision theory. It shows that TOPSIS also has a built-in
Revised 26 June 2017
multi-attribute value function that is not revealed explicitly. So far, this has been a hidden aspect of TOP-
Accepted 26 July 2017
SIS. Second, we incorporate the decision maker’s (DM) behavioral tendency into TOPSIS. The behavioral
Available online 27 July 2017
TOPSIS accommodates the loss aversion concept in behavioral economics. A DM can set the loss aversion
Keywords: ratio to reflect his/her choice inclination. Two cases are illustrated to show the efficacy of the behavioral
Multiple attribute decision making TOPSIS.
TOPSIS © 2017 Elsevier Ltd. All rights reserved.
Efficient frontier
Multi-attribute value function
Behavioral economics
Loss aversion
Endowment effect

1. Introduction havioral economics (Kahneman, 2011; Thaler, 1980). That is, to in-
troduce a behavioral TOPSIS.
The TOPSIS (Technique for Order Preference by Similarity to This paper is organized in seven sections. Section 2 shows
Ideal Solution), which was introduced by Hwang and Yoon (1981), how TOPSIS effectively screens the non-dominated alternatives.
became a classic multiple attribute decision making (MADM) Section 3 explains the choice logic of TOPSIS by way of the
method with more than 4500 citations, and appeared in recent classical decision theory. Section 4 proposes a behavioral TOPSIS.
books (Garvey, 2008; Ishizaka & Nemery, 2013; Tzeng & Huang, Section 5 demonstrates the efficacy of the proposed method using
2011; Yoon & Hwang, 1995). There is rich literature source on the two cases. Section 6 discusses the validity of the behavioral TOPSIS.
TOPSIS: see Behzadian, Otaghsara, Yazdani, and Ignatius (2012) for Section 7 offers concluding remarks. The appendix presents com-
a state-of-the-art survey of applications; refer to Buede and putation steps of the TOPSIS.
Maxwell (1995), Olson (2004), Opricovic and Tzeng (2004), and
Zanakis, Solomon, Wishart, and Dublish (1998) for comparison
studies with other MADM methods. 2. Reduction of dimensionality by TOPSIS
This paper has multiple objectives. First, to discover a rela-
tionship between TOPSIS and the traditional multi-attribute value The goal of the MADM method is to prioritize m alternatives
function (Keeney & Raiffa, 1976). This may unfold the hidden logic that are evaluated by n attributes. An alternative is denoted as
of the method. Second, to correct the misunderstanding that TOP-  
SIS always selects an alternative that is closest to the best alter- Ai = xi1 , . . . , xi j , . . . ., xin , i = 1, . . . , m, (1)
native and farthest from the worst alternative simultaneously. No
MADM method can. Third, to rewrite the TOPSIS steps using better where xij is the performance rating of the ith alterative, Ai , with
terminology. For instance, we will use ‘anti-ideal solution’ instead respect to the jth attribute, Xj .
of ‘negative-ideal solution’. Fourth, to implement decision maker’s The TOPSIS creates two artificial, infeasible alternatives to uti-
(DM) choice policy by adopting the gain and loss concept in be- lize them as reference points like the polar star in the heavens and
the nadir point in hell. The ideal solution, A + is composed by the
best attribute values across the alternatives, whereas anti-deal so-
lution, A − is consisted of the worst attribute values.

Corresponding author.  
E-mail addresses: yoon@fdu.edu (K.P. Yoon), wkkim@kyungnam.ac.kr (W.K. Kim). A+ = x+ + +
1 , . . . , x j , . . . ., xn (2)

http://dx.doi.org/10.1016/j.eswa.2017.07.045
0957-4174/© 2017 Elsevier Ltd. All rights reserved.
K.P. Yoon, W.K. Kim / Expert Systems With Applications 89 (2017) 266–272 267

Fig. 2. The efficient frontier.

Fig. 1. An alternative in (D+ , D− ) dimension (Chang et al., 2010).

where x+j is the best value for the jth attribute among m alterna-
tives.
 
A− = x− − −
1 , . . . , x j , . . . ., xn (3)
where x−j is the worst value for the jth attribute among m alterna-
tives.
The TOPSIS measures the Euclidean distances between alterna-
tive Ai and A + , which is denoted by D+ i
, and between Ai and A − ,
which is D− i
. Then an alternative is denoted by two consolidated
attributes.
 
Ai = D+
i
, D−
i
, i = 1, . . . , m (4)
The revised definition of alternatives in the Eq. (4) allows the
DM to reduce n-dimensional comparison to two-dimensional com-
parison. Fig. 1 depicts how an alternative in three benefit attributes
is transformed in two distances from ideal and anti-ideal points
(Chang, Lin, Lin, & Chiang, 2010). It is natural that we want to Fig. 3. Preference order by original TOPSIS.
maximize D − and minimize D + . Then alternative Aj is judged to
dominate alternative Ak when one of the following conditions are
met: Eq. (6) can be rewritten by
a. D−j > D−
k
and D+j < D+
k
D = V D+

(7)
b. D−j > D−
k
and D+j = D+k (5) Eq. (7) indicates a straight line that has a slope of V and y-
c. D−j = D−
k
and D+j < D+
k axis intercept of zero where the value increased, the line becomes
steeper. Then the most preferred alternative is one that meets the
Otherwise Aj and Ak do not dominate each other.
indifference curve (actually a line in our case) with the steepest
A collection of all non-dominated alternatives is called the effi-
slope. Fig. 3 shows A1 is the most preferred alternative and A3 is
cient frontier or the Pareto optimality set (Keeney & Raiffa, 1976). Ef-
the least preferred, which is concisely written as A1  A2  A3 .
ficient alternatives are marked with overlaying solid dots in Fig. 2.
According to Eq. (6), we can tell that Aj is preferred to Ak if
Now, can we prioritize alternatives in the efficient frontier? Yes, it
is possible only if a DM can assess a value function, V(D + , D − ) for D j− Dk −
each alternative. > (8)
D j+ Dk +
3. The original TOPSIS Eq. (8) is mathematically equivalent to the following equation:

It may be a very heavy burden to create a value function with D j− Dk −


+ − > (9)
n attributes, (X1 ,…, Xj ,…Xn ), but it may be a relatively light task to Dj + Dj Dk + Dk −
+

form a value function with two consolidated attributes, (D + , D − ). You may notice that the original TOPSIS has the value function
Since two attributes possess the opposite goals, that is, we want to of Eq. (9). That is, the relative closeness of alternative Ai to the
maximize D − and to minimize D + , we can come up with a simple ideal solution is defined as
value function:
Di −
V = D /D − +
(6) Ci+ = , i = 1, . . . ., n (10)
Di + Di −
+
268 K.P. Yoon, W.K. Kim / Expert Systems With Applications 89 (2017) 266–272

Note that 0 ≤ Ci+ ≤ 1, where Ci+ = 0 when Ai = A − , and Ci+ = 1


when Ai = A + . TOPSIS picks an alternative that has the maximum
C+ .
Since Eq. (6) is made by a ratio of two numbers, the
most preferred alternative by TOPSIS may be neither the clos-
est to A + nor be the farthest from A − . For instance, if
A1 = ( D+1
, D−
1
) = ( 0.2, 0.4 ), A 2 = ( D+
2
, D−
2
) = (0.3, 0.7 ) and A3 =
+ −
(D3 , D3 ) = (0.4, 0.8 ), then A2 becomes the best choice because
C1+ = 0.67, C2+ = 0.70 and C3+ = 0.67.

4. The behavioral TOPSIS

What are the physical and economical interpretations of Di +


and Di − ? Opricovic and Tzeng (2004) pinpointed that the original
TOPSIS did not mention it, and did not consider the relative impor-
tance of these distances. Chen, Kilgour, and Hipel (2011) imagined
different DM’s attitudes dealing with D − and D + : aggressive ef-
fort to minimize D + and conservative effort to maximize D − . Then
they formulate a weighted averaging index with D − and D + . We
Fig. 4. Preference order by behavioral TOPSIS.
think a DM should aggressively search in both situations.
Now we discover that there exists an interesting relationship
between them. The magnitude of Di − is viewed as the gain that
a DM earns from taking Ai instead of anti-ideal solution A − . The
magnitude of Di + is viewed as the opportunity loss (Raiffa, 1968)
or opportunity cost (Newnan & Johnson, 1995) that a DM pays from
taking Ai instead of ideal solution A + , thus Di + can be considered
as loss.
We all hate losses. Specifically speaking, losing something
makes us (roughly) twice as miserable as gaining the same
thing makes us happy. This means that a DM is willing to
give up one unit of loss only if he/she gets two units of gain.
Thaler (1980) called this DM’s biases as endowment effect, and
Kahneman and Tversky (1984) called it as loss aversion. Later these
behavioral economists have estimated from field experiments that
the loss aversion ratio is in the range of 1.5 and 2.5 (Novemsky &
Kahneman, 2005).
Let the global loss aversion ratio be denoted as
Change in Gain D −
λ = = (11)
Change in Loss D +
where the choice behavior is loss adverse when λ > 1; neutral Fig. 5. Choices with different loss aversion ratios.
when λ = 1; loss prone (or risk-seeking) when λ < 1.
Now, we want to implement the loss aversion ratio into a value
function. The value function of Eq. (6) cannot take this ratio be- both methods take the equal weights between D+ and D− . Though
cause the function itself is this ratio. We thus need another value we cannot prove this relationship mathematically, their extremely
function: close relationship will be demonstrated through cases in Section 6.
Lastly, we should recognize that Eq. (12) is one of value func-
V = D− − λD+ (12)
tions that aggregate D − and D + . Hence, it may be possible for a
Eq. (12) can be rewritten as better context suited value function to be formulated.
D = λD+ + V

(13) The Appendix shows the detailed steps of original TOPSIS and
behavioral TOPSIS. They have identical steps except Step 5.
Eq. (13) indicates a straight line with a slope of λ and an inter-
cept of V in the (D + , D − ) space.
Fig. 4 illustrates the choice of behavioral TOPSIS. Here an alter- 5. Two illustrative cases
native value increases as the y-intercept becomes larger. The pref-
erence ranking of three non-dominated of alternatives are A4  A5 5.1. Selecting an electrical power plant site
 A6 since A4 meets the value line with the largest y-intercept.
Fig. 5 contrasts choice patterns from different loss aversion ra- Stewart (1992) considered a site selection problem for electrical
tios. A DM, who is highly loss averse (represented by the steep power plant among six European countries. The selection commit-
value line), should choose A7 rather than A8 . Why? We observe tee identified six attributes that determine the final selection:
that A7 has much shorter distance of D+ than A8 . That is, A7 is
closer to A+ . The fear of loss for not getting the ideal solution X1 : work force required (cost attribute)
could be compensated by taking an alternative that closely resem- X2 : power generated in MW (benefit attribute)
ble the ideal solution. Whereas a DM, who is less loss adverse (rep- X3 : construction cost in 106 US dollars (cost)
resented by less steep line), chooses A8 . X4 : annual maintenance cost in 106 US dollars (cost)
We expect that both original TOPSIS and behavioral TOPSIS with X5 : number of villages to be evacuated (cost)
λ = 1 should yield the identical preference rating (or order) since X6 : safety level on an interval scale (benefit)
K.P. Yoon, W.K. Kim / Expert Systems With Applications 89 (2017) 266–272 269

Table 1
Candidate sites for power plant.

Countries X1 X2 X3 X4 X5 X6

Italy (I) 80 90 600 54 8 5


Belgium (B) 65 58 200 97 1 1
Germany (G) 83 60 400 72 4 7
United Kingdom (U) 40 80 10 0 0 75 7 10
Portugal (P) 70 72 600 60 9 8
France (F) 94 96 700 36 5 6

Fig. 7. Ten portfolios in return and variance.

Fig. 6. Six sites in (D+ , D− ) space.

Table 1 shows the rating of six countries in terms of six at-


tributes. The committee assumed the equal weight among all at-
tributes. Fig. 6 depicts locations of six sites in the (D+ , D− ) space.
We first notice that Portugal and Italy are dominated by other
countries, and four other countries belong to the efficient frontier.
Table 2 summarizes the priority ranking under the various
choice policies. First, we notice the two dominated countries, Por-
tugal and Italy are ranked 5th and 6th, respectively. Most policies
pick United Kingdom as the most preferred sites, but France be-
Fig. 8. Ten portfolios in (D+ , D− ) space.
comes the top choice by the behavioral TOPSIS with loss averse
policy (λ = 2). As we expected, the ranking from the original TOP-
SIS is identical with that of the behavioral TOPSIS with λ = 1.
non-dominated portfolio. Other portfolios are dropped from the ef-
5.2. Selecting securities ficiency frontier due to their relatively large standard deviation.
Next, we make a preference ranking using the TOPSIS procedure.
Investors today must choose among a staggering number of in- Table 4 summarizes the priority ranking under the various
vestment alternatives. They are searching for securities that ren- choice policies. First, we notice the rankings of bottom 5 portfo-
der the maximum return with the minimum risk. The risk aver- lios (F, G, H, J, K) do not change under any loss aversion ratio. Sec-
sion attitude is a conspicuous behavior we observe in the invest- ond, top five portfolios get varying ranks depending on the ratios.
ment market. An investment firm generates 10 portfolios from five Third, as we expected, the ranking from the original TOPSIS is al-
stocks (Hearth & Zaima, 2004). Table 3 lists 10 portfolios that are most identical with that of the behavioral TOPSIS with λ = 1. Actu-
measured by two attributes: ally, the correlation coefficient between two ratings is 0.99586.
Table 5 lists the ranking of top five portfolios under different
X1 : expected return in percent (benefit attribute) loss attitudes. We observe the best portfolio changes from A to C
X2 : standard deviation in percent (cost attribute) when the DM’s loss aversion ratio increases. In addition, portfolio
D jumps to the second rank when λ = 3.
Fig. 7 shows 10 portfolios in terms of the mean return and the
standard deviation. All portfolios belong to the efficient frontier be-
cause a portfolio with higher return has higher standard deviation. 6. Discussion
Can we identify the higher-level efficient frontier within the cur-
rent efficient frontier? Yes, the TOPSIS may render it. First, we would like to discuss the validity of using the Eu-
Fig. 8 shows same 10 portfolios that are converted in two dis- clidean distance as an exchange medium in the endowment theory.
tance measures. Here we assumed that return and risk have the A number of experimental studies exclusively used the amount of
equal weight in computing the distances from ideal and anti-ideal dollar as the magnitude of gain and loss (Thaler, 1980; Tversky &
points. Fig. 8 depicts only four portfolios (A, B, C, D) emerge as the Kahnneman, 1981; Morewedge & Giblin, 2015).
270 K.P. Yoon, W.K. Kim / Expert Systems With Applications 89 (2017) 266–272

Table 2
Site preference rankings under various policies.

Alternative sites Distance measure Original TOPSIS Behavioral TOPSIS

λ = 0.5 λ=1 λ=2


D+ D− C+ Rank V Rank V Rank V Rank

Italy 0.6604 0.4379 0.3987 6 0.1077 6 −0.2225 6 −0.8829 6


Belgium 0.7114 0.6819 0.4894 4 0.3262 2 −0.0295 4 −0.7409 4
Germany 0.5604 0.5633 0.5013 3 0.2831 4 0.0029 3 −0.5574 2
United Kingdom 0.6242 0.6588 0.5135 1 0.3467 1 0.0347 1 −0.5895 3
Portugal 0.6494 0.5351 0.4517 5 0.2104 5 −0.1143 5 −0.7638 5
France 0.5758 0.5966 0.5088 2 0.3086 3 0.0207 2 −0.5551 1

Table 3 Table 5
Ratings of 10 portfolios. Ranking of top five portfolios with (w1 , w2 ) = (0.5, 0.5).

Portfolio X1 X2 Rank Behavioral TOPSIS

A 21.23 9.68 λ = 0.5 λ = 1.0 λ = 2.0 λ = 2.5 λ = 3.0


B 23.29 10.95
1 A A C C C
C 24.31 11.68
2 B B B B D
D 25.49 12.73
3 C C D D B
E 26.53 14.13
4 D D A A E
F 27.57 16.11
5 E E E E A
G 28.78 18.63
H 29.84 21.46
J 31.07 24.55
Table 6
K 32.16 27.77
Ranking of top five portfolios with (w1 , w2 ) = (0.6, 0.4).

Rank Behavioral TOPSIS


However, there are some studies where non-monetary scales λ = 0.5 λ = 1.0 λ = 2.0 λ = 2.5 λ = 3.0
were successfully utilized to see the effect of endowment. For ex-
1 A C D D D
ample, Harbaugh, Krause, and Vesterlund (2001) showed children 2 B D C E E
also place a higher value on objects they own over objects they 3 C B E C C
do not own. They used a toy as a medium of value instead of the 4 D E B B B
5 E A F F F
monetary unit. In order to trace the origin of the endowment ef-
fect, Brosnan et al. (2007) and Lakshminaryanan, Chen, and San-
tos (2008) utilized chimpanzees as experimental subjects. They
used food such as a frozen fruit-juice stick and a PVC pipe filled MADM environment, the loss aversion indicates the aversion (or
peanut butter, which are favored by chimpanzees, as the medium discrimination) against the unattained amount toward the ideal so-
of exchange. lution. Hence, the higher aversion ratio means the heavier penalty
Thus, the magnitude of loss or gain is not restricted to mone- to inferior alternatives, not to the cost attributes. Accordingly, the
tary or financial costs. In the MADM context where attributes pos- greater weight on the cost attributes does not necessarily lead to
sess incommensurable units, it may be impossible to pick one unit the similar effect of the higher loss aversion ratio.
that represents all attribute ratings. In this paper, we utilized the We revisit the portfolio selection in Section 5.2 to observe the
nth dimensional Euclidean distance, which was constructed by an sensitivity of preference ranking due to changes of the attribute
amalgamation of n-normalized attribute ratings, in order to mea- weights and the loss aversion ratios. We may observe this re-
sure the magnitude of gain and loss. lationship more clearly, since this case has only two attributes.
The next question is how do the attribute weight and the loss Table 6 shows the preference ranking with (w1 , w2 ) = (0.6, 0.4),
aversion ratio effect the final selection? The behavioral TOPSIS al- that is 0.6 is assigned to the mean return and 0.4 is assigned to the
lows DMs to express their judgements twice during the course of standard deviation. Portfolio D becomes the first rank, and portfo-
decision processes. First, the DM assigns weights (or importance) lio A is dropped from the top five list when the DM chooses the
on each attribute and then decides the loss aversion policy. In higher aversion ratio. Table 7 shows ranking with (w1 , w2 ) = (0.4,

Table 4
Preference ranking under various loss aversion ratios.

Portfolio Distance measure Original Behavioral TOPSIS

TOPSIS λ = 0.5 λ=1 λ=2 λ = 2.5 λ=3


D+ D− C+ Rank V Rank V Rank V Rank V Rank V Rank

A 0.1262 0.3211 0.7179 4 0.258038 1 0.194935 1 0.068729 4 0.005626 4 −0.05748 5


B 0.1049 0.2995 0.7407 3 0.247105 2 0.194669 2 0.089798 2 0.037362 2 −0.01507 3
C 0.0973 0.2878 0.7473 1 0.239168 3 0.190494 3 0.093146 1 0.044472 1 −0.00420 1
D 0.0941 0.2715 0.7425 2 0.224417 4 0.177344 4 0.083199 3 0.036127 3 −0.01095 2
E 0.1023 0.2498 0.7094 5 0.198602 5 0.147449 5 0.045141 5 −0.00601 5 −0.05717 4
F 0.1259 0.2196 0.6356 6 0.156631 6 0.093704 6 −0.03215 6 −0.09507 6 −0.15800 6
G 0.1636 0.1842 0.5296 7 0.10239 7 0.020586 7 −0.14302 7 −0.22483 7 −0.30663 7
H 0.2108 0.1498 0.4153 8 0.044356 8 −0.06106 8 −0.27189 8 −0.37731 8 −0.48273 8
J 0.2643 0.1272 0.3249 9 −0.00495 9 −0.13709 9 −0.40137 9 −0.53351 9 −0.66565 9
K 0.3211 0.1262 0.2821 10 −0.03437 10 −0.19494 10 −0.51608 10 −0.67665 10 −0.83722 10
K.P. Yoon, W.K. Kim / Expert Systems With Applications 89 (2017) 266–272 271

Table 7 aversion ratios, and then choose the final one. This may be a nudge
Ranking of top five portfolios with (w1 , w2 ) = (0.4, 0.6).
offered to the DM from the normative decision scientists.
Rank Behavioral TOPSIS In the last two decades, due to the rapid growth of the fuzzy
λ = 0.5 λ = 1.0 λ = 2.0 λ = 2.5 λ = 3.0 set theory, we have seen so many papers that accommodate fuzzy
data in the TOPSIS procedure (Chen & Hwang, 1992; Mardani, Ju-
1 A A B B B
2 B B A C C
soh, & Zavadska, 2015). Likewise, we expect efficient fuzzy behav-
3 C C C A A ioral TOPSIS methods will emerge.
4 D D D D D
5 E E E E E
Appendix

The Revised TOPSIS algorithm


0.6). This time, portfolio D remains at the forth rank under all poli- Step 0: Define an MADM Problem.
cies, and portfolio B captures the first rank under loss adverse pol-  
icy. Ai = xi1 , . . . , xi j , . . . , xin , i = 1, . . . , m
It is natural that Tables 6 and 7 reveal quite different rankings
where each attribute may possess incomparable unit.
due to the different attribute weight assignment. When compar-
Step 1: Calculate comparable ratings using the vector normaliza-
ing preference rankings in Tables 5, 6, and 7, we can observe that
tion.
preference rankings are more sensitive to the changes of attribute
weights than that of loss aversion ratios. The tendency seems to be x
ri j =  i j , i = 1, . . . , m; j = 1, . . . , n
more conspicuous when we deal with small number of attributes. m
x2i j
i=1
The portfolio has only two attributes.
The third question is when to use the original TOPSIS, and Step 2: Calculate weighted normalized ratings.
when to use behavioral TOPSIS. Both methods are based on the
traditional multi-attribute utility theory (Keeney & Raiffa, 1976). vi j = w j ri j , i = 1, . . . , m; j = 1, . . . , n
We have demonstrated that the original TOPSIS and the behav-
ioral TOPSIS with λ = 1 render almost equivalent preference rat- where wj is the weight of the jth attribute assessed by the DM.
ing. The original TOPSIS has an intuitive, straightforward logic that Step 3: Identify ideal solution and anti-ideal solution.
should be easily understood by ordinary DMs, while the behavior
 
A+ = v+1 , v+2 , . . . , v+j , . . . , v+n
TOPSIS contains a little more sophisticated concept. Studies shows
that the endowment effect is not confined to private goods or the 


laboratory. People demand more to give up entitlements such as = max vi j | j ∈ J1 , min vi j | j ∈ J2 |i = 1, 2, . . . , m
time, intellectual property, public land, and environmental, health, i i
and safety regulations than they are willing to pay to acquire them
(Buccafusco & Sprigman, 2010; Hammack & Brown, 1974; Horowitz  
A− = v−1 , v−2 , . . . , v−j , . . . , v−n
& McConnell, 2002).
The behavioral TOPSIS is expected to render a better choice
than other methods (including original TOPSIS) particularly in the



= min vi j | j ∈ J1 , max vi j | j ∈ J2 |i = 1, 2, . . . , m
public domain, such as selection for nuclear power plant sites, oil i i
pipeline routes, next generation air-defense systems, etc. One rea-
son is that this method properly reflects the DM’s wishes to choose where J1 is a set of benefit attributes and J2 is a set of cost at-
the very best alternative that resembles the ideal solution, not to tributes.
mention of the worry of taking inferior alternatives. We are look- The benefit attribute has the greater preference with the larger
ing forward to seeing the successful applications of the behavioral attribute rating; the cost attribute has the lesser preference with
TOPSIS. the larger attribute rating.
Step 4: Calculate Euclidean distances to ideal solution and anti-
ideal solution.
7. Concluding remarks
n 
2
D+
i
= vi j − v+j , i = 1, . . . , m
In this paper, we tried to find a relationship between the TOP-
j=1
SIS and the traditional multi-attribute value function. It has shown
that the TOPSIS also has a built-in multi-attribute value function
n 
2
that is not revealed explicitly. This finding would enhance the
D−
i
= vi j − v−j , i = 1, . . . , m
adaptability of TOPSIS methodology, particularly in the public do-
j=1
main where the legitimacy of the selection is crucial.
Then we proposed a behavioral TOPSIS that incorporates the Step 5: Compute value for each alternatives.
concept of gain and loss in behavioral economics. A DM can im-
plement his/her loss aversion ratio in addition to attribute weights (a) For original TOPSIS
in the selection process. This descriptive-prescriptive linkage sug-
D−
gests a meaningful way of improving decision-making. Ci+ =  i
 , i = 1, . . . , m
Behavioral economists discovered DM’s inherent bias and er- D+
i
+ D−
i
rors that prohibit making rational choices. Thaler and Sun-
(b) For Behavioral TOPSIS
stein (2008) suggested some nudges to improve decisions on ev-
eryday affairs. Avoiding the loss aversion is one of them. The be- Vi = D−
i
− λD+
i
, i = 1, . . . , m
havioral TOPSIS may not offer the best rational choice, but renders
consequences resulting from a DM’s different attitudes toward the where λ is a loss aversion ratio chosen by the DM.
loss. A DM should compare solutions that come from varying loss Step 6: Rank alternatives according to Ci + or Vi .
272 K.P. Yoon, W.K. Kim / Expert Systems With Applications 89 (2017) 266–272

References Kahneman, D., & Tversky, A. (1984). Choice, values and frames. American Psycholo-
gist, 39, 341–350.
Behzadian, M., Otaghsara, S. K., Yazdani, M., & Ignatius, J. (2012). A state-of-the-art Keeney, R. L., & Raiffa, H. (1976). Decisions with multiple objectives: Preferences and
survey of TOPSIS applications. Expert Systems with Applications, 39(17), value tradeoffs. New York: Wiley.
13051–13069. Lakshminaryanan, V., Chen, M. K., & Santos, L. R. (2008). Endowment effect in ca-
Brosnan, S. F., Jones, O. D., Lambeth, S. P., Mareno, M. C., Richardson, A. S., & puchin monkeys. Philosophical Transactions of Royal Society B, 363, 3837–3844.
Schapiro, S. J. (2007). Endowment effects in chimpanzees. Current Biology, 17(9), Mardani, A., Jusoh, A., & Zavadska, E. K. (2015). Fuzzy multiple criteria decision–
1704–1707. making techniques and applications—Two decades review from 1994 to 2014.
Buccafusco, C., & Sprigman, C. (2010). Valuing intellectual property: An experiment. Expert Systems and Applications, 42(8), 4126–4148.
Cornell Law Review, 96, 1–45. Morewedge, C. K., & Giblin, C. E. (2015). Explanations of endowment effect: An in-
Buede, D. M., & Maxwell, D. T. (1995). Rank disagreement: A comparison of multi- tegrative review. Trends in Cognitive Sciences, 19(6), 339–348.
-criteria methodologies. Journal of Multi-Criteria Decision Analysis, 4(1), 1–21. Newnan, D. G., & Johnson, B. (1995). Engineering economic analysis (5th ed.). San
Chang, C. H., Lin, J. J., Lin, J. H., & Chiang, M. C. (2010). Domestic open-end equity Jose, CA: Engineering Press.
mutual fund performance evaluation using extended TOPSIS method with dif- Novemsky, N., & Kahneman, D. (2005). The boundaries of loss aversion. Journal of
ferent distance approaches. Expert Systems with Applications, 37(6), 4642–4649. Marketing Research, 42, 119–128.
Chen, S. J., & Hwang, C. L. (1992). Fuzzy multiple attribute decision making: Methods Olson, D. L. (2004). Comparison of weights in TOPSIS models. Mathematical and
and applications. Berlin/Heidelberg/New York: Springer-Verlag. Computer Modeling, 40, 721–727.
Chen, Y., Kilgour, D. M., & Hipel, K. W. (2011). An extreme-distance approach to Opricovic, S., & Tzeng, G. H. (2004). Compromise solutions by MCDM methods: A
multiple criteria ranking. Mathematical and Computer Modelling, 53, 646–658. comparative analysis of VIKOR and TOPSIS. European Journal of Operational Re-
Garvey, P. R. (2008). Analytical methods for risk management: A systems engineering search, 156, 445–455.
perspective. New York: CPR Press. Raiffa, H. (1968). Decision analysis. New York: Random House.
Hammack, J., & Brown, G. M. (1974). Waterfowl and wetlands: Toward bioeconomic Stewart, T. J. (1992). A critical survey on the status of multiple criteria decision mak-
analysis. Baltimore, MD: Johns Hopkins Press. ing theory and practice. Omega, 20(5–6), 569–586.
Harbaugh, W., Krause, K., & Vesterlund, L. (2001). Are adults better behaved than Thaler, R. H. (1980). Toward a positive theory of consumer choice. Journal of Eco-
children? Age, experience, and the endowment effect. Economic Letter, 70, nomic Behavior and Organization, 1, 39–60.
175–181. Thaler, R. H., & Sunstein, C. R. (2008). Nudge: Improving decisions about health,
Hearth, D., & Zaima, J. K. (2004). Contemporary investment: Security and portfolio wealth, and happiness. New Haven, CT: Yale University Press.
analysis. Mason, OH: Thomson/South-Western. Tversky, A., & Kahnneman, D. (1981). The framing of decisions and the psychology
Horowitz, J. K., & McConnell, K. E. (2002). A review of W TA/W TP studies. Journal of of choice. Science, 211, 453–458.
Environmental Economic Management, 44, 426–447. Tzeng, G. H., & Huang, J. J. (2011). Multiple attribute decision making: Methods and
Hwang, C. L., & Yoon, K. P. (1981). Multiple attribute decision making: Methods and applications. New York: CRC Press.
applications. Berlin/Heidelberg/New York: Springer-Verlag. Yoon, K. P., & Hwang, C. L. (1995). Multiple attribute decision making: An introduction.
Ishizaka, A., & Nemery, P. (2013). Multi-criteria decision analysis. New York: Wiley. Thousand Oaks, CA: Sage.
Kahneman, D. (2011). Thinking, fast and slow. New York: Farrar, Straus and Giroux. Zanakis, S. H., Solomon, A., Wishart, N., & Dublish, S. (1998). Multi-attribute deci-
sion making: A simulation comparison of select methods. European Journal of
Operational Research, 107, 507–529.

You might also like