189 / 330
臺大管理論叢
第
27
卷第
2
期
189
handling the multiple inputs and outputs of DMUs makes DEA an attractive alternative for
evaluating the performance of different suppliers. One of the early works that used DEA to
measure the efficiencies of alternative suppliers is the study of Braglia and Petroni (2000),
which demonstrated an application of the method using actual data obtained from a firm
operating in the bottling industry. Their study also showed that DEA is capable of handling
multiple conflicting attributes inherent in supplier selection while simultaneously trading
-
off
key selection criteria. Liu, Ding, and Lall (2000) presented a DEA model that aims at
selecting a supplier with greater diversity so that the number of suppliers can be reduced.
Narasimhan, Talluri, and Mendez (2001) adopted DEA to evaluate suppliers for a multi-
national corporation in the telecommunications industry. The method proposed by Garfamy
(2006) for measuring the overall performances of suppliers is developed from the concept of
ownership total cost, whereas Ross, Buffa, Dröge, and Carrington (2006) used DEA to
evaluate suppliers with respect to attributes of both buyers and suppliers. Talluri,
Narasimhan, and Nair (2006) presented a chance-constrained DEA approach to evaluate the
performance of suppliers in the presence of stochastic performance measures. Saen (2007)
proposed using an imprecise DEA to evaluate suppliers in the presence of both quantitative
and qualitative data.
Using DEA to evaluate and select suppliers involves several issues. First, the efficiency
of suppliers has to be defined. Different definitions of supplier efficiency were considered in
the literature, such as simple production efficiency, comparative efficiency and super
efficiency (Narasimhan et al., 2001). To avoid selecting a sub-optimal or false-positive
supplier, cross-efficiency index may also be used (e.g., Braglia and Petroni, 2000; Forker and
Mendez, 2001). Second, the decision-makers must specify multiple input and output criteria.
Distinct sets of input and output criteria were considered in previous studies on supplier
selection. Third, some DEA models require assumptions that may not hold in practice. For
instance, the CCR model, proposed by Charnes, Cooper, and Rhodes (1978), was developed
with the assumption of constant returns-to-scale, but this assumption can hardly be satisfied
in practical applications.
2.2 Mathematical Programming
Previous research formulated the supplier selection problem using various types of
mathematical programming models, such as LP, mixed integer LP, mixed integer non-linear
programming, goal programming, and multi-objective programming. Talluri and Narasimhan
(2005) proposed a LP model for evaluating and selecting potential suppliers with respect to