practice  users of e commerce. The basic idea of DEA is multi input and multi output oriented 

efficiency evaluation without any further assumptions about the structure (e.g., normal 

distribution) or side conditions. Unlike parametric methods, DEA can use all kinds of input and 

output data to analyze the production behavior. The DEA model used was non input  or output 

oriented because neither an input minimizing (input oriented) nor an output maximizing (output 

oriented) analysis was necessary to evaluate the observed, actual input/output relation identified 

in the survey. Moreover, the model assumes returns of scale for each DMU depending on the 

size and a concave function of decreasing returns. The software used for the data analysis 

together with a detailed description is available from Scheel (2000). 

DEA was chosen due to the unique alternative way of analyzing a set of data in comparison to 

the best performing data sets. A regression analysis, for example, only describes the deviation of 

best performing data sets from the average. Unlike parametric approaches, DEA optimizes on 

each individual observation independent of any distribution assumptions (Charnes et al., 1994, p. 

5; Cooper, Seifert & Tone 2003, p. 13). Different kinds of DEA models have been used in a large 

number of ways to measure the impacts of IT, e.g., in the banking industry (Barr et al. 2002), the 

manufacturing industry (Beck, Wigand & Konig, 2004), or in the distribution industry (Beck, 

Konig & Wigand, 2003). 

In this paper, the DEA was used as follows: As input variables for the DEA model, the results of 

seven questions are used (Table 16), measuring the number of e commerce technologies in place 

as binary variables. The variables are coded as 0 when an establishment uses the e commerce 

technology and 1 if it does not use it. The coding is equivalent to more costs of input when e 

commerce is not available or the other way around, i.e., firms using e commerce gain benefits by 

reducing their processing costs. 

Input variables (Internet usage indicator) = u (online advertising, online sales, online procurement,  , same formal 

business processes along supply chain)   

 s.t. 

v  

i

{ }

1

;

0

The ten output variables of the model are measured by a five point scale (Table 24) with 1 (no 

impact at all) to 5 (a great deal). The DEA model uses a linear program to analyze the ratio 

between low costs of input (using e commerce) and the resulting satisfaction output for each 

establishment. As a result, the DEA identifies the best practice cases or the most efficient 

establishments within the sample. Firms on this so called  efficient frontier line  are relatively 

more efficient users than other firms below the frontier line. For a better explanation of the 

results, the average of  efficient  and  inefficient  establishments was calculated. The seven 

input variables are aggregated to an Internet usage indicator, while the ten output variables are 

aggregated as an average e commerce satisfaction index. The input variables are used 

unweighted so that each e commerce technology has the same explanation weight or loading for 

the efficiency of DMUs. 

Output variables (E commerce satisfaction index) = v (internal process more efficient, staff productivity increased, 

 , competitive position improved)   

 s.t. 

u  

j

{

;

1

;

3

;

2

}

5

;

4

8