87 /342
臺大管理論叢
第
27
卷第
2S
期
87
majority of taxi drivers are alone in the street looking for passengers, with a lack of
necessary information for planning optimal routes unless they join a taxi fleet. Hou (2010)
described six innovative work practices and service inventory types based on a case study of
driver’s behavior in using an on-the-air queuing system of Taiwan Taxi Fleet.
3. Research Method
The on-the-air queuing system of Taiwan Taxi Fleet identify areas of high demand for
taxi service, and classify them as virtual scheduling points by analyzing historical data from
geographic information systems, and dispatch customers’ taxi request information to drivers
who are registered in the system. In the system, each vehicle will return a set of GIS
information to the dispatch center about every 30 seconds to a minute to show the current
state of the vehicle. This study collected data from the random sampling of 300 taxi drivers,
with 12 day operation data which accumulated to 4.44 million records. Two-stage data
analysis was used in this study. In the first stage, we used the MYSQL database to process
the huge amount of raw data so that it could be manipulated by relational data logic. Some
Java programs were also developed for data transform and analysis.
At the second stage, this study performed hierarchical and non-hierarchical cluster
analysis by using SPSS statistical software. Hierarchical cluster analysis is used to determine
the proper number of groups which can represent different types of driver behavior. This
group number also serves as the initial input for the non-hierarchical cluster analysis; here
we adopted the K-means algorithm to assign observations into respective groups. Based on
the two-stage analysis, this study obtained different types of taxi service inventory to
calculate their operational performance metrics for investigating the correlation between
them. Besides analyzing data collected from the on-the-air queuing system, this study also
collected 300 samples of the driver's age, education level, driving experience, the time of
joining the team and the willingness to post advertisements on the car body. Those data were
used for conducting a demographic analysis.
4. Data Analysis
According to the sampling size guideline suggested by Dillman (2000), the sample size
should be larger than 295, giving a confidence level of 95%; 300 observations were drawn
randomly from the population of about 3000 taxi drivers. The data profile shows that the
average age is 46.07; driving experience is 5 to 10 years; route distance without carrying