Page 232 - 360.revista de Alta Velocidad - Nº 6

Kim, Junghwa. Li, Yeun-Touh. Schmöcker, Jan-Dirk.

Taiwan with a population of around 6.6 million at the end of 2012. The station was dominated
by all kinds of travelers, including business travelers as well as local and some foreign tourists.
Zuoying (located in Kaohsiung city) and Taichung station represent the 2 and 3 largest
nd
rd
metropolitan areas located in the south and middle of Taiwan’s west coast respectively. The
rest of the stations listed in order of ridership are Hsinchu, Taoyuan, Tainan, and Chiayi, which
are considered as second or third-tier city, from the 3 group. Note that Banciao is located in
rd
the Taipei metropolitan area as a satellite station dispersing the mass demand from Taipei.
Therefore two cities, Taipei and Banciao were considered together as one region in this study.
We consider regional heterogeneity mentioned briefly in above and assumed that regional
heterogeneity would measured by some socioeconomic variables such as population, car
ownership (%), total length of road (km), fuel price and business scale by city as underlying
independent variables (X ) (. These were obtained and calibrated into the regional scale
itk
of monthly data from 2007 to 2015. For brevity we omit descriptive statistics as shown in
Figure 2.

The region of Taipei and Banciao shows the highest value of population and business scale
while the lowest ratio of car ownership compared to other cities. This might come from
Taipei being the capital of Taiwan. The highest road supply rate is shown in Taichung with a
high ration of car ownership. This implies that better mobility by car use would be in case
of Taichung even this is located in the middle of Taiwan. In addition, regarding explanatory
variables which explain THSR station demand, Li et al. (2015) we also considered two time
specific factors, including a variable for years (X ), which shows time since operation after
y
THSR opening, in order to reflect the trend of increased demand over time and a dummy
variable for summer vacation (D ), to reflect some induced demand such as leisure trips.
s
Thus the variables for regional heterogeneity also have a form of panel data which has a
double subscript, index i and time t as same as variables of accessibility.

Figure 2. Descriptive statistic of socioeconomic variables by major city

4. Methodology

This study uses a two-way error component model, a kind of panel model, to mitigate the
errors occurred from unobserved heterogeneity in terms of location regions and time. A two-
way error component random effect model was applied in this study where we control the

230 360.revista de alta velocidad