Log in / Register
Home arrow Economics arrow Bounded rational choice behaviour
< Prev   CONTENTS   Next >

3.3.4. Choices of Packaged Tours

Zhang, Hakoda, and Fujiwara (2008) introduced relative utility into a paired combinatorial logit (PCL) model (r_PCL model: Eqs. (3.21a)—(3.21c)), which represents both observed and unobserved inter-alternative similarities.


(3.21b) (3.21c)

where, indicate choice alternatives; refers to a similarity parameter between a pair of alternatives s and ); and are defined in Eq. (3.4) or (3.7).

One can see that observed similarities across alternatives are represented by and unobserved similarities are described by asq. A case study was conducted using data from two stated tour package choice experiments with respect to 331 Japanese tourists (1655 SP responses) and 325 Korean tourists (1610 SP responses), focusing on tourism by their own cars on the Asian Highway.[1] The surveys were conducted in Japan in December 2005 and in Korea in January 2006. Four packaged tour plans were prepared as a choice set in both countries, which are defined as combinations of major destinations (Japan: only the western area (Osaka, Hiroshima, and Fukuoka); Korea: Seoul, Busan, and Gyeongju). Attributes include visit sequence, sojourn time, visiting timing, and stopping behaviour during travel between destinations when visiting two or more destinations. It is assumed that Japanese (Korean) tourists visit Korea (Japan) by driving their own cars on the Asian Highway for a four-night and five-day tour together with their friends/acquaintances (in total, four persons, among which two have a driving license).

Three models were built: r_MNL model, PCL model, and r_PCL model. In case of Japanese tourists, adjusted McFadden's rho-squared values are 0.097 for the r_MNL model, 0.101 for the PCL model (4% higher than that of the r_MNL model), and 0.106 for the r_PCL model (5% higher than that of the PCL model). In case of Korean tourists, adjusted McFadden's rho-squared values are 0.087 for the r_MNL model, 0.096 for the PCL model (10% higher than that of the r_MNL model) and 0.112 for the r_PCL model (17% higher than that of the PCL model). These results suggest that introducing unobserved similarities is more effective to improve model accuracy than introducing observed similarities, and the joint representation of both observed and unobserved similarities is much more effective. As expected, the r_PCL model is better than the PCL model, but it is also worth noting that the model accuracies of r_MNL and PCL models are not remarkably different.

As for relative interest parameters, Japanese tourists attach the highest importance to (or show the highest interest in) the combination 'Seoul-Gyeongju-Busan' and Korean tourists to (in) the combination Osaka-Hiroshima-Fukuoka'. This indicates that both Japanese and Korean prefer to visit as more destinations as possible. Introducing the unobserved similarities results in the reversal of relative importance of different tours for Korean tourists.

  • [1]
Found a mistake? Please highlight the word and press Shift + Enter  
< Prev   CONTENTS   Next >
Business & Finance
Computer Science
Language & Literature
Political science