e-Review of Tourism Research (eRTR), Vol. 16, No. 1, 2018 http://ertr.tamu.edu 38 Chi Fung Lam The Chinese University of Hong Kong Jian Ming Luo City University of Macau Museumpark Revisit: A Data Mining Approach in the Context of Hong Kong It is important for tourism managers to understand the synergistic effects between museums in tourism management. This study raises is concerned with the "museumpark" and re- examines the synergistic effects among museums in Hong Kong. Utilising online photos from social media, over 2500 photos from more than 300 visitors were collected and analysed quantitatively. Through limited dependent variable regressions, the results show that once a visitor visits one museum in the "museumpark", the probability that this visitor will visit another museum nearby increase significantly. Based on the result of this study, museum researchers can further investigate the precise synergistic effects between museums. Keywords: Museumpark; Museum Demand; Spill-over Effects; Data Mining Chi Fung Lam Department of Finance The Chinese University of Hong Kong Hong Kong Email: kelvinlam@cuhk.edu.hk Jian Ming Luo Faculty of International Tourism and Management City University of Macau Avenida Padre Tomás Pereira Taipa, Macau. Phone: +853 8590 2535 Email: kenny.luo@connect.polyu.hk; kennyluo@cityu.mo Jian Ming Luo is an Assistant Professor in the Faculty of International Tourism and Management at City University of Macau. His recent publications have appeared in the highly prestigious international academic journals. http://ertr.tamu.edu/ mailto:kelvinlam@cuhk.edu.hk mailto:kenny.luo@connect.polyu.hk;%20kennyluo@cityu.mo e-Review of Tourism Research (eRTR), Vol. 16, No. 1, 2018 http://ertr.tamu.edu 39 Background The concepts of “industrial district”, “cluster” and “museumpark” are very similar. Becattini (1990) defined industrial district as “a socio-territorial entity with is characterized by the active presence of both a community of people and a population of firms in one naturally and historically bounded area” while Porter (2000) defined clusters as “geographic concentrations of interconnected firms, specialized suppliers, service providers, firms in related industries, and associated institutions in a particular field that compete but also co-operate”. Despite the lack of a formal definition of “museumpark”, in 1996, Jansen-Verkebe and Rekom studied a “museumpark” or “museumquarter” in Rotterdam. The “Rotterdam Museumpark” included a museum for modern architecture, a museum of natural history, a local art museum, and a gallery for temporary art exhibitions, and a fine arts museum and all of them were within walking distance (Jansen-Verkebe & Rekom, 1996). The authors argued that physical clustering would generate spill-over effects, such as synergistic marketing strategies, sharing common resources, and providing convenience. Yet, there is a paucity of research which tests this hypothesis and the significance of spill-over due to the “museumpark” is still unknown. This study provides a preliminary estimation of the effect of spill-over. The primary research question is centred on how likely visitors are to visit further museums after having visited one. While Hong Kong has not advertised a “museumpark”, there are several museums located within walking distance of each other in Hong Kong, i.e., the Hong Kong Space Museum, the Hong Kong Museum of Art, Hong Kong Science Museum, Hong Kong Museum of History, the Hong Kong Heritage Discovery Center and the Hong Kong 3D museum. Although Jansen-Verkebe and Rekom (1996) did not provide a formal definition of a “museumpark”, this area, spanning from Tsim Sha Tsui East to Tsim Sha Tsui, fits the criteria of “industrial district” defined by Becattini (1990) and “cluster” defined by Porter http://ertr.tamu.edu/ e-Review of Tourism Research (eRTR), Vol. 16, No. 1, 2018 http://ertr.tamu.edu 40 (2000). This research does not attempt to provide any formal definition, but to direct research interest on formalizing the definition. Empirical strategy and results This study applied P-DBSCAN (Kisilevich, Mansmann, & Keim, 2010) clustering technique to the initially collected dataset. The initial data collection process focused on photos from social media platform Flickr tagged with the keyword "museum." The choice of this social media platform was based on convenience, as Flickr provides free access to many important data, such as photos, locations, and timestamps (Vu, Luo, Ye, Li, & Law, 2018). We set the bounding box for the photo function with parameter values as , , and to cover the entire Hong Kong geographical area. The search covered a ten-year period from 2007 to 2016. All photos, whose tag field contained the provided keyword and taken time within the specified time frame, were included. The search returned 2,843 photos about the museum from 349 visitors. The museums included were the Space Museum, Museum of Art, Museum of History, Science Museum, and Heritage Museum as other museums did not yield sufficient data. After examining the locations, the clusters were marked using the names of the corresponding museums. Since the location between the Museum of Art and the Museum of History is relatively close, the location cannot be distinguished from the data. Therefore, we considered them as one attraction. All of the museums were located in Tsim Sha Tsui and Tsim Sha Tsui East area. Since some visitors can visit multiple museums at the same time, the total number of visitors can exceed 349. There were 321 visitors who visited one museum, 25 visitors who visited http://ertr.tamu.edu/ e-Review of Tourism Research (eRTR), Vol. 16, No. 1, 2018 http://ertr.tamu.edu 41 two museums, and three visitors who visited three museums. Table 1 provides a summary of visit frequency for one and two museums. Table 1: Frequency of Museum Visits Space Art History & Science No. of Visitors Visit 1 Museum 59 141 121 321 Visit 2 Museums 17 19 14 50 To analyze how prior visit affects the likelihood or probability of visiting other museums around the area, this study employed both a logit and a probit regression model. Since the dependent variable was a limited dependent variable, multiple linear regression would be subject to an out of boundary problem, i.e., the predicted value would be outside the 0 and 1 range, and heteroskedasticity problem (Kennedy, 2003). Despite the beautiful property of Ordinary Least Square (OLS) that it was robust to violations of some of the assumptions, OLS was not robust to the limited dependent variables unless the mean of the dependent variable was close to 0.5 (Walsh, 1987). Clearly, according to Table 1, the data in this study did not exhibit this property. The more appropriate way to study the relationship was to use a probit or logit model. A general linear model for limited dependent variable could be characterized where H(.) was the cumulative density function. When it was a logit model and when http://ertr.tamu.edu/ e-Review of Tourism Research (eRTR), Vol. 16, No. 1, 2018 http://ertr.tamu.edu 42 it was a probit model. For a given set of data, the likelihood function was The dependent variables here were whether the visitor visited one of the museums, i.e., Yi=1, if the visitor visited the corresponding museum and zero otherwise. Similarly, for the independent variables, Xi = 1, if the visitor visited the corresponding museum and zero otherwise. According to Kennedy (2003), as computational becomes relatively “cheap”, both logit and probit model become the most common model to address latent or limited dependent variables. Since we did not have a prior hypothesis on how the visitation spread from one museum to another, we investigated three possibilities. Particularly, The parameters were estimated using maximum likelihood estimation with R. The result of the two models are presented in Table 2 Table 2 Logit and Probit Regression Results Model 1 Model 2 Model 3 Logit Probit Logit Probit Logit Probit Intercept Coefficients 0.04753 0.0844 -0.8112 -0.47129 -0.6535 -0.3673 std error 0.18313 0.11471 0.1538 0.09261 0.1578 0.0961 p-value 0.795 0.462 1.34E-07 3.60E-07 3.46E-05 0.000132 Space Coefficients 2.3148 1.28169 2.6468 1.4052 std error 0.3423 0.19048 0.4076 0.2099 p-value 1.36E-11 1.71E-11 8.40E-11 2.16E-11 Art Coefficients 2.31478 1.21698 3.0158 1.666 std error 0.34233 0.18064 0.3297 0.1696 p-value 1.36E-11 1.62E-11 2.00E-16 2.00E-16 History Coefficients 2.64677 1.35913 3.0158 1.69477 std error 0.40762 0.20079 0.3297 0.17175 p-value 8.40E-11 1.30E-11 2.00E-16 2.00E-16 Log Likelihood -156.477 -159.499 -198.425 -201.403 -183.097 -186.792 Chi-square 87.63 81.587 151.14 145.19 155.1 147.71 http://ertr.tamu.edu/ e-Review of Tourism Research (eRTR), Vol. 16, No. 1, 2018 http://ertr.tamu.edu 43 p-value 2.2e^-16 2.2e^-16 2.2e^-16 2.2e^-16 2.2e^-16 2.2e^-16 As Table 2 showed, all regressions, regardless of the underlying distribution, or the dependent variables, were significant at 1% significance and the signs of the coefficients, along with the Chi-square statistics, were consistent with the spill-over hypothesis that upon the visitation of one or more museums, the probability of the visitors visiting another museum increases. Limitations and Conclusions The methodology has limitations, for example, due to limited data, many important variables such as the cultural background, duration of stay of the visitors, etc. are not included in the model, which can cause the endogeneity problem. However, the most important finding is that upon visiting one or more museum, visitors tend to visit more museums. This result sharply contrasts with the results from Jansen-Verbeke and Rekom (1996), where the physical clustering of museums (the “museumpark” in Rotterdam) and the visiting of one museum dis did not increase the tendency to visit another museum. To conclude, this study calls attention to researchers and highlights the need for a more precise definition of “museumpark” and further examination of the spill-over effect among “museumpark” around the world. http://ertr.tamu.edu/ e-Review of Tourism Research (eRTR), Vol. 16, No. 1, 2018 http://ertr.tamu.edu 44 References Becattini, G. (1990). 4 The Marshallian industrial. Industrial districts, 37. Jansen-Verbeke, M., & Rekom, J. (1996). Scanning Museum Visitor. Urban Tourism Marketing, Annals of Tourism Research, 23 (2), 364-375. Kennedy, P. (2003). A Guide to Econometrics. Cambridge, MA: The MIT Press. Kisilevich, S., Mansmann, F., & Keim, D. (2010). P-DBSCAN: a density based clustering algorithm for exploration and analysis of attractive areas using collections of geotagged photos. In Proceedings of the 1st international conference and exhibition on computing for geospatial research & application. Article no. 38, Bethesda, Maryland, USA. Porter, M. E. (2000). Location, competition, and economic development: Local clusters in a global economy. Economic development quarterly, 14 (1), 15-34. Vu, H.Q., Luo, J.M., Ye, B.H., Li, G., & Law. R. (2018). Evaluating museum visitor experiences based on user-generated travel photos. Journal of Travel and Tourism Marketing, 35 (4), 493-506. Walsh, A. (1987). Teaching, Understanding, and Interpretation of Logit Regression. Teaching Sociology, 15 (2), 178-183 http://ertr.tamu.edu/