“…Regarding the methodology, most previous studies used multinomial logit (e.g., Skinner, 1976;Harvey, 1987) or nested logit (e.g., Pels et al, 2000Pels et al, , 2001Pels et al, , 2003, while there are some that employed probabilistic choice set multinomial logit (e.g., Başar and Bhat, 2004) and weighted conditional logit (e.g., Ishii et al, 2009). In terms of data, previous work used either revealed preference data (e.g., Pels et al, 2001Pels et al, , 2003Başar and Bhat, 2004;Zhang and Xie, 2005;Hess andPolak, 2005a,b, 2006;Ishii et al, 2009;Pels et al, 2009), stated preference data (e.g., Skinner, 1976;Harvey, 1987;Proussalogloua and Koppelman, 1999;Hensher et al, 2001;Zhang and Xie, 2005;Hess et al, 2007;Loo, 2008;Ishii et al, 2009;Hess, 2010;Koster et al, 2011;Marcucci and Gatta, 2011;de Luca, 2012;and Drabas and Wu, 2013), or mixed data (Ortúzar, J. de D. and Simonetti, 2008). 3 In terms of methodology, the study that is closest to ours is the demand side of Berry and Jia (2010) who estimate a structural model with aggregate data.…”