Correspondence: Hai Jiang, 3150 Sabre Drive, Southlake, Texas 76092, USA. E-mail: hai.jiang@sabre.com

¹is a Senior Analyst in the Research Group at Sabre Holdings. His research interests include consumer behavior modeling and large-scale optimization. He received a PhD from MIT in 2006 and was the winner of the Anna Valicek Medal for Innovative Research in the area of Airline Operations Research from the Airline Group of the International Federation of Operational Reseach Societies (AGIFORS).

Received 22 September 2008; Revised 22 September 2008; Published online 16 January 2009.

Abstract

Understanding the relationship between itineraries displayed on an air travel website's shopping screens and the website's market share has never been so important nowadays. This paper presents an approach to measure 'screen quality', that is, the overall desirability of itineraries displayed by a website. At the centre of this approach is a nested logit model that generates a screen quality metric called 'selection probability' by comparing the screens of the host website to those of a reference website. This metric measures the probability an individual would select an itinerary on the host website. Recognising the fact that there are unobserved factors across websites that influence shopping behaviour, we introduce website-specific constants to capture the average of those effects, which directly translates our 'selection probability' to market share prediction. We present three models and conduct experiments to evaluate their performance. Results show that the model using shopping, booking and screen data simultaneously in calibration generates satisfactory prediction results.