Segmenting tourists by length of stay using regression tree models

Purpose: This study analyses how the socio-demographic profile of the tourist, trip-related characteristics, distance, and economic conditions in the source country affect pleasure tourists' length of stay behaviours in Barbados. Design/methodology/approach: The study uses “biggish” data (over 3.6 million observations), parametric models (OLS) and statistical learning models (regression trees) to develop a length of stay decision rule to segment pleasure tourists' length of stay. Our sample period is January 2004 to March 2013. Findings: The analysis revealed a great deal of heterogeneity in the impact of the predictors across segments, which would be typically hidden from simple parametric approaches often used to model length of stay (such as OLS). Practical implications: The main implication is that conventional models of length of stay should be complemented with segmentation analyses to shed some light on the heterogeneous length of stay behaviours of specific market segments. Originality/value: Many studies on small tourism-specialising states focus on modelling aggregate arrivals. By modelling micro-data for Barbados, we provide insights on this aspect of tourism demand for small states. Second, very few studies use classification tools to analyse length of stay. The study contributes to the literature through its methodological approach.