Developing a long- term global tourism transport model using a behavioural approach: Implications for sustainable tourism policy making

Tourism emits 5 per cent of anthropogenic carbon dioxide emissions of which about 75 per cent is caused by tourism transport (Scott et al., 2010). If historic developments continue, it will be very difficult for tourism to significantly reduce its emissions to a sustainable level (Scott et al., 2010). Technology-based efficiency improvements have so far been outpaced by volume and demand growth (Ch’eze et al., 2011; Owen et al., 2010; Sgouridis et al., 2010). Therefore, changes in demand and travel behaviour will be inevitable to achieve sustainable tourism development with respect to climate change. Further, most current tourism studies cover only international trips, just 16 per cent of all global tourism trips (Peeters and Dubois, 2010). Finally, a long-term horizon is needed, up to at least the year 2100 in most climate scenarios (Girod et al., 2009; Girod et al., 2012; IPCC, 2000; Rogelj et al., 2011) and even up to 2300 (Moss et al., 2010). The main reasons for such a long-term focus in climate change scen arios are the “long-term (decades to centuries) trends in energy-and landuse patterns” and because of “the slow response of the climate system (centuries) to changing concentrations of greenhouse gases” (Moss et al., 2010, p. 748). Most existing tourism demand models (Lim, 1997) and many tourism scenario studies cover only time horizons of 15-20 years (e.g. Forum for the Future, 2009; Schwaninger, 1984; UNWTO, 2011; WTO, 1998). Only a few studies take wider horizons, all dedicated to tourism and climate change (Ceron and Dubois, 2007; Mayor and Tol, 2010; Müller and Weber, 2007). Suitable systembased models for global tourism do not exist. Econometric models find increasing validity problems when describing longer-term futures; the coefficients defining such models are statistically derived, but not necessarily founded in the real world mechanisms of behaviour. This chapter’s goal is to create a tourism travel behaviour model founded in system dynamics, product diffusion and psychological mechanisms. System dynamics can model systems that lack data, proven theoretical foundations and need longer simulation periods (Sterman, 2000). The chapter explores a way to develop a novel tourism behaviour model that describes travel behaviour in terms of trips and distances travelled per transport mode at the global scale. Tourism’s CO2 emissions are, for a given level of technology, determined by trip numbers, distances travelled and transport mode (Peeters and Dubois, 2010). Therefore the model must provide estimates of trip numbers and distances per transport mode. Important model inputs are travel cost, travel time, income distribution, GDP/capita and population. Secondary inputs are transport infrastructure and technology that will affect both travel cost and travel time. The behavioural model has been created and tested with a dynamic version of the Global Tourism Transport Model (GTTMdyn). Two versions of GTTMdyn preceded the dynamic version: a basic version, GTTMbas programmed with Excel, with linear extrapolations and an advanced version, GTTMadv, programmed in Powersim Studio (version 7), mainly based on linear projections but with automatic scenario generation capabilities used for back-casting towards certain emission goals. The GTTMbas and GTTMadv models are described by Dubois et al. (2011) and Peeters and Dubois (2010). The ultimate goal of the GTTMdyn model is to provide insights into the impacts of tourism on greenhouse gas emissions and the effectiveness of policies to mitigate those emissions. The model will cover the period up to the year 2100. A consequence of that long time span is that we will need to calibrate the model over a similar period, i.e. from 1900 to 2005. The model must be able to handle the development of a completely new transport mode, civil air transport, that became available from c. 1920 (Ananthasayanam, 2003). Furthermore, GTTMdyn should be able to handle a wide range of policies governing travel cost, time or speed, infrastructure capacity and psychological factors in decision-making processes of tourists (Schäfer, 2012). The long-term and global character of GTTMdyn forms a challenging combination to the behavioural (demand) part of the model (Schäfer, 2012). A common approach in transport modelling is the “four-stage” model (Bates, 2008). The stages are trip generation, trip distribution, modal split and assignment to the grid or infrastructure. In GTTMdyn we need the trip generation, distribution and mode-choice stages, but not the grid assignment stage, as detailed global networks are not defined in the model for the main transport modes. In most transport models trip generation is a function of population characteristics including income, age, household and trip properties such as motive. Generalised cost (a combination of cost and monetised travel time and sometimes discomfort) is ideally taken into account, but often ignored (Bates, 2008). Trip distribution and modal choice generally are modelled as (multinomial) logit models (Bates, 2008). Multinomial logit models are used in many studies for tourism demand (Huybers, 2003; Lyons et al., 2009; Nicolau, 2008) and tourism transport demand (Bieger et al., 2007; Pettebone et al., 2011). Such models determine the probability of choice for each alternative using an exponential function of utility (Morley, 1994; Papatheodorou, 2006). Another line of modelling is based on the use of constant elasticity for travel cost and travel time (Schäfer, 2012). Schäfer (2012) shows that most large-scale transport models use a constant elasticity of substitution (CES) or price elasticity as the basic demand function, and in some cases, additionally, a logit type of model to govern distribution of trips over transport modes. Distances are generally determined from distances between (world) regions as given by Schäfer (2012, p. 31). The problem with elasticity-based models is that elasticities are more a statistical artefact than a factor that represents any specific “psychological” behaviour. Elasticities differ when taken over different time periods and general validity is low which is shown by the very wide range of values obtained from different studies for the same kind of behaviour, e.g. choice between air and car transport (Oum et al., 2008). The kinds of modelling described above are founded in the standard economic model (SEM). The main axioms of SEM are that economic agents make rational decisions, are motivated by utility maximisation, are purely selfish, ignore the impact on others’ utility, are Bayesian probability operators, have consistent time preferences (i.e. the discount rate is constant over time) and consider all income and assets to be completely fungible or freely interchangeable (Wilkinson, 2008, p. 5). Mounting criticism of SEM claims that almost none of the above axioms seems to be valid in the real world and result in different strands of thinking like behavioural economics (Wilkinson, 2008) including prospect theory (Kahneman and Tversky, 1979), evolutionary economics (Dopfer, 2005) and ecological economics (Daly and Farley, 2004). It seems risky, specifically in the context of a systems model for a long-term analysis, to ignore known discrepancies in human economic behaviour. Therefore, the behavioural model of GTTMdyn has been founded on insights from prospect theory (Kahneman, 2011; Kahneman and Tversky, 1979) as will be further elaborated in the next section.