Making Modelling Worthwhile: Revisions to the Modelling Framework
M Logie, Minnerva Ltd, UK
This paper considers the topic of the cost-effectiveness of transport modelling. It analyses the issues that determine the value of modelling and the costs of developing and using such models. The paper puts forward a modelling framework that is considered to offer improvements over arrangements that are currently used. Features of this new framework include a tighter integration of data and models, that is, the distinction becomes less clear than usual; a greater emphasis of cultural and consumer issues in modelling, with ?transport? viewed to a greater extent as one product amongst other consumer products; improved approaches to handling variations with time in modelling; and an emphasis on communicating the results of modelling in varied ways.
The paper reflects experience on a number of projects connected with transport data collection, model development, and research into methodologies. Some of the ideas on data have been examined in work for Transport for London, but are due for further research and development in the 3-year OPUS project starting early 2003. The ideas on modelling largely identify weaknesses that point the way to future work, while the ideas on communication of results are exemplified in new web-based software that has recently been developed.
The case for transport modelling is best made when a policy maker or network manager makes a decision because of insights gained from the modelling. This happy state of affairs does arise, but only infrequently, and the majority of the models are developed because procedures that release money demand their use. These procedures see models as a means of introducing an element of rational and systematic analysis into a process, transport planning, that is otherwise prone to the vagaries of political and institutional decision-making. Although small in price relative to the total costs of the decisions that they are used to support, the production of models requires significant funds and their value can be questioned, not least by those who only use them simply to meet external requirements. This situation is most clearly demonstrated in the case of the British multi-modal studies that emanated from the UK Government?s 1998 White Paper on Integrated Transport and subsequently enacted into legislation. These studies have now covered the majority of the population in England (and Scotland) at a cost around £20m. The resulting proposals from the multi-modal studies have met with a mixed response, with limited marks awarded by most observers for insights gained.
The multi-modal studies provide good, but far from unique, examples of the complexity of transport matters for which the systematic analysis of modelling is clearly required, yet the value of the models when translated into decision making is more questionable.
A more technical issue that the paper also considers relates to the nature of the solutions provided by modelling. Historically, transport models have sought static equilibrium solutions and, significantly, most evaluation and assessment frameworks require this form of solution. Indeed, it poses problems for decision makers when presented with more complicated solution forms. However, prodigiously powerful computing now makes more complex and dynamic solutions more commonplace, typically generated from simulation-based modelling. The issue for modelling now becomes how to understand and make use of these more complex solutions.
An Improved Modelling Framework
The modelling process is naturally decomposed into three stages of:
1) data preparation and input,
2) modelling (solution generation, including traveller choice making and matching transport demand to transport supply),
3) presentation and analysis of results, including evaluations and appraisals.
The improvements proposed by the paper affect each of these stages in various ways.
The most expensive element of modelling is usually associated with data, unless it is fortuitously available from another study. The specification of modelling is often strongly conditioned by the availability of data, which places major constraints on what the modelling may achieve. The paper proposes that greater emphasis is placed on synthesised data and, moreover, that data is synthesised at a high degree of disaggregation. For synthesised data to be meaningful, the synthesis process must be based on real observations. The key to the methodology is to use data that is inexpensive to obtain. The widespread use of trip matrix estimation techniques that exploit readily-obtained travel flow counts is an example of the process, but the proposal is to extend this concept. In part, this is done through methods that are better able to ?accumulate knowledge?. This approach has the merit that it is easier to use cheaper, small scale surveys, rather than costly major one-off surveys. One way of accumulating knowledge is through mathematical modelling, which may be viewed as a concise way of representing otherwise voluminous data. Developments in Markov Chain Monte Carlo (MCMC) simulation methods provide a powerful means of encapsulating all manner of data. Another set of statistical processes, namely Bayesian-type methods, provide a natural means of improving the accumulated knowledge in the light of new information. Maximum Likelihood methods, as used in matrix estimation and elsewhere, can also contribute to the process.
The consequence of these methods is that it is possible to conceive of synthesised populations (of would-be travellers) whose details, through synthesis, are known. This allows details to be introduced into transport modelling that current modelling frameworks must, perforce, omit. An important set of details relate to cultural and consumer issues. These are reflected in part through socio-economic categorisations that are typically used in transport modelling, but the emphasis is quite different from analysis in other fields, notably of market research. An instructive example is provided by comparing the modelling approaches used by the travel industry (tourism, business travel, etc) and by transport planners. The former has not only structured its modelling but also its operations on a deep appreciation of the different traveller types. Here ?types? reflect cultural and personal lifestyle factors, rather than travel purposes.
The standard transport modelling framework is the ?four stage? model. The paper encourages the existing tendency for the first stage, ?trip generation and attraction?, to be extended to incorporate land use and (regional) economic modelling in a more whole-hearted manner, so that the framework becomes better appreciated as a ?five stage? model. The stages of ?trip distribution? and ?mode choice? have increasingly become the province of logit-choice modelling. These techniques, while powerful, are strongly data-driven, so that the availability of much richer synthesised data considerably extends their scope. The modelling using in the data synthesis process can be exploited by these stages of the modelling framework. In particular, Bayesian methods can be used to take better account of people?s choices, given their current circumstances.
The fourth modelling stage of ?assignment? links travel demand with travel supply. The issue of balancing demand when it exceeds the capacity of the transport network infrastructure has been approached by a number of methods, including recent UK Department for Transport research work. These methods include, implicitly or explicitly, the notion of ?equilibrium?. As previously intimated, static equilibrium is helpful to decision making (recalling that this is the purpose of transport modelling), but there is a growing awareness and application of dynamic modelling techniques. One approach is to aim for solutions corresponding to a ?dynamic equilibrium?. This term is close to an oxymoron and explains why satisfactory mathematical definitions have been hard to find. Progress has recently been made, which opens the way for improvements to this aspect of modelling.
Although advances can be made in equilibrium methods, it is not-infrequently observed that the notion of equilibrium is misplaced in modelling the real world that is continuously subject to systematic and random changes. This is particularly the case when simulation modelling is used, as simulation modelling does not typically concern itself with notions of equilibrium solutions. The paper expresses the view that non-equilibrium solutions are antithetical to good decision making as non-equilibrium implies unstable and undependable solutions. An approach is proposed in to this situation in which greater notice is taken of experience from non-linear dynamics where random processes can be found to have either stable or unstable properties, depending on the settings of parameters that define the processes.
While stable solutions may help in presenting and understanding results, time-varying solutions can still pose difficulties to understand, although they frequently make attractive animated screen displays. The modelling framework needs to match dynamic information with unambiguous results for the decision maker to use. The paper proposes that results are understood in the form of functions that may be integrated over (future) time.
The value of a model is severely constrained if its results cannot be communicated effectively to those taking decisions. The modelling framework needs to embrace new technologies to disseminate results. The paper presents a system for presenting transport information using web browsers coupled with GIS technology not only to provide widespread access to information, but also to offer it in context according to policy interests and past performance as provided by monitoring studies.
The paper sets out a comprehensive agenda for changes in the modelling framework. The changes from present practice are considerable, but each of the changes is impelled by existing and exciting developments, which means that the realisation of the changes is within the bounds of practicality.
Association for European Transport