Objectives:

General discussions concerning issues common to spotlightsTN four discussion lines.

Send us your
 opinion!!: info@mcrit.com

Scientists, by nature, are not creatures who commonly seek out or enjoy the public spotlight...

...our training predisposes us to avoid any kind of overt behavior that might encourage two-way communication with with the masses. Instead, we are content to pursue our truth in windowless laboratories, accountable only to members of our highly exclusive club. And although presenting papers at professional meetings is encouraged, in fact required, it's rare to find one of us holding sway to standing-room-only crowds, laughing, telling jokes and giving away trade secrets.

Even though I am a long-standing member of this club and bona fide insider myself, I cannot say that it has been my trademark to follow the rules. Acting as if programmed by some errant gene, I do what most scientists abhor: I seek to inform, to educate, and inspire all manner of people, from lay to professional. I try to make available and interpret the latest and most-up-to-date knowledge that I and my fellow scientists are discovering, information that is practical, that can change people's lives. In the process, I virtually cross over into another dimension, where the leading edge of biomolecular medicine becomes accessible to anyone who wants to hear about it. This mission places me in the public spotlight quite often"

Candace Pert, "Molecules of Emotion: The Science behind mind-body medicine"

Glossary: What's a model?

1.0 Generally speaking, a "model" is just an "algorithm" which predicts unknown data, often "forecasting" uncertain "futures". A model is an intelligent simplification of reality. The paramount modelling goal to achieve the maximum simplicity representing reality with the minimum error. Science is about discovering the simple laws governing reality.

1.1 Compared with other scientific fields (e.g. natural sciences and physics), transport modelling and social sciences in general (those fields pretending to model human behaviour), are far from having convincing explanatory theories and predictive models, and it is reasonable to doubt that never they will because understanding and predicting human behaviour is for humans themselves an ontological impossibility. For instance, social and personal experimentation involves ethical aspects which are not present in natural sciences and physics. All taken, from many decision-makers point of view, transport models not only provide poor predictions: they use obscure formulations based on oversimplified assumptions and there is no much hope future research may significantly improve the situation.

1.2 Empirical evidence shows that transport model forecasts often turn out from the actual flows, even if good practice is followed (Nielsen mentions Skamris & Flyvberg, 1996). This fact is creating since early sixties doubts of the interest of using transport models for planning purposes. Critics say that transport models use to work with poor data, apply unnecessarily sophisticated and cryptic formulations in order to get wrong results.

1.3 But, however, transport models are being developed and applied to evaluate almost all important transport policies, at urban, regional and continental scales. Most of the members of the EU have, or are developing, either national or regional transport models. While structures vary, a majority of these take the form of a traditional four stage model based on aggregate data. In some cases these are only road models, and there are often separate freight models. However, the inclusion of more features is increasing with time of day choice, mode choice, elastic trip generation and the use of tours or trip chains becoming more common. A few models, notably the Dutch National Modelling System, use a different form of transport model, a disaggregate one, which uses information on individual people and households rather than averages for zones to predict travel behaviour: This type of model is becoming more widely used particularly in Scandinavia and Italy. At the more local level, microsimulation allows individual cars to be modelling travelling in a ‘real world’ traffic environment.

1.4 The analysis of human behaviour gained scientific attention during last decade (see Himanen et alt., 1998). Linear programming models, gravity models, spatial interaction and entropy models, discrete choice models, non-linear dynamics, genetic models, agent-based models and many more. Transportation research in particular has shown the genesis of a fascinating diversity of models (Himanen et alt., 1998). Despite all these variety of paradigms, it seems that there is no alternative to substitute the classic modelling paradigm (the so-called "four steps") and all recent research developments use to be refinements, improvements, extensions and complements to this classic approach, or academic work not easily applicable on conventional decision-making processes, despite their value as knowledge effort.

1.5 Shortly speaking, models could be clusted on three major paradigms: statistically-based (then data becomes an indispensable starting point), theoretically-based (then the abstract formulation, e.g. based on scientific analogies is the starting point and data is used mostly to validate) or expert-based (e.g. rules of thumbs, heuristics... and then comparative cases and expert’s panels are key modelling procedures). In social sciences, almost any model has a component belonging to each one of these paradigms.

1.6 All considered, accurate predictions and transparent meaningful explanations alone, are not the more important model requirements (or at least not the only ones) for using transport models in decision-making processes. If the model has to be used as planning assessment tool (as a decision-making tool) it has to provide for robust results, in the sense that each run with the same input data yields to the same final results, and marginal changes in input variables do not produce huge variation. This has crucial conceptual implications, since it requires models to be deterministic (even if they include internal schocastic mechanisms) and assumes the existence and unicity of an equilibrium point.

1.7 Contrary to intuition, the "predictive" and the "explanatory" attributes of scientific models are not always coincident: Better explanations not necessarily produce more accurate predictions. Outputs from models with wrong explanatory formulations may produce better predictions (e.g. the famous Kepler formulations in relation to Newton gravitatory laws). And the opposite may also be true. Recent developments on non-linear dynamics show the actual limits of any scientific model predicting not just complex human behaviour but even much simpler physical systems. While evolutionary biology is well ranked for providing right explanations and bad ranked for predictions, quantum mechanics is in just in the opposite situation (excellent, amazing predictions but no clear explanation of why Shrodinger equations are so accurate).

1.8 An "acurate prediction" is usually obtained using statistically-based paradigms, supported by large volumes of data (it is even possible "to let the data speak for itself" and give the computer the capacity to "learn by itself"). But accuracy predicting short-term trends based on statistic adjustments not always provide meaningful explanations: There is the misperception to consider that strong correlation implies something about causal connections between the variables correlated. Furthermore, overstimating a model formulation with the available data may reduce the apparent "error" (between model outputs and data samples) but may also increase the "real" error (between model outpus and the evolution of the system being modelled, specially in the long-term).

1.9 On the other hand, a meaningful explanation is usually obtained by applying a formulation derived from a more general theoretical framework which is independent from a particular set of data. A meaningful explanation may provide less accurate predictions that a meaningless explanation. It is fair to say that a number of important policy questions are not yet solved by the transport economic theory. Fundamental behavioural hypotheses, such as considering people as rational agents having perfect information, have not being substituted yet by more explanatory hypotheses such as considering people as adaptative agents using local and temporary information to satisfy (rather than to optimase) utilitie’s threholds. The practical impossibility to consider fram-equilibrium dynamic stable solutions (instead of a single static equilibrium) restricts available theories to analyse marginal and short-term changes. Radical changes having with long-term structural impacts, such as building transport infrastructures, have insufficient theoretical support, as well as other related issues such as costs and benefits redistribution in networks, etc. In part resulting from this weak theoretical support, some fundamental transport modelling issues (such as forecasting induced traffic) use to be poorly treated by most models.

1.10 Transport models add a crucial third dimension to the prediction/explanatory dichotomy: Models have to be robust to be applied in decision-making, in the sense that the model has to produce the same outputs each time runs with the same inputs. This is crucial to make comparisons between alternatives reliable. This requirement implies, for instance, that model’s iterations have to converge towards an unique equilibrium point, and even if there are internal stochastic algorithms, the overall model has to be deterministic. In fact, the whole architecture of the classic "four-steps modelling paradigm" applied in transport modelling, was conceived to assure this goal; different modellers using the same datasets and applying the same modelling technique should get rather similar stable results (avoiding the periodic, complex and chaotic solutions which may happen even in deterministic models with marginal modifications of the initial values). Needless to say, there is a trade-off between the accuracy of predictions, the explanatory character of the formulations and its robustness.

1.11 Transport models may have three types of purposes: Strategic models (e.g. to evaluate large infrastructure projects), Tactic models (e.g. to evaluate new pricing policies), Operational models (e.g. to optimise service logistics). Each type of purpose requires a more or less detailed information, a different complexity of the model formulation applied as well as a different time horizon. Next table (from Nielsen 1999), summarises this point. While Operational models use to produce accurate results and not always convincing explanations since they are supported by large databases (on-line often) and based on statistically advanced formulations, strategic models are expected to be exactly in the contrary situation.

1.12 Transport models applied on decision-making situations have to be robust (to provide reliable and comparable outputs), and provide reasonable explanations (at least not misleading, supported by a well established general theory) and realistic predictions (according to expert judgment and available comparative cases). However, for short-term operational decisions (e.g. service optimisation), accurate predictions is also a critical goal to be achieve (and then statistically-based models become more advantageous).

1.13 Because of the required robustness of models, sensitivity analysis is an indispensable part of the modelling process. Given the likely complexity of any advanced model formulation (with continuous feed-backs and iterations of non-linear mathematic expressions) there is an obvious risk that marginal changes in inputs lead to non-stable solutions (periodic, complex or even chaotic) or simply to very different outputs. Testing the statistical reliability of more sensitive parameters and variables becomes then a crucial quality control to be carried out. Needless to say, when it takes days (or many hours) to run a model, the capacity to carry on sensitivity analysis is seriously constrained. In this cases, the use of well-known and validated algorithms is almost indispensable.

1.14 Transport models have to be run by computers, so models are software products. It is naïf to consider computers as scientifically or cultural neutral tools: they influence the way models are build and therefore the way real problems are looked (see the evolution of computers during last decades, from main frames only devote to compute large mathematic models in isolated laboratories, to personal communication devices integrating all multimedia capabilities; or the development of non-linear dynamic models). As computers and associated technologies become ever faster, it can be tempting to suppose that they will eventually become "fast enough" and that the appetite for increased computing power will be sated. However, history suggests that as a particular technology satisfies known applications, new applications will arise that are enabled by that technology and that will demand the development of new technology. This is certainly true for transport modelling where improvements in processing power have been accompanied by large, more complex models.

1.15 Traditionally, transportation analysts have been faced with the problem of utilising many different software packages, all of them having different interfaces. Some of the legacy software long in use for activities such as travel demand forecasting and air emissions forecasting still utilise what are largely script and control file based interfaces. Locally developed software for custom applications often have their own unique interfaces. More up to date packages include Windows or similar GUI based interfaces. Often, complex projects may involve using all of these types of software in combination. From a management perspective, this complicates the issues of staff training and retraining and, with the attendant investment in staff experience in a particular package and interface, limits consideration of new, improved software and related capabilities that may appear in the marketplace. This situation is likely to continue as more specilised software tools will appear, therefore, instead of pretending that a transport modelling tool be, at the same time, a good statistical tool, a good database management tool, a good GIS, a good transport network manager and having user-friendly interface, an "open multi-software system" solution seems indispensable. Bridges research (1997-2000) provide for the harmonisation formats and routines able to support such a open-support systems to be used in transport modelling.

1.16 The advent of the graphical Windows style interface has revolutionised the use of personal computers by making them more accessible and eliminating many of the technical barriers to their use by the less technical user. Similarly the rise of the Internet and the associated remote access capabilities, including easy to use World Wide Web browser software, email and other facilities, has begun to change the perception of separation and distance. Already, the Internet, particularly in technical and scientific fields, is widely used for co-ordinated far-flung project participants and activities. Using these capabilities, Bridges research aimed to bridge the gap between the policy-maker and the model, by somehow interfacing them. Two tools were developed to make this feasible: in the one hand a tool for developing Expert Systems able to translate policy questions into model’s inputs and interprete modelling output’s. The second is a tool able to develop powerful user-friendly interfaces, including together with multimedia and Internet browser capabilities routines needed to handle large databases and complex transport topologies.

1.16 In ASSEMBLING a Executive Support System for European policy makers is being developed in Internet. It includes on-line access to transport models (the Dynamic System model developed by IWW for SCENARIOS) as well as a number of knowledge-tools (tools with friendly interfaces and interactive modelling capabilities based on results and algorithms previously developed in advanced models). On the other hand, based on the databases and results of the forecast models developed by NEA and Mkmetric/IWW for the Phare countries, a so-called "Toolbox" has been developed following the same "knowledge-tool" approach and disseminated in CDRom. These experiences show the interest and feasibility of the "interfacing" strategy being developed in European transport modelling since early 1995, when the so-called UTS (Mcrit, 1996) was developed as a user-friendly tool for free dissemination providing interactive GIS visualisation and analysis of pre-calculated accessibility models.

1.17 In the project Bridges a software-tool to develop Expert Systems for advanced models was created with the aim to bridge advanced models to end-users (decision-makers). A similar case of policy-strategic interface (based on a different approach) is the PACE-FORWARD policy-interface developed by RAND for the Dutch Government.

1.18 The policy relevance of a model is directly linked to the inclusion in the model of key indicators related to both the policy instruments to be evaluated and the goals to be achieved. For instance, to be policy relevant in relation to Kyoto’s agreement, a model has to produce as output the total CO2 emissions (to be compared with the –8% Kyoto’s reduction goal) and give the user the capacity to modify policy instruments such as road pricing, standards for vehicle emissions etc.

1.19At European level, the policy relevance of transport models can be referred to the following set of questions: Construction of new infrastructures, Introducing transport pricing strategies, Changing transport service provision, Spatial Development strategies and Environmental regulations. The key gap between scientists and policy-makers is that both have different starting points: scientists start with a given question and a provisory hypothesis to be validated or rejected, and policy-makers start with many answers each one looking as a definitive position, and have as objective not to validate any one in particular but to negotiate and agreement among the groups lobbying for each one. How to integrate scientific models and rational ways of thinking in a political process requires a kind of intelligent mediation.

Taxonomy: How to classify models?

 

Type of models	Strategic	Tactic	Operational
Main policy-relevance	Infrastructure capacity expansions	(TDM) Demand management strategies	Service optimization
Samples	New roads and public transport extensions	Innovative supply (park&ride, car-pool, access restrictions...), dynamic toll, fares, new bus services...	Renewal of vehicles in the fleet, change of routes and services, scheduling...
Time horizon	Long-time (10-20 years)	Medium-time (1-5 years)	Short-time
Typical interested agent	Territorial planner	Transport regulator or planner	Transport operator or traffic controller
Paramount goal	Multicriteria evaluation across sectors and geographic scales	Social Welfare. Socioeconomic and environmental profitability. Equity issues.	Financial profitability
Assessment	Assuring minimum quality thresholds (e.g. minimum accessibility to networks, e.g. minimum level of service in main links)	Optimizing key strategies and monitoring overall quality thresholds.	Optimizing profitability
Impacts	Spatial development impacts. Regional equity. Global environmental impacts.	Local environmental impacts. Social equity.	Traffic control.
Modelling paradigm	4steps integrated in demographics, land-use and economic models.	1-4steps and/or Discrete Choice integrated with an optimisation module	Agent-based models
	Aggregate models (few periods, few market segments) with static equilibrium	Desegregate models, statistically-based with/without dynamic equilibrium	Dynamic micro-simulation at vehicle level.
Specialised software	MEPLAN, ESTRAUS	GAMS, TRIO	TRANSIM

Internal electronic conferences based on Q&A:

Glossary: What's a transport model?

Taxonomy: How to classify transport models?

Consistency: Which type of models are more consistent when modelling the EU transport system at strategic level? (to be launch in August 2000)