4. Modelling and Forecasting
4.1 Difference between modelling and forecasting
Modelling and forecasting are both used to estimate demand for transport facilities. Demand estimates, for base and project cases, then become important inputs to the appraisal of transport initiatives.
The principle differences between forecasts and models lie in structure and complexity. A transport model that includes the traditional four-step model encompasses forecasts, but a forecast does not necessarily entail modelling. Forecasts - for example, those in which current demand is projected from some existing base demand using assumed changes in population or trip incidence - are more likely to be relevant to simple appraisal problems, such as an extension to an existing pedestrian/cycle path. A forecast might also be made drawing growth rates from analogous initiatives elsewhere that have already been subjected to some form of ex-post appraisal.
On the other hand, transport models (with their much greater complexity) are more suited to analysis of large initiatives such as the development of an active travel network or a ‘missing link’ active travel initiative that is significant enough to have network consequences. In these circumstances, a transport model would facilitate an estimation of the active travel initiative on demand for other modes. Estimation of those impacts can be a useful input to the appraisal of network-significant active travel initiatives.
4.2 Difference between walking and cycling
There are limitations to modelling pedestrian activity given the variations in length and purpose of trips. The variation in trip purposes involved can also limit modelling accuracy; for example, the amount of commuting and leisure travel will respond differently to an intervention.
4.3 Relationship to other modes
The main difference between modelling for active transport and other modes (private vehicles or public transport) is that the zone systems and networks used in an active transport model would need to be significantly smaller as the majority of trip lengths would be much shorter (0.4 - 1.2 km). Active travel models need to have greater spatial detail to account for the various paths and routes available.
Models may be used to estimate future demand for transport facilities where walking and cycling schemes form part of a larger set of transport proposals. In this context, demand and spatially aggregate models[1] would be relevant. Different modelling approaches may be required where walking and cycling schemes are promoted separately from other modes of transport. Modelling may not be required for active travel depending on the extent and scale of the intervention proposed. Forecasting may be sufficient to determine the impact of an active travel intervention.
4.4 Range of approaches
A range of approaches to modelling/forecasting active travel demand have been successfully applied. All of the approaches described in this section would be equally applicable to walking and cycling travel modes,but the choice of one approach over the other will depend on the:
- Scale of the initiative
- Availability of existing models and data
- Budget and time available
- Level of accuracy required.
4.4.1 Comparison studies
This method aims to predict the active travel demand of a facility or intervention by comparing it to usage and surrounding population and land use of a similar or comparable facility. The aggregate data from comparable facilities can be assessed in an attempt to identify variables that contribute to the different levels of usage between areas, time or facility.
An example of the comparative study method is given in the United Kingdom Department for Transport (DfT) online Transport Analysis Guidance (WebTAG) for active travel mode appraisal. The demand impact is estimated by reference to before and after demand surveys conducted for an already completed active travel initiative in a similar area. The comparative demand impact estimated from the before and after surveys of an analogous initiative in a similar area is used to derive ‘with project’ (project case) demand estimates. Underlying demand growth rates in the catchment area of the proposed active travel initiative are used to estimate ‘without project’ (base case) demand.
In this example, the existing base year survey data was increased using different growth rates for the ‘without’ and ‘with’ scheme scenarios. The ‘without’ improvement scheme uses local network or city wide annual growth rates (such as 0.25% for cyclists and 0.5% for pedestrians). The ‘with’ improvement scheme scenario uses growth rates taken from the comparative study, which showed a significant increase in pedestrian and cycle demand. The difference in use predicted with the intervention can be determined by subtracting the ‘without’ scheme estimates from the forecast demand predicted ‘with’ the scheme in place.
4.4.2 Aggregate behaviour studies
This approach relates active travel in an area to its local population, land use and other characteristics, usually through regression analysis. The model equations can be used to predict demand in other areas. Some data used for these studies is obtainable from census or geographical data sets (such as car ownership, income, average age, gender and journey to work data). This approach may be useful for a large network study, but otherwise may not be cost effective. Also, data such as bicycle ownership is not readily available at the local or zonal level. Geographical information systems (GIS) may be used to obtain data on the topographical profile of an area, network or city, which may be a useful characteristic for cycling demand.
4.4.3 Sketch planning method
This method predicts active travel demand of a facility or in an area based on simple calculations and rules of thumb about trip lengths, mode share and other aspects of active travel behaviour. A series of ‘back of the envelope’ calculations are used to estimate the number of pedestrians or cyclists using a facility. Sketch planning relies generally on existing data or easily collected data (such as population or census data, traffic counts, pedestrian counts, cyclist counts, zoning or land use data, trip length or crash data).
The accuracy of this method is sometimes questionable given that the parameters are generally derived from previous studies that may not necessarily have transferrable results.
4.4.4 Discrete choice models
These models aim to predict an individual’s trip decision, including choice of mode or route, as a function of any number of variables. Discrete choice models can be used to estimate the total number of people who change their behaviour in response to an intervention. The change in active travel demand can then be predicted. Parameters from these models can also be used to estimate the elasticity of demand (that is, the percentage change in pedestrian or bicycle activity) in response to a given change in another particular variable.
Discrete choice models can be calibrated using stated preference survey data and, as such, can be a cost effective way of estimating active travel demand for new facilities in areas with little or no relevant data.
As an example, a discrete choice model may be used to predict the probability of taking a trip by bicycle or by car based on the following three factors:
- Time difference between the two modes for the trip
- Gender of the individual
- The extent of the cycle facilities available.
The weight assigned to each factor (that is, the coefficient) can be used to derive elasticities. These can indicate the change in mode choice based on one of the three factors, while keeping the other factors constant. Transferring this data to other schemes may be difficult, but the elasticity may be used to estimate the change in users as a function of a change in a facility.
A discrete choice model could also be applied to an entire affected population (for example, one proximate to an active travel facility). The model would be used to estimate the total number of people who would change their behaviour as a result of the new facility.
4.4.5 Traditional demand models
These models employ the traditional four-step travel demand modelling approach using land use conditions, transport network characteristics and relevant travel behaviour variables to predict future active travel patterns. They predict total trips generated by trip purpose, mode and origin/destination and distribute these trips through a network of transport facilities.
Part T1 of the Guidelines provides more detailed discussion of four-step models.
As noted in section 4.3, the main difference between modelling for active transport and other modes is that the zone systems and networks used in an active transport model would need to be significantly smaller as the majority of trip lengths would be much shorter (e.g. 0.4 to 1.2 kilometres). Active travel models need to have greater detail to account for the various paths and routes available.
Some advance models have been developed to include an active travel component. These primarily deal with bicycle travel and are based in Europe. A British consultancy firm, MVA, has developed two such models:
- START, a mode choice model that includes both walking and cycling as options
- TRIPS, a network model package that includes a bicycle network option called MVcycle.
Each of these models requires small zone sizes to account for the shorter trips associated with active travel and the variations in the active transport network.
A similar model for cycling activity, QUOVADIS, has been developed in the Netherlands.
4.4.6 GIS based approaches
Geographical Information Systems (GIS) are information management tools with graphical display capabilities that can be used in many ways to appraise potential demand. GIS can also be used to enhance active travel demand forecasting and initiative assessment.
An example of this approach was developed in a University of Queensland study (Hutchinson 2000) of cyclists travelling to the campus. An initial multiple criteria analysis was undertaken on survey data to determine the important factors in cycle route choice. This information included directness, quality and topography of the route. Data on these factors were used to construct a layer in the GIS map of the area from which predictions of cyclists’ preferred routes could be made.
4.5 Limitations of modelling
For small scale initiatives with limited impact (such as a pedestrian refuge island or path widening), the comparative study might be the most appropriate approach to forecasting/modelling. A medium sized initiative (such as sizeable extensions to a cycle network or construction of an off-road shared path) might use the sketch planning method or a discrete choice model. A fully specified four-step network based model would only be justifiable for a major infrastructure initiative (such as a bridge over a river connecting with large residential or working populations).
[1] A strategic road network model is an example of a spatially aggregate model.