Improving the Measurement of Congestion: Interim Report

Introduction

This paper sets out the interim thoughts of the Congestion Measurement Working Group, incorporating comments from the Motorists Forum & CfIT. The Group was formed in 2001 under the auspices of the Motorists Forum and has met twice.

The Group comprises representatives from the CBI, RAC, AA, Highways Agency, ACPO, LGA, DTLR and officials from the Motorists Forum and Commission for Integrated Transport. Its remit has been to:

• look at how motorists perceive congestion
• assess DTLR's proposed criteria for measuring/reporting changes in congestion
• see whether the proposals could be made more meaningful to motorists.

The Department has recognised that its chosen definition is not intuitive and that there are alternative, albeit imperfect and problematic, ways of measuring and expressing congestion. The Group's tentative conclusions are that:

• the DTLR choice of indicator (average delay per vehicle kilometre) is broadly sound but limited
• there are five general ways in which the DTLR indicator is not sufficiently meaningful to motorists
• there are six possible sets of improvements which would be desirable, covering the ways in which congestion is measured and expressed
• the possible improvements appear technically feasible, but the costs and benefits associated with them are not yet known
• there are a number of possible next steps to follow.

What do motorists want?

Motorists expect quality service from the road network. The degree of expectation will depend upon the real and perceived cost of road use, and the perceived quality of experience compared with other utilities and services.

Quality service can generally be taken to mean, among other things:

• the ability to carry out a range of journeys, on demand
• the strong likelihood that any given trip will be completed safely, within a reasonable time and with minimum interruption.

Congestion (technically, the incipient and actual saturation of road capacity), irrespective of its cause - be it regular (eg recurrently high traffic flows) or irregular (eg roadworks, accidents or major traffic-generating events) - reduces service quality. It increases journey times, variation in journey times and interruptions to journeys.

But motorists do not have a single view of what they actually mean by congestion, nor how it should best be measured. Emerging results from ONS surveys suggest that people think of congestion as situations characterised by any one of the following:

• traffic jams with complete stops
• stop-start driving
• moving very slowly
• travelling at less than the speed limit.

Focus group research suggests that motorists are in fact sceptical of the value of measuring congestion. Our view is that motorists, as both road users and taxpaying members of society, will perceive the issue in different ways. Ultimately, measurement is only likely to be seen as important if it leads to action - by motorists, policymakers or providers - to reduce effect of congestion. Thus:

• measuring congestion will be of most direct use to motorists if it provides good information on journey time predictability, which can be used to plan ahead
• measuring the actual impact of congestion will be of less direct value, but is still important as a way of holding policymakers and providers to account
• modelling the likely impact of congestion will be of less value still to motorists, but is important in shaping policies and measures implemented for their benefit.

Generally, any approach to measurement also needs to be:

• accurate - adequately capturing conditions on the ground
• relevant - reflecting the breadth of road users' motoring experiences
• understandable - expressed in terms which are clear to motorists
• practical - in terms of data collection and analysis
• serial - enabling actual performance to be compared fairly between different years
• forecastable - to enable policymakers and providers to identify appropriate action.

The DTLR approach to congestion measurement and its presentation

Description

The DTLR has a PSA target "to reduce road congestion on the inter-urban network and in large urban areas in England to below current levels by 2010". The target will cover parts of the network accounting for an estimated 85% of the time lost due to congestion in England. DTLR's chosen indicator to measure progress against the target is average delay per vehicle kilometre, compared to driving at speeds typical when traffic is light. It regards this indicator as both forecastable and measurable.

Methodology

Data on average traffic speeds is collected through floating car surveys, which have been operated in London for many years and which have recently been extended to other major urban areas and the trunk road network.

The reference speed against which actual speeds are due to be compared will be the speed limit, in the case of trunk roads, and night-time measured speeds in urban areas. Data will cover the inter-urban road network as operated by the Highways Agency; Greater London; the 6 PTA conurbations; and the 12 urban areas with populations greater than 250,000. Urban areas outside London and trunk roads will be surveyed in alternate years.

Survey results are normalised through some modelling "infill" (to deal with gaps and outliers in the data). Extra surveys have been carried out to cover Monday mornings, Friday afternoons and weekends. School holidays, Bank Holidays and periods when it is known in advance that major roadworks or traffic generating events are due to take place, are avoided.

Shortcomings

We support the DTLR view that its chosen indicator is measurable and forecastable, and that it has merit as a measure of congestion, but believe it to be insufficiently meaningful in its current form for motorists in five ways.

First, there is a concern that the treatment of the data, while understandable from the point of view of statistical analysis, will render the results less "real" than the variety of conditions and journey types experienced by motorists.

Second, the concept of average delay is not immediately attractive to motorists. Average journey time (eg in minutes per mile or km) might be more important. Average delay certainly seems less important than distribution of journey times. One of the major sources of dissatisfaction for motorists is the gross divergence of journey times from what might be expected as the norm. Information on the distribution of journey times (from shortest to average to longest) would more accurately capture an issue of particular concern to motorists.

Third, the chosen indicator does not provide motorists with practical information on predictability of journey time. Focus group research conducted for the Department suggests that motorists believe predictability to be a relevant concept - though the research also indicates that there are major challenges in finding ways of expressing predictability in ways that are meaningful to motorists.

Fourth, while motorists (when asked) appear to find the DTLR's indicator easy to understand, they prefer other forms of expression - but there is considerable variety in what motorists regard as the best alternative. ONS survey work indicates that most (just over half) respondents preferred a measure expressed as extra time taken compared with free flow time. Focus group research for the department reveals that an indicator expressed as the amount of time spent in traffic jams was the most popular measure among participants.

Fifth, there is a danger that however well measured and expressed, hard data on the effects of congestion may not register with what motorists think is happening on the roads. There would appear to be merit in policymakers - given their commitment to improve the service delivered by the road network - having a good grasp of motorists' perception of progress towards the target.

Conclusions

In reviewing the DTLR's proposed approach, we have had to recognise that, for the present, soundly-based and comprehensive measures of congestion will have to relate to relatively large networks and be produced retrospectively so that their value for journey planning will be very limited.

Looking ahead, particularly as new techniques of measurement become available, it will be important to keep in mind the desirability of producing more specific journey-related information consistent with the general measures.

A vital caveat, particularly given motorists' scepticism, is that efforts to improve the measurement of congestion should not detract from real action to reduce its impact. Nevertheless, we believe that more could be done to make congestion measurement more meaningful to motorists - hopefully avoiding the risk of pursuing the perfect at the expense of the good - by tackling two broad sets of issues:

• what is measured - and how
• the presentation of any form of measurement.

Potential areas for improvement

Improve the methodology on measuring average delay

We believe it would be desirable to include, in the data capture and analysis, more of the breadth of journeys experienced by motorists. There is a good case to include traffic conditions affected by schoolruns, school holidays, Bank Holidays, incidents and events - on the basis that "what gets measured gets managed". We recognise that this poses challenges, among other things, in terms of how the information is presented, but these do not seem insurmountable in principle (see below).

A further challenge - particularly for predictability - is that while the incidence of school terms is known and their effects reasonably uniform over time, road accidents and incidents are unpredictable, both by location and time. The very treatment of incidents (their definition and management) is complex and we note that another Motorists Forum working group is examining this issue.

To make sense of some of the additional factors (eg schoolruns), there is also an implication in terms of the regularity of data collection and its presentation. Annual figures will not be as meaningful as, say, quarterly ones. This would appear to imply a significant additional cost, though we would note that the approach adopted in Australia appears to involve data captured three times a year.

In the future, it is expected that data provided through the floating car survey approach could be usefully supplemented by GPS-generated data. Our understanding from the Highways Agency is that at present, the latter is not yet sufficiently robust. We understand from the Agency that other sources (such as freight vehicle tachographs and Trafficmaster data) also appear to be unsuitable.

We note the use of two different reference speeds for different parts of the road network: we have no particular view on the merits of the specific baselines adopted. We further note the Highways Agency has shown interest potentially in using an alternative baseline for the inter-urban network, such as some "intervention" level of speed. We understand this would be used for internal Highways Agency purposes only, aimed at assisting it in targeting the most appropriate parts of the network for improvement. Again, we have no particular view on this, other than that it would be highly desirable to use the same baseline for the whole of the inter-urban network.

Develop an additional measure of the variability of journey times

We believe it desirable to measure the distribution of journey times to indicate the extent to which there is variability, and particularly divergence from some defined norm. We note, again, that variability of journey time is reported in Australia, indicating that at least technically this is feasible.

Measure average delay and variability for "typical" journeys

The floating car survey is already due to collect data by road and area type - but the measurement of average delay, even if disaggregated into "all large urban areas" or "the inter-urban network" will not feel particularly relevant to individual motorists.

Some progress already appears to be under way. We understand that additional runs of the survey have recently been conducted to enable the Highways Agency to measure conditions on specific parts of the motorway network (such as the M25, M4, M1). Extending this to other parts of the network would be desirable.

But we believe it would be desirable to go significantly further and to measure performance (not just in terms of average delay but, as suggested above, in terms of variability) against a basket of typical journeys (by purpose and length) by region.

This approach could mirror that used for measuring inflation (eg the RPI) and could draw upon the data and methodology underpinning the existing National Travel Survey. Inevitably, there are challenges in determining a "typical" basket of journeys, but there is no reason to assume this is any more difficult in principle than measuring RPI. And while there is no such thing as a "typical" motorist, such measurement would approximate closer to his or her actual road use. We also understand that, from next year, the sample size of the NTS is being tripled.

Finally, we note that congestion measurement in Australia and in the USA does appear capable of presentation in a disaggregated manner.

Establish a programme to develop indicators of predictability

We note that there is a distinction between measuring variability and measuring predictability. This might be analogous, for example, to the Met Office on the one hand collecting data on weather patterns and on the other hand providing weather forecasts which individuals can use to plan their activities.

One of the challenges in developing predictability indicators is that many journeys involve using roads belonging to more than one highways authority. Progress on this front is likely therefore to need the involvement of many players, such as DTLR but also potentially other organisations, such as the AA and RAC, who already provide comprehensive traffic information. We would encourage work to continue in this field and for players to be given the right incentives to do so.

Measure perception

We believe it would be desirable for the Department regularly to survey motorists' perceptions about changes in congestion, alongside "harder" measurements of road performance. We understand that the Department is already considering instituting satisfaction surveys.

However, we recognise that caution is needed in this field. We also note that there might be merit in this regard in drawing on other similar activities (eg the RAC and CfIT annual surveys, and the AA's own surveys of congestion and incidents).

Make the explanation of any measurements clear and motorist-friendly

However the hard data is measured - be it on average delay or variability, as proposed - considerable attention needs to be given to how the information is presented to motorists, and we know that the Department is keenly aware of this challenge.

Focus group research for the Department indicates that motorists find expression of delay, either in terms of seconds per vehicle kilometre or aggregated into hours lost per year, difficult to relate to their own experience. Equally, motorists surveyed to date appear to feel uncomfortable with figures expressed as percentages.

Ultimately, the most important thing is likely to be the expression of any change (and ideally, improvement) in performance of the network over time. Although the group has not been yet able to test this sufficiently, this would tentatively suggest using some sort of index, rather than nominal figures, to express congestion. Again, experience with use of RPI may prove instructive in this regard.

If, as proposed above, the measurement of average delay includes a wider range of journeys, it is important to explain the figures which result - for example, highlighting how far any set of figures are influenced by seasonal factors (such as the schoolrun) or by unusual levels of incidents/events. Although the group has not tested the idea, the practice of seasonally adjusting employment figures may again provide some pointers.

Finally, it is important to handle carefully any presentation of the figures. Reducing congestion is but one aim of Government transport policy, among others. Measures that help achieve other goals (such as improved road safety or modal shift) might also be desirable, even where they have a negative impact on the congestion measures used. For example, the average delay indicator could be adversely affected by the introduction of speed cameras which result in greater compliance with speed limits.

Potential next steps

Carry out some initial cost-benefit analysis

Our understanding is that the improvements identified above are technically possible but carry resource implications. In many cases, for example, there is an issue about developing sufficiently large samples to provide the disaggregated level of measurements proposed. In the absence of sufficiently robust GPS-generated data, this would require many more floating car survey runs to be done.

The urban and inter-urban surveys cost annually about £100,000 and the "infill" survey has a similar cost. However, we have not established the level of additional cost that might be involved, in absolute terms or relative to other elements of the DTLR research budget, nor of the benefit that might accrue. Initial work, possibly by the Department, to establish at least in outline the possible costs and benefits of moving forward on the improvements suggested above, would be a useful next step.

Consider the needs of other road users

Our remit has been to focus specifically on ensuring that the official measurement of congestion is meaningful to motorists. Consideration should be given to looking at whether recommendations in this respect would be equally applicable to other road users, including bus operators/users. Freight users, for example, may place more importance on measuring journey time variability. It is also worth noting that the data capture does not extend to non-Highways Agency inter-urban roads, smaller urban areas and rural roads.

Understand better the approaches adopted overseas

We have alluded to work done on congestion measurement in other countries, notably Australia and the USA (the Department has also provided information on experience in other countries). There might be merit in researching further the extent to which the approach in these two cases is well-received by motorists.

Establish a phased programme of additional measurement and modelling

Even if the proposed improvements to measurement are practical and affordable, implementing them will take time. Developing a robust methodology to model what is measured, to help establish future scenarios, is likely to take even longer. This would suggest that the DTLR should continue to use its chosen indicator and remain committed to its target as currently expressed, but that it should keep these under review as its measuring and modelling capability improves.