I posted a blog last year reporting a revision to the Department for Transport’s road traffic statistics that found traffic in minor roads to have increased by 26% between 2010 and 2019. This compared to an increase in traffic on major roads over the same period of 12%. There were two things going on: a statistical sampling problem and a likely real increase in traffic on minor roads.

DfT practice is to monitor a representative sample of minor roads, scaled up to estimate traffic on all minor roads on the network. Because any sample may become less representative over time, a benchmarking exercise is carried out every ten years using a larger sample. This allows data inferred from the original sample to be adjusted retrospectively and a new sample to be established. In the previous 2009 benchmarking exercise the adjustment was fairly small, but not so in the most recent investigation.

Transport for London queried the DfT estimates, which did not square with its own estimates of traffic on the capital that showed a declining trend. Prompted by this discrepancy, the DfT statisticians carried out a deep dive into the methodology, the finding of which were published last September. The outcome has been to revised downward the estimated increase in minor roads traffic from the previous 26% to 10% over the period 2010 to 2019 (see chart above). This amounts to an unusually large revision, and has been accompanied by relegation of the previous report to the archives.

The explanation offered by the statisticians mainly involved comparison of the samples employed in the 2009 and 2019 benchmarking exercises, using GPS data to compare traffic flows, dichotomising into ‘high’ and ‘low’ flow links, and finding that there were 5% more high flow links in the 2019 sample. This was then included as an additional stratification factor for recalculating both exercises. In addition, the benchmarking estimates for London were recalculated, distinguishing Inner London and Outer London, which had not been done previously.

Transport for London, in its Travel in London Report 15 (page 141), notes that the revised DfT data for minor roads traffic in London, although smaller than previously estimated, still represents a substantial upward revision of this component of road traffic in London, whereas the DfT has not revised its traffic estimates for major roads in the capital. The outcome is that whereas in the estimates published before 2019, minor road traffic was put at around 33 per cent of all road traffic in London, the latest estimates have increased this proportion (of a correspondingly larger total) to around 40 per cent.

I have been interested in these estimates of traffic on minor roads since there is much anecdotal evidence that the widespread adoption of Digital Navigation (generally known as ‘satnav’ in the roads context) allows these roads to be used by those without local knowledge, leading to more traffic than desirable in the neighbourhoods affected. The new estimated of the pre-pandemic growth of traffic on minor roads brings this into line with growth on major roads, which might seem to argue against the significant impact of Digital Navigation.

Yet the size of the correction prompted by the benchmarking exercise points the other way, since had the sample of minor roads used to track traffic growth remained representative over the ten-year period, the increase in traffic would have been seen year on year. The fact that this was only recognised from the benchmarking suggests some increased heterogeneity in the sample. My hypothesis is that use of Digital Navigation would lead to growth of traffic on minor roads located adjacent to congested major roads, where the former offer a time-saving opportunity, while minor road located elsewhere would be less affected. The London data showing a growth of traffic on minor roads compared with that on major roads, noted above, is consistent with my hypothesis, given the extent of congestion in this city.

I have been in correspondence with the DfT statisticians on this question. What remains unclear to me is whether they regard controlling for flow characteristics as simply a matter of controlling for mismatched samples, or whether they accept the possibility that use of minor roads changed over the decade in a way that increased the heterogeneity of traffic growth. I have been told that the heterogeneity of traffic growth is not something that their data is currently suitable to verify beyond road class, vehicle type, and regional differences. However, they are looking at other ways of stratifying the sample and at other options for minor road sampling and counting in the future, including use of GPS data.

I have noted previously the likelihood that the use of Digital Navigation is facilitating the diversion of local traffic to new capacity on motorways. More generally, Digital Navigation is changing travel behaviour in ways that need to be better understood for decisions on road investment and other policies, in particular the promotion of walking and cycling for which minor roads are well suited. It is regrettable that the DfT is not able to shed light on these changes as part of its otherwise extensive and comprehensive compilation of road traffic statistics.

According to press reports, the European Commission is in discussion with providers of Digital Navigation services to protect people in neighbourhoods from noise and emissions by adjusting the routing algorithms, which is good news.

The British government has long been keen on encouraging the prospect of what were initially known as ‘connected and autonomous vehicles’ (CAV). In 2015 it set up the Centre for Connected and Autonomous Vehicles to stimulate the development of the technology and steered funds into the sector via the Transport Systems Catapult. Typical of recent bullish statements by DfT ministers is this from Grant Shapps, when Transport Secretary, earlier this year:

‘The benefits of self-driving vehicles have the potential to be huge. Not only can they improve people’s access to education and other vital services, but the industry itself can create tens of thousands of job opportunities throughout the country. Most importantly, they’re expected to make our roads safer by reducing the dangers of driver error in road collisions. We want the UK to be at the forefront of developing and using this fantastic technology, and that is why we are investing millions in vital research into safety and setting the legislation to ensure we gain the full benefits that this technology promises.’

There’s no doubting the enthusiasm, but is this just hype, or is there a reality underpinning the vision? And what exactly is the vision anyway? It seems to be principally focussed on re-imagining the existing private car as a vehicle that no longer needs a driver behind the wheel, and which can whizz its occupants around in robotic mode, avoiding collisions and improving safety, whilst allowing the passengers to get on with other tasks, or simply enjoying the ride and the view – or maybe having a nap.

Trials of such driverless cars on public roads are taking place in a number of countries, mostly with a human ‘safety driver’ still on board, but this requirement has been dropped in some US and Chinese locations where traffic conditions are deemed favourable. Yet the prospects remain uncertain, so it is timely that the House of Commons Transport Committee has initiated an inquiry into self-driving vehicles.

I am specifically exploring here the prospects for such driverless cars ever becoming widespread, acceptable and effective in the foreseeable future. I’m not considering the possibilities of driverless multi-user pods or rapid transit buses on at least semi-segregated roads, where some deployment has already taken place. Nor do I discuss the scope for freight vehicles to operate in platoons on inter-urban motorways, where a recent UK trial found the hoped-for improvements in fuel use to be disappointing.

The government published a substantial policy document in August of this year entitled Connected and Automated Mobility 2025, jointly presented to Parliament by the Secretary of State for Transport and the Secretary of State for Business, Energy and Industrial Strategy.

Note that Automated has replaced Autonomous, perhaps reflecting a scaling down of expectation. Note also that Mobility has replaced Vehicles, hence CAM rather than CAV, a sensible but small generalisation. But note also the absence of reference to Connected within the new document: the whole emphasis is on automation of individual vehicles, with little mention of the possibilities of connectedness between vehicles. In this, the government is following the prevailing mindset of the automotive industry, tech start-ups as well as established vehicle manufacturers, who see autonomy easier to achieve than connectedness between vehicles of different manufacture and type.

The self-driving vehicle concept is seen as the application of CAM technologies with greatest commercial potential. But for deployment to happen, safety must be assured. So in 2018 the government asked the Law Commission to develop the legislative framework for a safety regime. The Commission’s final report, also published earlier this year, will be the basis of legislation that the DfT will bring forward.

Legal consequences

Because, to be useful, a self-driving vehicle will need to be able to drive itself for at least part of a journey, there are profound legal consequences. The human driver can no longer be the principal focus of accountability for road safety. Instead, new systems of safety assurance are needed, both before and after vehicles are allowed to drive themselves on roads. The legal framework has to cover both self-driving functions, where a user must be able to take over control when the vehicle cannot cope – for instance in poor visibility-  as well as more advanced automation when the vehicle is able to cope under all conditions.

The Law Commission proposes three new legal actors:

  • Authorised Self-Driving Entity (ASDE) – the vehicle manufacturer or software developer who puts a self-driving vehicle forward for authorisation to the regulator.
  • User-in-charge (UiC) – the human in the driving seat while the vehicle is driving itself, who can take charge if the self-driving function cannot cope.
  • NUiC operator, a supervisory system responsible when there is no user in charge, to navigate obstructions and deal with incidents.

There will need to be regulators for new functions:

  • Vehicle type approval.
  • Authorisation to self-drive.
  • In-use safety regulation.
  • NUiC operator licencing.

The proposals of the Law Commission will need to be enacted in legislation, to make possible the government’s intention to make the UK an attractive place for the deployment of self-driving vehicles, even though the main development of the technology is taking place in the US, where there is considerable diversity of regulatory regimes across the states. However, the comprehensive legal framework may deter some prospective developers.

Ride-hailing operators such as Uber, who have aspirations for a driverless future, might hope to save the costs of the driver. But under the proposed regime they would need to own the vehicles and accept responsibility as the NUiC operator for oversight of operations, and possibly also to be the ASDE. This would be a large and costly change from the company’s current capital-light business model, where ownership of the vehicles lies elsewhere.

Electric vehicle maker Tesla’s approach to automation has meanwhile been to incorporate the hardware into existing privately-owned vehicles and to progressively improve the software over time to get to the point of implementing what is being marketed as ‘Full Self-Driving’ capability. Taken literally, this would also require Tesla to be licenced as a NUiC operator, with oversight of its whole fleet of vehicles on the road, which is hardly conceivable. So if they use the automated mode, Tesla drivers would need always to stay alert and be prepared to take charge in the event of the unexpected.

Safety

A key question in this discussion is the appropriate level of safety required for a self-driving vehicle. The Law Commission took the view that this is best decided by ministers, in the light of judgement of the public’s acceptance of risk. The view of the government, set out in the recent policy document, is that self-driving vehicles should be held to the same standard of behaviour as that expected in road traffic legislation for human drivers: ‘competent and careful’. This standard is higher than that generally enforceable now on the average human driver – who include, for example, drivers who are fatigued, distracted or under the influence of drink or drugs- despite this not being the behaviour that is expected of them.

The ‘competent and careful’ standard appears to be the minimum required politically. Anything less would hardly be publicly acceptable. But does it do the trick? On occasion, self-driving vehicles will inevitably be involved in crashes with fatalities. ‘Competent and careful’ implies that the self-driving vehicle could not be blamed. The fault would be attributed to the other drivers (‘human error’) or to vehicle- or road-related defects. Yet would it be possible to demonstrate such blamelessness in practice, when the other parties to the crash would be seeking to shift responsibility?

In Britain there is one fatality per 140 million miles driven, so deaths involving self-driving vehicles could be expected to be exceedingly rare events. But when they do occur, they will surely  attract much public attention. As a result, there may well be pressure to tighten safety standards further, which, done in response to public anxiety, may undermine the general acceptability of self-driving vehicles. There is therefore a good case for a higher initial safety standard than ‘competent and careful’ to be set.

Benefits

The practical development of vehicle automation for general deployment is proving more difficult than the pioneering technology optimists had hoped. Apart from getting the technology to the point of being publicly acceptable, there is an evident need to develop viable business models consistent with the expected regulatory regime. And that means asking who will buy the vehicles? This in turn will depend on how purchasers perceive the benefits. In this regard, the practical benefits of vehicle automation remain unclear.

Proponents point up the safety benefits, especially in the United States where almost 39,000 people were killed in motor vehicle crashes in 2020, a fatality rate of 12 per 100,00 population, compared with 2.3 for the UK. Given that human error and risky behaviour is said to be responsible for 90% or more of fatalities, it would seem a reasonable expectation that the ‘robot driver’ of a self-driving vehicle could do better than fallible humans. On the other hand, robots suffer from their own shortcomings, tending to be less effective at perceptions involving high variability or alternative interpretations. In particular, robots would find it difficult to engage in the kind of visual negotiation that occurs between human drivers to settle which gives way when space is tight. Besides, the driving performance of a robot would need to be very similar- or better- than that of a human driver to ensure public acceptability. A self-driving vehicle that proceeded particularly cautiously to meet safety requirements could be unattractive to the purchasers, and an irritation to other road users . So the robot driver would effectively need to learn how to drive like a human.

Another claimed benefit is that automation might increase the capacity of existing roads by allowing these safety-conscious vehicles to move with shorter headways, that is, with a smaller distance between them than the recommended two-second gap on fast roads. The more precise control exercised by a robot might also smooth traffic flows and allow the use of narrower lanes. However, such increases in capacity would seem likely to be possible only on roads dedicated to self-driving vehicles, since the presence of conventional vehicles, not to mention cyclists, motorcycles and pedestrians, would require standard spacing to be maintained. In any event, in line with current experience, any increase in capacity would be expected to attract additional traffic, so that long term congestion relief would not be expected.

Automation that allowed an increase in road capacity might anyway be of interest to a road authority, as a public benefit, but not to individual vehicle owners if they had to bear the cost of the necessary technology. Because self-driving vehicles would be capable of operating empty, for example when returning to base after dropping off the occupant, and thus not needing to pay to park, they could add to the traffic miles generated and hence to congestion. Conventional taxis operate without a passenger while seeking a fare, of course, but privately owned vehicles without occupants would be a new source of traffic. All this is before any consideration of how they would be licenced and the fee, and/or any road user charges they must pay- whether the same as, or different to, conventionally driven vehicles.

Prospects

One obvious problem of driverless vehicles is how they can happily enter the vehicle population alongside conventionally-driven vehicles The prospects for widespread vehicle automation on existing roads with mixed traffic seem very uncertain. The driving task on motorways might be lessened in good visibility and in the absence of road works, but the driver would need to be immediately available to take control in adverse situations – and when going into general traffic. There are some low-speed environments that might accommodate driverless vehicles, including campuses, business parks and other new developments with extensive road space, and some such deployment has taken place; in these circumstances driverless shared-use pods could be a cost-effective transport technology. Yet it is hard to see driverless vehicles successfully negotiating historic towns and cities with complex road layouts, often narrow streets, extensive kerbside parked cars, cyclists and randomly moving pedestrians.

This impediment to automation in general traffic in urban areas is particularly relevant to the idea of driverless taxis that could be attractive to ride-hailing operators such as Uber, to avoid the cost of the human driver and spread the extra capital cost of the vehicle through intensive use. On the other hand, there would be the potential cost of ownership of fleets of automated taxis, rather than the current model of ride-hailing operators where drivers own their own vehicles. Besides, a driverless taxi system operator would need to exercise oversight of the activity of each vehicle in their fleet, to deal remotely with incidents and navigate obstructions, which would add to costs. It is certainly possible that the relatively unskilled human taxi driver in their own car may remain a lower cost option than the robot driver in a specially provided one.

The way forward thus remains unclear. Indeed, after more than a decade of development of autonomous vehicles, the early excitement and optimism have been followed by some disillusion as the problems of achieving and implementing an acceptably safe product have been recognised. The recent decline in the value of technology stocks and venture capital investments suggests that finance to support further development of vehicle automation may be harder to come by. Indeed, Argo, an autonomous vehicle tech start up funded by Ford and VW, is shutting down, evidence of uncertain prospects for the technology. Accordingly, it would be premature to predict the eventual outcome, both in timing and extent of deployment. Who can judge whether, as enthusiasts for the technology assert, children born today will not need to learn to drive a car?

The major transport innovations of the past have been those that made possible step-change increases in the speed of travel – the railway and the modern bicycle in the nineteenth century, the internal combustion engine for road vehicles and the jet engine for aircraft in the twentieth century. These step-change increases in speed, and thus reductions in journey times, permitted increased access to people and places, opportunities and choices, which were the benefits of harnessing the energy of fossil fuels for the motive power of mobility. In contrast, it is unlikely that automation would increase the average speed of travel on the existing road network, which is constrained by safety and congestion. Accordingly, the benefits of vehicle automation are much more likely to take the form of improved journey quality, and hopefully safety, and the possibility of doing other things whilst on the move – particularly those that have come with the digital information and communications era . It remains to be seen to what extent purchasers of driverless vehicles will be willing to pay for these benefits – or if they can achieve them by adapting to wider new mobility thinking in other ways, particularly at a time of changing inter-generational attitudes to the ownership of vehicles personally, and their use.

The UK government has been very supportive of vehicle automation, in particular aiming to put in place a comprehensive legislative regime based on the thorough analysis by the Law Commission. Yet there is only limited commercial development of the technology underway in Britain, so the benefits for both industrial and transport policy do not seem to be that great. A more important technology priority for road transport policy has surely to be the switch to electric propulsion and the quest for transport decarbonisation.

This blog is the basis for an article in Local Transport Today dated 14 November 2022.

I made a presentation to the Highways UK Conference held at Birmingham 2-3 November 2022. These are the main points.

The widespread use of Digital Navigation (DN) (generally known as satnav) is changing travel behaviour (see my recent paper). One impact is to divert local users to major roads to take advantage of increased capacity. I have analysed two Smart Motorway investments in detail: M25 Junctions 23-27 and M1 Junctions 10-13. The Smart Motorway concept involves converting the hard shoulder to a running lane, originally during the periods of morning and evening peak demand, or, as has been recent practice, throughout the day. The advantage is that capacity can be increased without the cost of additional land take or rebuilding bridges.

Monitoring the traffic flows and speeds 3-5 years after opening of the two schemes showed that the forecast increases in speed had not occurred, and hence the economic benefits, which largely depend on the value of time savings, were not obtained. Something had gone badly wrong with the traffic modelling that informed the investment decisions. Accordingly, I sought and obtained copies of the relevant reports.

The traffic modelling in both cases employed regional variable demand models that utilised the long established SATURN software. Traffic flows and speeds for the with- and without-investment cases were compared, and the outputs fed into the economic model, the Department for Transport’s TUBA model, which forecasts the economic benefits. In both cases, substantial travel time saving benefits were projected for business users, offset by a small increase in vehicle operating costs (VOC). There were also substantial time savings for non-business users (commuters and others) but these were very largely or entirely offset by increased VOC. Hence there was no net economic benefit to non-business users. It seems likely that increased number such users, above that forecast, pre-empted the increased capacity intended for longer-distance business users.

Examination of the routing information offered by Google Maps, for a journey between two locations in the neighbourhood of a Smart Motorway scheme, shows that diversion to the motorway can save time, but at the cost of increased distance and hence fuel cost (see screenshot at top). This is consistent with the modelling, on the basis that road users are likely to take the faster route and be less concerned about VOC.

Those making local trips have a variety of options, using the motorway as well as local roads, while long-distance users are likely to stay on the motorway. In the past, local users would have made routing decisions based on recent experience of congestion and  broadcast traffic information. But with the widespread use of DN, the choice of the fastest route is clear. The impact of DN is to increase local use of motorway capacity, to the disadvantage of longer-distance users. This seems likely to be an important contributory factor to the failure of the M25 and M1 Smart Motorway investments to deliver the expected travel time savings.

Although detailed information is available only for two Smart Motorway schemes thus far, it is likely that these may not be unrepresentative. The Strategic Road Network (SRN) is under greatest stress in or near areas of population density, where local users and longer-distance users compete for road space. Remote from such areas, the traffic generally flows fairly freely. So opportunities for investment appear to be where local users are best placed to take advantage of new capacity.

If the motorway system operated as a toll road, as in France or Italy, tolls would deter use by locals. The one example in the UK is the M6 Toll road in the West Midlands, built and operated with private finance, where daily traffic is half that on the adjacent M6 proper, doubtless due in part to the toll that local users do not choose to pay. But this is the exception that proves the rule: which is that attempts to alleviate congestion by increasing the capacity of major roads experiencing marked peaks of traffic at commuting times, as with Smart Motorways, must be expected to result in increased use for local trips, to the disadvantage of longer-distance users.

To better appreciate the benefits of road investment, it would be important to understand the impact of DN on road user behaviour, so that this can be incorporated into the traffic modelling that informs investment decisions. It would also be important to get a more granular evaluation of outcomes of investment. Traffic and economic modelling of prospective investments distinguishes between business and non-business users, the former split between cars, light goods and heavy goods vehicles, and the latter between commuter and other journey purposes. In contrast, monitoring of outcomes only tracks total traffic, volume and speed. However, it is now possible to employ DN to distinguish between local and longest-distance traffic, as exemplified by the TomTom Origin/Destination analytical service. Making such a distinction is important for evaluating the economic benefits of investment since the total volume of traffic might be close to that forecast, but if the share of local users is greater than forecast, the economic benefit will be less than expected.

We are at present midway through the second five-year road investment programme, known as RIS2, worth £27bn over the period 2020-2025. RIS3 is now being planned. But there are headwinds:

  • The potential economic benefits are likely to be overstated, as discussed above.
  • Any increase in road capacity is inconsistent with the Net Zero climate change objective since both tailpipe and embedded carbon would be increased.
  • There are public anxieties about the safety of Smart Motorways in the absence of the hard shoulder, reflected in a critical report from the House of Commons Transport Committee, to which the government responded by halting new schemes until five years of safety data is available.
  • The government’s Levelling Up White Paper, published in early 2022, identified a dozen ‘missions’ across departments. The single mission for the Department for Transport is aimed at improving public transport in regional cities towards that achieved in London, a sensible political and social objective. There was no reference to road investment, which is sensible given that congestion delays on the SRN are less in the Midlands and North than in the South East.
  • Current pressures on medium term public expenditure.

Given these impediments, there is a good case for treating the SRN as a mature network, with a focus on operational efficiency. This is the case for urban roads, which in the past were enlarged to accommodate more traffic, but nowadays the trend is to reduce capacity allocated to general traffic, to encourage active travel and facilitate public transport. Similarly, the aviation sector focuses on operational efficiency – airlines maximising use of aircraft, allocated routes and passenger load factors; airports (struggling recently) optimising throughput of passengers and baggage; and air traffic managment making best use of crowded airspace. The underlying discipline is operations research, not civil engineering, plus the modelling and economic analysis of operations, not long-lived investment.

A focus on operational efficiency of the SRN would naturally prompt consideration of how best to take advantage of the huge investment in DN that has been made, both by providers of the service and by road users. Here a very odd phenomenon is the apparent disregard of DN by road authorities, at least a judged by their publications – no reference to satnav in those of National highways, the Department for Transport, or local authorities (with the one exception known to me, Transport for London’s collaboration with Waze). Why is this? Possibly because of the preoccupation of highways engineers with civil engineering works, the need to spend the large budget allocated to the SRN, the lack of professional background to cope with digital technologies, and that fact that road authorities are monopolies, so not subject to competitive pressures?

The one constituent of road users that is highly competitive is road freight, particularly that forming part of integrated logistics businesses, which makes extensive use of digital technologies to manage fleets on the SRN and delivery vehicles on local roads. We are well aware of this when we order goods online, with a specified delivery date and often a time slot, the ability to track packages, delivery confirmed on the doorstep, and our feedback sought on the experience – all done by algorithm. This kind of operational efficiency needs to be brought to bear on the totality of traffic on the road network.

Experienced network operators would naturally want to take advantage of DN, which is vehicle-to-infrastructure connectivity that is changing travel behaviour on a massive scale. One aim would be better to cope at times of stress – major incidents, bad weather, peak holiday flows. A second would be to optimise use of the network in normal times, including avoiding routing traffic through unsuitable minor roads.

There is a maxim that you can’t build your way out of congestion, which we know from experience to be generally true. The Smart Motorway case studies exemplify this truth and provide an explanation: increased capacity is taken up by local users, pre-empting capacity intended for longer-distance business users, with no overall economic benefit, and restoring congestion to what it had been before. However, when road users are asked why congestion is a problem, their main concern is the uncertainty of journey time. Digital Navigation provides estimates of journey time in advance, so those who need to be at their destination at a particular time can decide when best to set out; those who are more flexible can avoid the worst of congestion; and all can choose the fastest route.

Digital Navigation is vastly more cost-effective as a means to mitigate the impact of road traffic congestion than costly civil engineering investment.

I have previously drawn attention to the impact of Digital Navigation (DN) of travel behaviour. The providers of this service, commonly known as satnav, respond to requests from users by providing routing options and journey times that take account of prevailing traffic conditions. Here I want to consider how this is achieved.

There is only fragmentary published information on how the routing algorithms function. It appears that a model of travel behaviour on the road network is constructed from trip data derived from users of the navigation service: trip origins, destinations, routes through the network, time of day/week, prevailing traffic conditions, journey times. Such a model may be analogous to a microsimulation model, but using observed trip data rather than synthetic data. It could also be viewed as combining the trip generation, distribution and assignment stages of the standard four-stage transport model (the mode split stage not being relevant for committed road users seeking routing advice).

Providers of DN offer predictions of journey time in advance of setting out. Comparison of predicted and outturn journey times provides a check on the validity of the model. Machine Learning has been employed to improve the accuracy of journey time predictions of Google Maps.

The type of model developed by DN providers is novel and powerful in that it can utilise huge amounts of trip data, both real time and historic. A question is whether such models could be used to inform decisions on road investments and other interventions aimed at improving experience on the road network. That is, could the DN models replace conventional transport models for planning purposes?

The DN models already exist. Their cost of construction and operation is met by the income generated from sales, whether of direction services to business premises (e.g. Google Maps) or to vehicle manufacturers that fit DN as standard equipment (e.g. TomTom). So, the cost of using these models for planning purposes could be less than for building and using conventional models.

TomTom offers Origin Destination Analysis as a service and may therefore be open to suggestions for use of the underlying model for planning purposes.

Another possibility would be to create an open-source, crowd-funded DN model – a kind of not-for-profit version of Waze, a provider that encourages user input. The funders might be road authorities that would gain access the the underlying model for planning purposes.

A further possibility arises from the likelihood that some form of electronic road user charging will be introduced, as electric propulsion replaces the internal combustion engine, to replace revenue from fuel duty. This is likely to involve technology similar to DN, and might therefore be the basis of traffic modelling for other purposes.

DN is both changing travel behaviour and generating new travel models to inform public policy. We may be at the beginning of a new era of travel and transport analysis.

I post a blog a few months ago about the Office of Rail and Road (ORR), which was consulting on its involvement in the third road investment strategy (RIS3). The outcome of this consultation was recently announced.

I had responded to the consultation, suggesting that the ORR should take an interest in the benefits of road investment as experienced by road users, since there was doubt about who mainly benefited, local users (commuters and others) or business users. My contribution was published by the ORR, anonymously, as ‘Response from member of the public D’ (I would have been happy to be named, but was not asked).

In its detailed response to the consultation, the ORR responded to my suggestion at paragraph 3.7, saying that it is not within its remit. It is unusual for an industry regulator to take no interest in the consumer experience. This is a central concern for the regulators of other industries – energy, water, telecoms, financial services.

The ORR also ducked out of accepting other suggestions for extending its responsibilities, including in relation to other important government objectives such as Net Zero and Levelling Up. The detailed critique of the Transport Action Network (page 78 of the ORR response document) is well worth reading.

Altogether, the ORR is a pretty feeble regulator in respect of roads.

I have a new article published in a peer reviewed journal:

Metz D. The impact of digital navigation on travel behaviour. UCL Open: Environment. 2022;(4):05. https://doi.org/10.14324/111.444/ucloe.000034

Abstract

Digital navigation – the combined use of satellite positioning, digital mapping and route guidance – is in wide use for road travel yet its impact is little understood. Evidence is emerging of significant changes in use of the road network, including diversion of local trips to take advantage of new capacity on strategic roads, and increased use of minor roads. These have problematic implications for investment decisions and for the management of the network. However, the ability of digital navigation to predict estimated time of arrival under expected traffic conditions is a welcome means of mitigating journey time uncertainty, which is one of the undesirable consequences of road traffic congestion. There is very little available information about the impact of digital navigation on travel behaviour, a situation that needs to be remedied to enhance the efficiency of road network operation.

An article in Local Transport Today of 13 June broadens the consideration.

The Department for Transport (DfT) has started planning its third Road Investment Strategy (RIS3), a five-year investment programme for the Strategic Road Network (SRN) for the period 2025-2030. The approach is conventional – a programme of projects, with little overview of how societal objectives will be advanced by the likely substantial expenditure. Yet there are five major issues that need to be addressed for the programme as a whole.

First, there is a need to reconcile the government’s Net Zero objective with the carbon emissions from both the tailpipes of the additional traffic arising from increased road capacity and the embedded carbon in the cement, steel and asphalt used in construction. Recent presentations by the DfT’s Transport Appraisal and Strategic Modelling (TASM) division indicated an intention to tackle this issue at scheme level, but this is misconceived. What matters is the overall contribution of RIS3 to carbon emissions and how this is to be offset or otherwise justified.

Second is the question of how RIS3 advances the government’s Levelling Up agenda, where the recent, well-received White Paper identified twelve medium-term ‘missions’ to be pursued across all departments. The one specific to transport states: ‘By 2030, local public transport connectivity across the country will be significantly closer to the standards of London, with improved services, simpler fares and integrated ticketing.’ Although the rate of progress implicit in ‘significantly closer’ is vague, the direction of travel is clear and the objective is not in dispute.

There is no mention of investment in the SRN in the Levelling Up White Paper. This is appropriate since there is, if anything, an inverse relation between the performance of the road network and economic prosperity across the nation, given that delays on the SRN due to congestion are greater in London and the South East than in other regions of England.

The implication of the White Paper approach is that there should be a substantial switch of DfT funds from road investment to improve public transport beyond London, if the Department is to play a full role in supporting the government’s the Levelling Up agenda. Yet the Department’s recently issued Levelling Up Toolkit is essentially a pro forma for a box-ticking exercise aimed at justifying investments already forming part of agreed expenditure programmes. There is palpable inconsistency here.

Third, we have the problem of the safety of smart motorways. These require conversion of the hard shoulder to a running lane as an economical means of increasing capacity without the expense of rebuilding bridges. Generally, new roads are safer than older roads, which meant that adding road capacity yields a modest safety benefit. But this is not obviously the case for smart motorways, and there has been considerable pushback from the public and the House of Commons Transport Committee. As a result, the DfT has paused the roll out of new smart motorways until five years of safety data is available for schemes introduced before 2020. A decision on the generic safety of smart motorways will be an important factor in developing RIS3.

Fourth, and less recognised, there is a question about the economic benefits from additional road capacity. There are two published evaluations of smart motorway schemes where the traffic flows after opening were very different from those that had been forecast. For the M25 Junctions 23-27 scheme, the traffic flowed faster one year after opening but subsequently delays reverted to what they had been before opening on account of greater traffic volumes than forecast. For the M1 J10-13 scheme, traffic speeds five years after opening were lower than before opening. Since the main economic benefit of road widening is the saving of travel time, both schemes had negative benefit-cost ratios (BCR) at outturn.

Examination of the reports of the traffic and economic modelling of these two schemes showed substantial time-saving benefits expected for business users, offset by a small amount of increased vehicle operating costs (VOC) arising from additional traffic volumes. There were also time savings to non-business users (for commuting and other local travel) but these were entirely offset by increased VOC – because these were local trips that rerouted to the motorway to save a few minutes of time, at the expense of additional fuel costs.

The scope for rerouting local trips to take advantage of increased motorway capacity is likely to be underestimated in modelling. Local users have the flexibility to vary routes whereas long distance business users will stay on the motorway unless there is a major holdup. Moreover, the general use of digital navigation in the form of Google Maps and similar offerings makes choice of minimum time options commonplace.

Even when the outturn total traffic flows are a reasonable match to those forecast, the scheme economics could be much worse than predicted if there is more local traffic, and hence less long distance business traffic, than projected. Traffic and economic modelling involve recognition of different classes of road user with different values of travel time: cars, LGVs, HGVs, business, local commuters, and other local users. However, the monitoring of outturn traffic flows does not distinguish between these classes of users. GPS tracking make such distinctions possible.

The DfT has emphasised the importance of evaluation of outturns of investments. Yet the failure to appreciate the need to break total traffic flows down into the segments that had been modelled reflects a serious professional shortcoming. As a result, we cannot be at all confident that investments to increase SRN capacity do more than facilitate rerouting of short trips by local users, of nil economic value. Likewise, we do not have the kind of detailed evaluation data that would allow traffic models to be better calibrated for future use.

The fifth issue for RIS3 is that the widespread use of digital navigation by drivers prompts questions about the continued focus of DfT and National Highways on major civil engineering expenditure. Contrast the aviation sector, where new runways or terminals are occasional efforts, not regular business. The main focus of airlines and air traffic control is to improve operational efficiency, to sweat the assets employing the techniques of operational research. We have a mature road network in Britain. It’s time to focus on operational efficiency. Yet it seems not to occur the National Highways that working with Google Maps, TomTom and other providers of digital navigation services would be a cost-effective means of improving the performance of the network.

More generally, the DfT is trapped in its box labelled Transport Analysis Guidance (TAG), a thousand pages of prescription to which more text is added when some new issue or policy arises, such as Net Zero, Levelling Up, inequalities or gender. The task for those promoting a scheme is to tick all the boxes and flex the modelling to generate BCRs that represent good value for money. Evaluation of outturns is inadequate to distinguish between success and failure.

Although the DfT pays lip service to the need to think at the strategic level, the TAG framework does not facilitate this in that the detailed analysis is at project level. Other interested parties do not challenge the Department’s approach. The consultants and local authorities do not bite the hand that feeds them. The professional societies, institutions and think-tanks do not engage. The National Audit Office carries out good analysis of road investments on occasion, but not systematically. The Office for Rail and Road scrutinises the management of the SRN, including how well new investments are delivered, but does not see its role as enquiring into how investments benefit road users. This is quite unlike the regulators of other infrastructure industries – electricity, gas, water, telecoms – that are focused on how consumers benefit from investment.

The DfT is stuck in its box and seems unlikely to break out. The best bet for a strategic view of RIS3 may come from the National Infrastructure Commission, which has begun the development of its second National Infrastructure Assessment. The Commission’s advice was the basis of the government’s £96 billion rail investment programme for the North and the Midlands. This required fresh thinking about the benefits of transport investment at the level of the whole programme, an approach clearly needed for RIS3.

This blog post formed the basis of an article in Local Transport Today of 25 March 2022.

The Department for Transport’s National Transport Model (NTM) was first constructed two decades ago and has subsequently undergone a number of phases of development. The main function of the model has been to provide projections of travel demand as the basis for justifying investment in the road network. The model has also been used to project future carbon emissions, to inform the Department’s Transport Decarbonisation Plan, as well as to explore the impact of technological developments such as electric vehicles.

An account of the latest version, effectively a new model known as NTMv5, was released recently in the form of a 250-page ‘Quality Report’, oddly, two years after completion. NTMv5 is a spatially detailed conventional four-stage transport model structure, iterated so that congestion feeds back into demand. The model has been implemented using the standard commercial software, PTV Visum. The intention is that the model should be transparent to external stakeholders, a very welcome development given the opacity of previous versions of the NTM. The complexity of the model means that a single run takes around ten hours, with a number of iterations needed to achieve convergence of outcomes.

However, there are some notable limitations to the model. There is no detailed treatment of public transport capacity. Car ownership data derives from a separate model, which has not been updated. And the primary source of growth of travel demand is the DfT’s National Trip End Model data set that projects expected changes in demography and land use, which are problematic of account of uncertainty of economic growth, population growth and distribution, and planning policy.

A number of potential applications of the model have been identified, of which the most immediate is the preparation of new national road traffic forecasts. Also recognised is a need to project future transport carbon emissions, and for the analysis of packages of road schemes at national level, including value for money.

The purpose of the succession of NTM versions has been to support the traditional ‘predict and provide’ approach to road investment. This viewpoint persists in the latest version where the stated rationale for analysis of packages of road schemes is to identify ‘gaps in the network… where the road capacity in future may be insufficient, leading to unacceptable rises in congestion and journey times.’ (section 2.4.2). Yet we do not adopt that approach when considering urban roads, and the scope for enlarging peri-urban motorways at acceptable cost by converting the hard shoulder to a running lane is now problematic on account of public concerns about safety. Besides, the scale of induced traffic has been persistently under-estimated in traffic modelling, so the aim of avoiding unacceptable congestion seems naïve, even before addressing the Net Zero objective.

The model builders struggled to treat the complexities of urban traffic. It was accepted that a full link-based modelling of urban road capacity and related journey time responses could not be achieved, and therefore a simplified approach had to be applied. This involves assuming general fixed speeds on urban networks for the Base Year, which were reduced over time based on assumed growth of demand. (sections 4.7 and 11.4). This simplification has implications for projections of traffic in London, as recognised by the peer reviewers.

Peer Review

The DfT has published a Peer Review and an Audit of NTMv5. The 120-page peer review, led by the seasoned practitioners John Bates and Ian Williams, drew attention to a number of apparent shortcomings in the methodology (too technical for me to appreciate sufficiently to offer comment). These led to counterintuitive results when sensitivity tests were run, notably for London.

The reviewers advise caution in application of the model, primarily due to the focus of the NTMv5 being on the more strategic highway network, whereas many of the potential applications focus on urban travel policy and public transport interventions. In particular, the reviewers are critical of the treatment of urban traffic, observing that the assumed relation between traffic speed and demand growth lack validity, and that the range of policies aimed at reducing urban car use are not taken into account. Besides, it is noted that the DfT’s car ownership model has not recognised that ownership in dense urban areas has been declining for many years in response to increasing population density, notwithstanding rising incomes.

The reviewers find that for London, the model results are not convincing. The observed car (driver + passenger) trip mode share is 38% from the National Travel Survey in 2015/16, whereas that in the model in 2015 is 50%. Moreover, the model projects a future gain of car share, whereas over the period 2005-16 a major decline of 5.6% was observed (para 4.3.5). The reviewers concluded that the model could not be safely used to examine policies that relate specifically to London, and query whether this relates more generally to rapidly growing dense urban areas across England. They took the view that the model should be suitable for use in forecasting the growth of road traffic in most areas other than those adjacent to or within major urban areas (section 6.3.24), which is a pretty major qualification.

Audit

The 260-page audit of NTMv5, carried out by consultants Arup and AECOM, drew attention to a number of shortcomings in both documentation and substance, including that some of the model components and tools used to process the data are not owned by the Department, which limited access to some of the key processes and data used in model development – not consistent with the aim of transparency to external users. The auditors advised that users of model outputs should be cautious because of problems in reaching convergence to a stable outcome as the model is run through repeated iterations, a concern also of the peer reviewers.

What next?

The NTM documents recently published are two years old. No doubt, further development of the model has been taking place to respond to the issues raised in peer review and audit. In its Transport Decarbonisation Plan published last July, the DfT stated its intention to review the National Policy Statement on National Networks, the basis of strategic planning of road and rail investment, and to update the forecasts on which it is based. NTMv5 will presumably be used for this purpose. Yet the modellers will be stretched to meet the divergent needs of their client policy makers, between bullish forecasts of travel demand to justify continued infrastructure investment and bearish projections of transport carbon emissions. Given the uncertainties of the model illuminated by peer review and audit, it will be hard to be confident about the validity of carbon forecasts out to 2050 and 60-year investment appraisals.

While the DfT’s intention to make NTMv5 available for use beyond the Department is praiseworthy, this seems problematic in practice. Doubtless the large transport consultancies could master the software and data, but given their complexity, clients would need deep pockets to fund the work. That would rule out non-government bodies that might want to challenge particular schemes. Regional transport undertakings have their own bespoke models. I am not aware of any academics who would be likely to buy into the NTM, a situation unlike national energy modelling where government and a substantial group of university researchers work with the same model. The DfT would be well advised to support academic researchers and others wishing to use NTMv5 to explore a range of policy scenarios.   

This blog was the basis for an article in Local Transport Today of 11 March 2022.

The Law Commission has recently completed a review of the law relating to automated vehicles. The review had been requested by the Centre for Connected and Autonomous Vehicles, which reports to the Secretary of State for Transport. This was the first occasion in which the Law Commission had been asked to develop legal reforms in advance of future technological development. The government’s aim is to avoid regulatory barriers that might impede innovation while managing potential risks of harm.

The review process involved three rounds of consultation that fruitfully engaged a wide diversity of interested groups. It was a good example of how the need to capture policy intentions in legislation generates close scrutiny and clarification of the issues involved. The 300-page report is comprehensive, except that it does not consider V2V connectivity for AVs, a slight surprising omission given who commissioned it.

Because an automated vehicle (AV) will be able to drive itself for at least part of a journey, there are profound legal consequences. The human driver can no longer be the principal focus of accountability for road safety. Instead, new systems of safety assurance are needed, both before and after vehicles are allowed to drive themselves on roads.

The legal framework has to cover self-driving functions where a user has to be able to take over control when the vehicle cannot cope, for instance in poor visibility, as well as for more advanced automation when the AV is able to cope under all conditions.

The Law Commission proposes three new legal actors:

  • Authorised Self-Driving Entity (ASDE) – the vehicle manufacturer or software developer who puts an AV forward for authorisation to the regulator.
  • User-in-charge – the human in the driving seat while the vehicle is driving itself.
  • NUIC operator, responsible when there is no user in charge, to navigate obstructions and deal with incidents.

There will need to be regulators for new functions:

  • Vehicle type approval of an AV.
  • Authorisation of automated functions.
  • In-use safety regulation.
  • NUIC operator licencing.

The proposals of the Law Commission will need to be enacted in legislation to have effect. The government’s intention is to make the UK an attractive place for the deployment of AVs, even though the main development of the technology is taking place in the US, where there is considerable diversity of regulatory regimes across the states. However, the comprehensive legal framework may deter some prospective developers.

Ride-hailing operators such as Uber might hope to save the costs of the driver, but would need to own the vehicles and accept responsibility as the NUIC operator for oversight of operations, and possibly also to be the ASDE. This would be a large and costly change from the current capital-lite business model.

Tesla’s approach to automation has been to incorporate the hardware into existing privately-owned vehicles and to improve the software over time to get to the point of implementing what is being marketed as ‘Full Self-Driving’ capability. Taken literally, this would require Tesla to be licenced as a NUIC operatory, with oversight of its whole fleet of vehicles on the road.

Developing vehicle automation for general deployment is proving more difficult than the pioneering technology optimists had hoped. As well as getting the technology to the point of being publicly acceptable, there is an evident need to develop viable business models consistent with the expected regulatory regime.

I have previously discussed the widening of the M25 motorway between Junctions 23 and 27, where the economic benefits forecast did not materialise. Another example has now arisen.

Conversion of the hard shoulder of the J10-13 section of the M1 motorway to dynamic running was intended to reduce congestion by allowing the hard shoulder to be used as an additional running lane during busy periods. The scheme, one of the earlier ‘smart motorways’, opened in 2012 and a report of its first five years of operation, up to 2017, was published in 2021. The cost was £489m for 15 miles of widened motorway, adjacent to Luton, to the north of London.

Electronic signs tell drivers when it is safe to use the hard shoulder as a running lane, but then speeds are limited by the level of congestion, with a maximum of 60mph. This, together with some traffic growth on the route, meant that journey times were in fact longer than before the road was converted.

The main economic benefit of road investments is taken to be travel time savings. In the present case, the forecast had been for an average travel time savings of 1.5 min per vehicle in the opening year, increasing to 2.25 min by 2028. But time savings were not observed. The forecast benefit-cost ratio (BCR) had been 1.4, whereas the estimate based on the five-year outturn was negative, -0.8.

The stated conclusion, five years after opening, was: ‘In this case, the monetisation of journey time benefit is not a good measure of value for money and the qualitative evidence presented in the evaluation is considered a more robust measure.’ The failure of forecasting was attributed to limited prior experience of such smart motorway conversions.

In view of this marked discrepancy between forecast and outturn, I made a Freedom-of-Information request to see the detailed reports of the traffic modelling and economic analysis of the proposed investment. The original modelling had been for a widening from three to four standard lanes in each direction. The model was adapted for the use of the hard shoulder for the extra lane. The model drew upon the East of England Regional Model, a variable demand model, for data to input to a local traffic model for the section of the M1 involved. As elsewhere, the traffic modelling employed the established SATURN package.

The traffic modelling projected increased traffic volumes, comparing the investment case with the ‘do minimum’ case without the investment, as well as journey time reductions in the range 4-15% for the opening year, depending on section of the road and time of day. The economic appraisal used the output of the traffic model as input to the standard TUBA economic model to project the economic benefits. The main benefits were time savings to business users of £456m pv, offset by increased vehicle operating costs (VOC) of £61m. There were time savings to consumers of £170m, more than offset by increased VOC of £197m. This suggests that the increased road capacity is attracting local users, such as commuters, who save a few minutes of their journey by rerouting to the motorway, at the cost of more fuel use for a longer trip, as illustrated in the screenshot from Google Maps above. A similar situation arose in the M25 case.

The forecast BCR from the TUBA model was 3.5, which is different from the forecast of 1.4 provided in the year five report (above), apparently on account of a change in how increases in revenue from fuel taxation are required to be treated, whether as offsetting the scheme costs or as an element of the benefits from the investment.

Conclusion

Transport models are complex and opaque. Generally, little effort is made to valid forecast against outturn. The present M1 case demonstrates a marked failure of a model to forecast the observed traffic flows and speeds five years after opening.

More generally, monitoring traffic flows and speeds provides only limited information about the validity of a model that projects economic benefits for different classes of road user. The outturn of a widening scheme that matched projected flows might arise if all the increase in traffic volumes arose from more local users taking advantage of the increased capacity to save time on local trips, thereby pre-empting benefits to long distance business users. Effective monitoring needs to track the travel behaviour of different classes of road users.

It seems likely that there is often a bias in traffic modelling of road investments to underestimate the growth of local traffic and hence to overstate the economic benefits to business users.