Your questions answered on our LTN report

It’s true that many boundary roads are very busy with motor traffic. The median (typical) value is around 11,000 motor vehicles daily and some see more (a few much more) than that.

The report is not saying that boundary roads are fine in terms of motor traffic, but rather that LTNs aren’t making the problem worse. Over the period analysed, a similar number of boundary roads saw decreases in traffic as saw increases in traffic.

The report also makes explicit that transport authorities must introduce measures to address the burden of traffic on boundary roads - including the likes of more ambitious clean air zones which are proven to substantially reduce air pollution.

Many external factors can affect individual boundary road count sites across London. For instance, a hospital nearby might have meant traffic levels stayed the same or even increased during the pandemic, or a major transport scheme may have led to a reduction in capacity and hence traffic. This wouldn’t mean the first scheme was a ‘bad LTN’ nor that the second scheme was a ‘good’ one.

Averaging across all the 174 boundary road count sites gives us a typical or aggregate picture, helping to reduce the impact of any externally caused variation. (It should be noted that it’s not ‘half of all schemes’ but ‘half of all count sites’ - so some LTNs saw reductions in motor traffic on some boundary roads alongside increases on others.)

We also don’t think it’s usually feasible to attribute large reductions or increases in traffic on boundary roads mainly to an LTN. Where there’s a drop of around 1,000 motor vehicles daily in an LTN, for instance, it’s hard to see how that might have led to a rise or a fall of 2,500 motor vehicles on an LTN boundary road. Of course, high levels of motor traffic and/or a big rise in motor traffic is still a problem - but the data here suggests that removing LTNs is unlikely to help.

Motor traffic reduction measures are urgently needed for many reasons. One of the most pressing concerns is the safety of vulnerable and marginalised road users - for instance, disabled pedestrians, who are four-five times more likely than non-disabled pedestrians to be injured by a driver.

There are of course disabled people who absolutely need access to a motor vehicle. And while LTNs do restrict through motor traffic from crossing the LTN area, they are distinct from full pedestrianisation in that all addresses should be reachable by any motor vehicle, although routes may change.

This report suggests that generally, LTN boundary roads do not become busier, which is good news. However, some journeys might still take longer because some drivers have to go a longer way around than previously. Local authorities must consider the needs of essential car users in relation to this, for instance, whether Blue Badge holders should have access to exemptions from some or all camera closures. This may lead to journeys getting faster for these essential car users thanks to there being substantially lower traffic on these routes.

There also needs to be more engagement with disabled people, and other groups like businesses, early on so design issues can be ironed out before they happen - and so that while LTN works are being done, other improvements can be made like adding more dropped kerbs.

One problem that local authorities have is a lack of resources to do meaningful, long-term engagement with local communities. Other pieces of work produced by Possible have called for such engagement to be improved and properly resourced, including for instance paying disabled people for their time to be involved in such co-design.

During this century London has seen a substantial shift away from motor vehicle usage, thanks to the implementation of road pricing measures and road space reallocation - the introduction of bus lanes in the early 2000s is one such example.

The problem is that in a dense “mega-city” like London, which experiences suppressed demand for travel, spare space for any mode of transport fills up quickly at peak times.

Central London’s new cycle tracks have quickly become congested at peak hour at key pinch points, due to what planners call ‘induced demand’ - in this case, for cycling. If that space was allocated back to cars, it would quickly fill up with cars instead.

Reducing motor vehicle usage, and reducing the harms caused by the remaining vehicles - by taking the most polluting cars off the road, for instance - needs to be the priority. It’s also vital to remember that without action, things will get worse, because a growing population will otherwise lead to more and more cars on the road.

About half the LTNs introduced during the period under study could be included in the study. This is because they had to have comparable data for the analysis being conducted. To be included an LTN had to, at a minimum, have data from one geographically located count site in the LTN or on the boundary road, with data from a specific month before the scheme’s introduction (2017 or later) and another specific month after introduction.

Some boroughs didn’t produce any monitoring reports with data on traffic changes at all, so those LTNs couldn’t be included. Other boroughs, like Ealing, did produce reports but only gave graphs and as such, didn’t provide the numerical data needed. Others just had problems with their data such as having no ‘pre’ data for a specific month.

Everything that could be included was and it is worth pointing out that this is still the biggest study of LTNs and motor traffic to date. Unfortunately not all boroughs provided sufficient quality - or even any - data for us to include them and in the report the need for more high quality and publicly available ‘open’ data is emphasised.

The report includes both means and medians and both are useful in giving an overview of what is a complicated picture.

The medians or middle values provide a measure of what is ‘typical’ for a given metric which can be very useful, for instance, traffic levels before an intervention on a boundary road. However, medians do have limits. For example they:

Don’t aggregate so you can’t multiply them up to give a picture of all the traffic counted.
Don’t capture skewing where - for instance - negative values are systematically larger than positive values or vice versa.
Can be confusing as they refer to different cases for different metrics. So the median ‘pre’ count added to the median change will not produce the median ‘post’ count, as the median for each of those is likely referring to a different count site. By contrast, a mean ‘pre’ count added to a mean ‘change’ should - if there are no errors! - produce the mean ‘post’ count.

The mean has its limits too. Unlike the median, it does get skewed by more extreme values, so is less good as a measure of a ‘typical’ case. Salaries in the UK’s highly unequal society, where the ‘median’ income is much less than the mean, because of a small number of very high earners is a good example of where median values are more helpful in providing a picture of typicality.

In order to aggregate and draw conclusions across a full set of cases, it is the mean that is more suited - if you wanted to estimate total earnings in the UK and you had the median and the mean, you’d use the mean.

This is why in considering whether there is traffic evaporation across these schemes as a whole, it makes sense to compare the mean values rather than the medians. The means take into account the size of reductions and increases in motor traffic across the full set of schemes studied, and are more appropriate to use for this than the medians - especially given the distribution of values is so different for internal and boundary roads.

Pneumatic tubes - rubber tubes across the road - are the most common type of traffic counter used in our data. Like other options their accuracy is not perfect and their performance is dependent on how they are set up, but we have confidence that the data is accurate enough for the purpose of this analysis.

For instance, a study by McGowen and Sanderson who described themselves as ‘suspicious’ of pneumatic tube data found that when using short counting periods (e.g. 15 minutes), error rates were much higher than manufacturer claims of around 1%. Errors would sometimes involve over- and sometimes under-counting. However, these errors cancelled out over longer periods such that ‘for daily counts, the road tubes have small error rates consistent with those reported by the vendors.’ This research uses daily averages based on longer counting periods, so accuracy should be much better than if short counting periods were used.

As mentioned in the report, there are known problems with pneumatic tube data. Parked vehicles can create problems, particularly on residential streets, which is why you will often notice that the tubes do not run to the edge of the road where motor vehicles are commonly parked. This also contributes to their high error rate for cyclists, also due to cyclists weighing much less than other vehicles and often being missed even if they do cross the tubes.

As these tubes can detect vehicles at different speeds, the 10mph question comes from the default settings used by some counter providers to generate reports. Metrocount’s manual states that 10kph (6mph), or in some cases 10mph, is the default lower limit set when reports are generated by the software.

This sits alongside other default settings such as which vehicle classes to include, and what to set as the included directions. Metrocount says that if operators want to incorporate vehicles moving at speeds above or below a default range, they should change these defaults such that the reports generated will include the appropriate set of vehicles.

Hence, there are three potential issues related to slow speeds/congestion. Firstly, in Enfield, a contractor initially reported data for a scheme that included traffic under 10kph (6mph), but at follow-up failed to change the default setting such that the follow-up report excluded such traffic. This is an error which should not have happened. One should always compare ‘like with like’ in terms of counting parameters, before and after.

When Enfield re-analysed the 2021 ‘post’ data, including the additional vehicles travelling at very low speeds, traffic did increase at some locations but by a relatively small amount. Vehicles recorded as travelling at under 10kph made up 2% of all vehicles recorded as passing the four counters sited on boundary roads in the Fox Lane monitoring report ‘post’ 24 hour counts. Including these had minimal impacts on Enfield’s overall assessment of impacts.

The second potential impact on results might be if speeds of under 10kph, or in some cases another cut-off, had been excluded from both ‘pre’ and ‘post’ count reports. This would at least be comparing ‘like with like’, reducing any bias caused by the exclusion of what is likely to represent a small proportion of vehicles both ‘pre’ and ‘post’. In Enfield, such slow-moving traffic made up around 2% of all ‘post’ traffic and had such vehicles been excluded ‘pre’ as well as ‘post’, the subsequent data loss would have been smaller than 2%.

The impact of excluding very slow-moving traffic might also lead to bias occurring in either direction in the study, depending on whether motor traffic was rising or falling. In our study, the background trend, overall, between ‘pre’ and ‘post’ counts was for a decline in traffic volumes and likely in congestion, as ‘pre’ counts were generally taken before any Covid impacts. Thus, were very slow-moving traffic to be excluded, this background change would on average artificially reduce ‘pre’ counts more than ‘post’ counts.

Thirdly, results may be impacted even if very slow-moving traffic is included, because this likely small proportion of motor traffic will be less accurately counted than motor traffic travelling at more typical speeds. Operators seek to reduce such bias by placing tubes away from junctions, which seemed generally to be the case in our study. As before, the use here of ‘pre-post’ comparison will also reduce the impact of such bias as where such factors exist, they will likely be present to at least some extent both ‘pre’ and ‘post’.

It is also worth noting that congestion and low speeds can bias results towards over- as well as under-counting. A recent analysis in New York by a company that sells video data collection alternatives found that congestion and low speeds led to ATCs over-counting and hence overestimating the number of motor vehicles, by 13%:

The over-counting in that case could be due to the presence of freight vehicles which at low speeds were detected as multiple separate cars, because of the time taken for each axle to cross the tube. By contrast, under-counting might happen where two cars cross the tube at exactly the same time, which is more likely with multiple lanes of general traffic and/or very high volumes of motor traffic.

Despite not being perfect, pneumatic tubes continue to be used globally as a popular method of gathering count data and analysing changes in volumes and speeds across a range of settings. Many other methods (e.g. radar, CCTV, manual counts) themselves have drawbacks including accuracy as well as cost. Like pneumatic tubes, other methods will suffer from operator error from time to time.

However, we have confidence that this data is sufficiently good to be able to draw the conclusions drawn in the report, while noting the need for improved data collection and reporting. Additionally, the sensitivity analysis excluding one borough at a time helps us to check that results are not being unduly influenced by one authority whose counter placement may have been less careful than in other boroughs. Within the context of a systematic review of data from many count sites managed by different organisations, ultimately there is neither the resource nor the expectation for raw data to be separately verified by the review team.

The growing use of more accurate methods for monitoring, such as machine learning sensors, is positive, especially as these are able to count pedestrians and cyclists accurately as well. However, no method is inherently error free and machine learning sensors require operators to carefully consider camera placement, for instance, to avoid occlusion from trees in summer time.

As explained above, the data looks good enough to draw conclusions from. Perhaps the most ‘perfect’ data might in theory come from recorded video data, then manually coded by several different coders to ensure reliability. To account for variation, then as is usually done with the ATCs, one would want a couple of weeks of 24/7 data for each count site. For 587 count sites, both ‘pre’ and ‘post’, this would then entail collecting and double coding 394,464 hours of video data, an undertaking that would be infeasible in time and expense. Other studies mentioned in the report have looked more in depth at more granular data (e.g. Yang et al.), whereas the contribution of this research is in the wide range of schemes it considers.

These LTNs were introduced during Covid-19 times which did not have typical traffic.

In terms of the research, the changes (or the lack of changes) found are not just due to Covid-19, nor other factors like seasonality, because background changes were taken into account. Data on monthly changes in traffic from TfL in London’s three functional zones were used to achieve this. It has helped us to be confident in our results, as the TfL data does clearly show fluctuation not just due to Covid-19 but also to the time of year and wider longer-term trends, which could then be controlled for as well.

For greater transparency the report presents both unadjusted figures and figures that take into account the background changes in traffic that might have been expected during the monitoring periods. This allows us to paint a picture not only of the impact of LTNs once background trends had been accounted for, and what changes were actually experienced on the monitored roads, which is also interesting and important.

The boundary roads vary, some are main roads and red routes, some are not. It depends where local authorities monitored, based on where they thought might potentially be affected by the LTNs.

This dataset was cross-checked against another dataset of where LTN boundary roads were thought to be, and in a few cases the decision was made that the local authorities were wrong - for instance, they were monitoring a road well outside the scheme area that couldn’t plausibly be affected by it.

But in nearly all cases it made sense for them to monitor those sites for potential displacement effects. Arguably it’s even more important for them to monitor smaller boundary roads which may be particularly unsuited to deal with any increase in motor traffic.

The phrase ‘traffic evaporation’ can be a bit unhelpful - it does imply something magic. However, it is just the reverse of the well-established phenomenon of ‘induced traffic’ - i.e. if you build something, a new motorway or a new bike track, then, as long as it goes somewhere people want to go, it quickly fills up often with new trips that weren’t being made before.

Traffic evaporation has been shown in a range of studies, including most recently a study into superblocks in Barcelona by Samuel Nello-Deakin. It relates to people changing how, when, and where they travel. This might mean someone cycling rather than driving the exact same trip, because driving is a bit more difficult and cycling a bit easier and more pleasant.

But it could also quite easily refer to changing where we shop, for instance, walking to the local high street rather than driving to a supermarket for some of our shopping trips, because the local high street has become easier to access on foot.

Or even to not making some trips at all. That would mean not driving to the local shop for a pint of milk as it is a bit more difficult (maybe there is less car parking), and instead making sure in advance that we have enough milk to last the week.

Many short trips are still driven, even in London, but not all of them need to be. Changing the balance a little bit so that other options become more attractive can make a big difference, as in research into London’s Mini-Holland schemes, which had a large impact on people’s local walking.

This report features the research findings produced through ATA's project partnership with climate charity Possible. The study has been carried out to a suitable academic standard for publication in a peer reviewed journal, and the research has been submitted to one, in the form of an academic article (shorter and more concise than the report, as more background methodological knowledge would be expected of an academic audience). This is not unusual for research-policy collaborations, and similar happened when Possible released a report by ATA on LTNs’ equity distribution, with an article based on the analysis later published in Journal of Transport Geography.

Possible and ATA work together to identify research topics and agree a scope and investigative team, and this report has had editorial and presentational input from Possible's communications team. However, Possible have no influence over the methodology used to conduct the research or the results, and are not involved in any academic outputs such as journal articles or conference papers.

Your questions answered on our LTN report

Look, I live on a boundary road around an LTN, and I’m telling you it’s packed with traffic. Why won’t you believe me?

OK, so half the LTNs reduced traffic on boundary roads but half increased it, so that’s a lot of bad LTNs.

What about disabled people and people who need home visits? Don’t we matter?

If all this is working, why is London still one of the congested cities in the world?

Hang on, you haven’t looked at my LTN. Is that because you’ve cherry picked the data?

Why have you used the mean and not the median? What’s the difference?

I heard that traffic counters don’t capture vehicles going slower than 10mph. Is that true? If traffic is just queuing in congestion, does that mean you missed loads of cars?

Why didn’t you collect manual count data instead, which would have been better?

Why is this data from during the pandemic? That doesn’t seem like typical traffic to me.

Your definition of ‘boundary roads’ doesn’t include the main roads and red routes. Isn’t that where the traffic actually ends up?

Traffic can’t literally ‘disappear into thin air’. If the mean reduction on internal roads is ten times higher than the mean rise on boundary roads, the traffic must just be somewhere else or the data must be wrong.

Is it peer-reviewed? Why should we believe it when a climate charity produces research which provides conclusions which it clearly agrees with?