Changes in motor traffic in London ’ s Low Traffic Neighbourhoods and boundary roads

This paper meta-analyses traffic data extracted from monitoring reports for 46 Low Traffic Neighbourhood schemes in 11 London boroughs introduced between May 2020 and May 2021. Schemes are controversial with still limited academic evidence on impacts. The analysis covers internal and boundary roads, looking at actual changes in motor traffic, and at what changes might have been expected based on background trends in London ’ s three ‘functional zones ’ (Central, Inner, and Outer). All metrics show substantial relative declines in motor traffic on internal roads. For instance, a mean ‘pre-LTN ’ traffic volume of 1,780 dropped to 930 ‘post-LTN ’ , against a small projected decline to 1,745 if background trends were followed. By contrast, the schemes are only marginally associated with change in traffic volume on boundary roads. While there are inevitable data limitations associated with the use of routine monitoring data, this research provides some support for the position that LTNs can form an effective part of wider strategies to reduce motor traffic and its associated disbenefits. Monitoring data should be publicly shared regionally and nationally to aid analysis.


Introduction
Since the start of the Covid-19 pandemic, Low Traffic Neighbourhoods have been introduced in London, and to a lesser extent, other parts of England.LTNs are schemes that seek to reduce motor traffic in primarily residential areas, using traffic management measures such as 'modal filters' to block general motor traffic while permitting walking and cycling. 1 They seek, like many other traffic management measures, to combine 'carrot' and 'stick' by encouraging use of sustainable transport while discouraging car use.The design principle of LTNs is not new: in the Netherlands, this approach to urban planning is called 'unbundling', referring to the goal of separating much motor traffic from people walking and cycling (Schepers et al., 2013).However, research is still emerging on the retrospective introduction of these measures, with stronger evidence for increased walking and cycling than for decreases in car use (Aldred et al., 2021a,b).
London offers an opportunity to study the impacts of such schemes at scale.Within six months of the pandemic starting, newly introduced LTNs in London covered 4% of the population (Aldred et al., 2021a,b).By March 2022, a hundred schemes had been introduced, although a minority had been removed.This scale is noteworthy given the city's complex governance.In London, most transport planning is done by 33 individual districts, which control more than 90% of the roads.Transport for London (TfL) manages 'strategic roads' (many but not all 'A' roads) and oversees much funding to districts, including that provided during Covid-19 for emergency schemes such as LTNs.Generally, districts decide where such schemes should be placed, in negotiation with TfL, which produces strategic analyses that boroughs may refer to in guiding scheme placement.
Planning approaches differ across London districts.One-third chose not to implement any LTN schemes and others vary both in the number of schemes and in the extent to which those schemes are located in more or less deprived or diverse areas (Aldred et al., 2021a,b).Processes by which districts decide whether and where to implement LTNs can be opaque and differ depending on, for instance, technical capacity as well as political leadership (Furlong et al., 2023).This has led to differences in the equity of their distribution (Aldred et al., 2021a,b) and may also potentially mean their impacts vary by district.Approaches to monitoring and evaluation also vary substantially, with a few London districts producing lengthy reports covering a range of potential LTN outcomes including metrics such as injury or crime numbers, most using only traffic counts to measure changes in vehicle numbers at specific locations, and other districts not reporting on any monitoring or evaluation metrics.Given this and the devolved nature of London transport planning, there is no city-level monitoring and evaluation of these schemes.
Within the UK (and beyond it, e.g.Nello-Deakin, 2022 on Barcelona's Superblocks), much debate around the impacts of these types of schemes has focused on possible disbenefits for residents and road users on roads to which motor traffic may be displaced from inside an LTN (usually called in the UK 'LTN boundary roads'; Hickman, 2021).Previous studies examining motor traffic capacity reduction suggest such schemes typically induce a mixture of motor traffic redistribution and traffic evaporation, for instance, through mode shift (Cairns et al., 2002).However, Cairns et al. (2002) found that the balance varied substantially from almost all redistribution to almost all evaporation.The nature of this balance will make a large difference to the overall impacts of a particular group of schemes.In response, this study systematically reviews and meta-analyses routine monitoring and evaluation data from 46 London LTN schemes to draw conclusions about changes in motor traffic within the schemes and on their boundary roads.

Background
LTNs sit within a broader set of interventions that aim to reduce motor traffic and encourage mode shift away from the car.Specifically, LTN schemes seek to cut such traffic both on streets directly affected by traffic reduction measures and across a wider area, and also aim to increase walking and cycling by creating more pleasant neighbourhood streets.There is increasingly evidence demonstrating various benefits of such schemes, much of it from London (Yang et al., 2022).In London, evidence covers schemes introduced in Waltham Forest borough (Outer London) and from London's 2020 'emergency' LTNs.This suggests that Waltham Forest and/or emergency LTNs in London have (i) increased levels of walking and cycling among LTN residents (Aldred et al., 2021a, b;Aldred and Goodman, 2020;Aldred and Goodman, 2021), (ii) reduced car ownership among LTN residents (Goodman et al., 2020a,b), (iii) reduced injuries and risk on roads in LTNs, particularly for pedestrians (Laverty et al., 2021, Goodman et al., 2021a,b,c), (iv) reduced street crime within LTNs (Goodman andAldred, 2021, Goodman et al., 2021a,b,c), and (v) reduced kilometres driven per vehicle by LTN residents (Goodman et al., 2023).However, no study has yet analysed pan-London impacts on motor traffic inside LTNs and/or on their boundary roads.
Unlike low emission zones or road pricing schemes, LTNs do not involve direct financial disincentives to driving.Instead, they seek to make driving less convenient while making active travel more pleasant.Many of the more worrying concerns about the fairness of LTNs have centred on motor traffic being displaced to LTN boundary roads, leading to disbenefits for those living on, visiting, or using these roads, as well as knock-on impacts on essential car, taxi, or bus journey times.One option in studying these issues is to measure possible disbenefits of increased motor traffic directly.Encouragingly, the papers referenced above on road injuries did not find displacement of injuries to boundary roads.Similarly, analyses of Waltham Forest and 'emergency' LTNs focused on fire response times found no detrimental impact (Goodman et al., 2020a(Goodman et al., , b, 2021a,b,c),b,c).
More recently, Yang et al. (2022) analysed three Islington LTNs, concluding that the LTN implementation led to reduced pollution both inside LTNs and on their boundaries.However, the three Islington schemes might be atypical of LTNs introduced across London; and/or impacted by factors peculiar to that borough.Yang et al's (2022) study used detailed air pollution data provided directly by the local authority, which is more feasible if dealing with a small number of cases.In contrast, Bosetti et al. (2022) discussed 9 scheme reports with information about traffic changes on scheme boundary roads; however, due to their use of already summarised data (rather than data from individual count sites) and the lack of normalisation to background traffic trends, few conclusions could be drawn.
Other research covers related types of scheme in Barcelona, Spain.A recent article by Nello-Deakin (2022) examined road space reallocation in Barcelona's Eixample district.The Eixample schemes were implemented at similar times to London's emergency LTNs and similarly involve substantial reallocation of road space from through motor traffic, with related concerns about traffic displacement.The study used open traffic data provided by the local municipality, generally monthly average traffic counts.Nello-Deakin (2022) reports finding significant 'traffic evaporation' across the intervention area after accounting for background changes, with little change on parallel neighbouring streets (in a grid system, these are analogous to LTN 'boundary roads' with similar concerns about traffic displacement).Specifically, there was a very small relative mean traffic increase on neighbouring roads of 0.7% (median + 3.9%).Nello-Deakin's study is most similar to this research, although the methods differ somewhat due to the spatially and temporally dispersed nature of London interventions and their follow-up, and the lack of collated publicly available data, hence the need in London to conduct a systematic review of such material.
This study explores changes in motor traffic on primarily residential streets within London LTN schemes and on generally busier 'boundary roads', on which concerns about traffic redistribution have focused.The aim is not to come up with a yes-or-no answer regarding traffic displacement versus traffic evaporation, given wider evidence suggests there tend to be a combination of these effects (Cairns et al., 2002), although the balance may vary substantially.The paper represents an early assessment of LTNs in London, considering underlying differences in the changes in overall traffic within Greater London experienced during the pandemic.This comparator is based on available data from Transport for London's designation of London into three large heterogeneous areas (Central Activities Zone, Inner London, and Outer London).

Materials and methods
This paper examines levels of traffic reduction within London LTNs, and whether there is a systematic and/or substantial impact on motor traffic on boundary roads.The research question guiding the systematic review conducted was:

What does data from existing monitoring and evaluation reports indicate about the impact that Low Traffic Neighbourhoods in London have had on motor vehicle traffic, on both internal and boundary roads?
The analysis separately examines impacts on internal and boundary roads in response to the framing of the debate and questions about unintended consequences.Both critics and promoters of LTNs typically agree that LTNs if successful can produce substantial reductions in motor traffic inside their borders, while disagreeing about impacts on what both call 'boundary roads' (Hickman, 2021).Specifically, the argument centres on whether most or all motor traffic removed from LTNs simply relocates to their boundaries, or whether there is sufficient 'traffic evaporation' (for instance, through mode shift, or other mechanisms) that traffic impacts on boundary roads are small or minimal (Aldred et al., 2021a,b).The analysis conducted here does not exclude the existence of other, perhaps more complex or longer-term impacts that were not measured (e.g., whether LTNs in East London may over time encourage more car-dependent residents to move to West London) However, it does begin to study an under-researched issue that cannot be easily resolved at scheme or district level, due to many factors affecting traffic in any one site and the typical lack of controls or comparators within local authority analysis.
A. Thomas and R. Aldred

Document search
By mid-2022, an increasing number of individual reports on London LTN schemes had been published.Local authorities implementing LTNs have been much more likely to monitor motor traffic than to monitor other transport-related outcomes such as bus journey times, cycling flows or (even more so) walking flows.This provided an opportunity to analyse this data and hence draw conclusions across scheme about LTNinduced changes in motor traffic.However, little work has been done so far to analyse this monitoring and evaluation data.Perhaps this is because the format, use and presentation of data, and calculation of impacts (where provided) varies so substantially, making synthesis challenging across multiple schemes and boroughs.These reports also sit on different websites and website sections belonging to London's 33 district councils (known as boroughs).
In May and June 2022, we systematically searched London local authority websites for monitoring and evaluation reports of all low traffic neighbourhood schemes from the 21 (of 33) London districts known to have installed LTNs since March 2020. 2 The search primarily used Google Advanced search tools, i.e., searching within local authority websites using LTN-related keywords, including the names and 'brands' used to refer to specific schemes or by specific boroughs (e.g.'People Friendly Streets').We had a pre-existing dataset of 100 London LTNs introduced up to March 2022 and so were able to search for each of these individually.
In total we found 106 documents covering 69 LTN schemes from 16 local authorities (Fig. 1).Of these, 57 schemes had reports with both before and after data on motor vehicle traffic on boundary and/or internal roads.For 46 of these schemes, traffic data was extractable for meta-analysis (our minimum requirement was tabular data, with at least one internal or boundary road data point providing 'pre' and 'post' count data for that specific site from a defined month from 2017 onwards, and the locations of count sites).The earliest of the schemes we could include was introduced in May 2020, and the latest in May 2021.

Data extraction
The 46 schemes that we were able to include in this report were drawn from 11 different local authorities.From these reports, we extracted measures of pre/post changes in motor vehicle volumes at each included site.We extracted this data at the traffic counter level, recording the spatial location and whether the count point was indicated by the local authority as lying on a boundary road or within the LTN area. 3We recorded this information on spreadsheets and a GIS file linked by internally generated count point IDs and held regular meetings between the two authors to discuss uncertainties or disagreement from borough classification.

Count sites
Across all 46 scheme reports that met our inclusion criteria, we found 641 count sites with spatial location data.We then had to remove 40 points (6%) due to data issues that meant that they could not be analysed, primarily the lack of a valid 'pre' and/or 'post' count date including a month.Some minor data issues were identified (such as using 5 day counts rather than the more typical 7-day average).These were included in the main analysis but marked for exclusion during a sensitivity analysis.Districts almost always classified sites as being 'internal' or 'boundary'.As discussed above, much concern about LTN impacts centres around possible negative impacts on boundary roads, and so this distinction was crucial to our analysis.Districts usually mapped the extent of LTNs, showing where they considered a scheme's boundaries to lie.Often, this extended to an entire 'traffic cell', in the case where a new LTN was immediately adjacent to an area that already had some traffic restrictions.As the Active Travel Academy (ATA) we had for other analysis already collated our own dataset of LTNs introduced since March 2020, although our dataset focused on identifying newly quietened areas.Fig. 2 shows an example in Hackney where new LTNs are coloured in yellow and boundary roads in purple.Typically, these were immediately adjacent roads to which a journey previously going through an LTN would most likely divert.
The ATA dataset was used to cross-check districts' classifications for internal and boundary roads, and in almost all cases the districts' classifications were retained.The cross-checking did lead to the removal of 16 points (2%) from our main analysis, as both reviewers were confident that these lay outside a scheme area of influence.Ten were originally classified by a borough as a boundary road and five as internal roads (one was not stated).All 16 we felt were neither subject to potential traffic displacement, nor within the traffic cell of a scheme, and in some of the 16 cases we felt the local authority might be intending to monitor the road more for background information on traffic in the area rather than to study LTN impacts.This left us with 587 points from the 46 reports that could be included in the analysis.These used a variety of 'pre' and 'post-intervention' months, and we took the latter as being the latest available monitoring point where more than one was provided.Baseline counts were taken between January 2017 and May 2021; while post-intervention counts were taken between July 2020 and February 2022.

Fig. 3 below demonstrates the lack of included count points in
Central London (dark green) and Outer London (pale green).While both City of London and Westminster introduced some (albeit atypical) LTNtype schemes, we could not find reports for these with traffic count data. 4Outer London is more sparsely covered than Inner London, due to a lack of monitoring reports available for some boroughs (e.g.Croydon, Merton), others not providing tabular count data in reports (e.g.Ealing), and Outer boroughs having been less likely than Inner boroughs to implement schemes. 5Within Inner London, count sites cluster North and North-East (Camden, Islington, Hackney), and South-Central (Lambeth, Southwark).
From the 587 count points, 175 (30%) were boundary road count sites, and 412 (70%) were internal road count sites.Of these, we have reclassified 14 (2%) between boundary and internal, only done where (i) there was disagreement between our mapping and the borough mapping and (ii) upon double-checking and discussion, we believed that the borough had made an error.One was reclassified from boundary to internal, and 13 reclassified from internal to boundary.We conducted sensitivity analysis to explore the impacts of these reclassifications.
As might be expected, almost all (404/412) of the internal streets are classed as local or minor streets (Table 1).However, there is more variability for boundary roads.While around two-thirds are more major A and B roads (110/175), one third are classed as local or minor streets.  Westminster's schemes were primarily aimed at supporting outdoor dining, so motor traffic reduction would not have been a main aim. 5 As we are using TfL's Functional Zone classification which is based on the GLA definitions of Inner and Outer, Newham and Haringey count as Inner, and Greenwich as Outer London.ONS uses a different classification when analysing and presenting Census data.

Calculating background trends
We obtained data from Transport for London (TfL) (Appendix A and Fig. 4) to allow us to calculate background trends month-by-month, and hence, estimate the change that might have been expected to occur anyway in that area's functional zone (Central, Inner, Outer) if no scheme had been implemented, between 'pre' and 'post' intervention months.London's functional zones have different traffic characteristics and showed different trends in terms of traffic volumes during Covid times and longer term.Most notably, Central areas saw very large Covidera changes, which were less sharp and less sustained for Inner or Outer London (see Fig. 4).While these zones are large in the context of a global mega-city, we take some comfort from being able to separate out very different trends in the Central Zone, where relatively few LTNs were implemented, from the much more similar trends in the Inner and Outer Zones.The use of an albeit imperfect comparator/control already is an additional contribution that an academic study in this area can make, compared to typical local authority reporting.Having more comparator zones would be better, in allowing a more accurate normalisation based on borough-specific trends for example; but the analysis was constrained by the available data.
The TfL dataset covered January 2017-June 2022, showing the typical percentage difference between average daily motor traffic in that  We used this dataset to calculate expected average daily motor traffic flows for each count site at the post-intervention point, based on the background trends experienced in its functional zone between baseline and post-intervention month.We did this by dividing the adjustment factor for the post-intervention date by the adjustment factor for the baseline date and then multiplying the resultant background trend factor by the observed baseline flow: Expected Flow = (Post-Intervention Adjustment Factor/Baseline Adjustment Factor) × Baseline Flow.
The Expected Flow can then be subtracted from the actual postintervention flow to calculate the difference between observed and expected traffic levels.This allows us to estimate deviation from the background change expected in each count point's particular functional zone (Central, Inner, Outer), between 'pre' and 'post' intervention count months.

Analysis
Our main analysis presents before and after changes for internal and boundary roads, providing both actual changes and changes relative to background change in the relevant functional area of London.In other words, it separates traffic changes inside and on the boundary of LTNs introduced between May 2020 and May 2021 in London (for which data were available) from changes that might have anyway been expected over the relevant monitoring period in that functional area.The analysis provides both means and medians and distributional plots giving further information on variation and skews.Both absolute and percentage changes are generally provided to give a rounded picture of the impacts of schemes.We provide analysis showing how many count points have 1,000+ motor vehicles per day 'pre' and how many 'post', i.e. the extent to which monitored streets are pedestrian-priority, in line with the level estimated in the UK Department for Transport's Manual for Streets (DfT, 2007, 83).
Further analysis splits results by functional area, although noting that Outer and particularly Central London have relatively few count sites compared to Inner London.It also compares traffic changes on boundary roads depending on whether these are A, B, or local/minor roads.Finally, we briefly summarise key points from a set of results for sensitivity analysis re-running the main analysis firstly eleven times to sequentially exclude one of the boroughs, followed then by the main analysis re-run but (i) changing the 14 points reclassified by us back to the borough definitions (ii) using the ATA's definition of LTNs and boundary roads (which results in some points being classed as 'external'), (iii) removing data where the counts referred to 5-day rather than 7-day averages or where baseline counts had been normalised or estimated using telemetry services, (iv) including points deemed 'external' in the main analysis using their original borough definitions as either boundary or internal.The sensitivity analysis is presented in full in Appendix C.

Strengths and limitations
The focus of the paper is an aggregated analysis across multiple sites, as we believe that is where the power of the dataset lies (its ability to examine the typical or total impact of many schemes across London).It avoids looking at impacts at an individual-LTN level, where we there are handfuls of count points at most and where results may be atypical but which are likely to have very little or no impact on overall means or medians.The sensitivity analysis includes looking at the effects of removing individual districts, finding little impact.
Meta-analysing 46 schemes from 11 districts provides a broad coverage and includes some schemes that have since been removed (e.g. Hills and Vales, Greenwich) as well as many still in place.However, there are inevitable limitations.Firstly, data gaps are not random.While we found data issues across districts, some either did not provide any reports at all or provided them without spatially located tabular count data.Hence, we cannot include schemes from Tower Hamlets or Ealing, for instance.Across London during this period, many different schemes such as School Street closures and temporary cycle lanes were happening (Thomas et al., 2022), alongside factors related to Covid-19 and other non-transport issues, and ongoing major schemes such as gyratory removal.Hence, it is challenging to attribute change to a specific intervention.By aggregating results from many schemes across functional zones, controlling for background trends, we improve our ability to identify possibly causal relationships.However, it is still possible, for instance, that LTNs tend to be co-located near to other interventions, such as temporary cycle lanes, and that some effects found may be due to these.LTNs were not introduced randomly across London but were concentrated in more deprived areas (Aldred et al., 2021a,b), and differences in area composition may affect trends in motor traffic levels.
There are known issues with data quality.Usually, reports used Automatic Traffic Counters (ATCs) to monitor traffic, in most cases 'tubes' across the road.These are imperfect.Parked or very slow-moving motor traffic may affect results; although in most cases, count sites were placed away from junctions where queueing is likely, which should reduce this problem.Data problems due to parking may be more an issue on internal residential roads than on boundary roads.Adjusting for expected changes should help control for such bias as that data too is largely based on ATCs.We have not accessed raw data directly from counters, as this would not be feasible for so many count points, schemes, and boroughs (and data may be held by contractors).It is possible that authorities or contractors have made errors 6 (in one report a clearly wrong count was given for one site, for instance).We believe that a small number of undetected errors should not bias the overall results; and sensitivity analyses assess if any borough's exclusion substantially changes the results.
Our background trend analysis has limitations.Disaggregating by functional zone is clearly superior to using background trends for Greater London, which would be strongly skewed by extreme changes in the Central Zone.However, it is still broad brush and individual district trends may vary; and it is derived from the TfL network of counters which tend to be on larger roads, so may be more reliable in relation to boundary than internal roads.The normalisation does however make only a small difference to the results overall, probably because districts tried to avoid use of count data collected when Covid measures were at their peak and background variability highest.Finally, there are limitations to what one can conclude from average daily motor traffic counts.For instance, changes in motor traffic volumes do not correlate linearly with congestion, which is time-specific, nor do they directly map onto air pollution where speed and other variables play an important role.

Schemes included in the study
Table 2 shows the count sites included for each borough.More detail on each borough's schemes is provided in Appendix 2.

Main analysis
Of the 412 internal road count sites, 303 (73.5%) saw any decline in motor traffic, compared to 109 (26.5%) which saw any increase.For the 175 boundary road count sites, 83 (47.4%) saw any decline in motor traffic, against 92 (52.6%) which saw an increase.Below we present the results of the main analysis, using summary statistics as well as distributions.Table 3 presents the median and mean results together for internal and boundary roads, also outlined in accompanying text.

Internal roads
Whether measured through median or mean averages, internal roads see a substantial relative reduction in motor traffic -of almost half for mean averages, and almost a third for medians.Adjusting for expected changes made relatively little difference to the percentage reduction.The robustness of results to different measures highlights the systematic impact of LTN schemes within their area, shown also in Fig. 5.
Specifically, a median baseline of 1,220 vehicles per day fell to 662.Had the expected trend been followed, the median 'after' count would have been 1,199.The actual median change was a reduction of 363 motor vehicles per day, and the median difference from the predicted change was a reduction of 321 vehicles per day.The actual median percentage change was a 33.3% reduction, and the median difference from the predicted percentage change was a reduction of 31.9%.
A baseline mean average of 1,780 vehicles per day reduced to 930.Had the expected trend been followed, the mean 'after' count would have been 1,745.The actual mean change was a reduction of 850 motor vehicles per day, and the mean difference from the predicted change was a reduction of 815 vehicles per day.The actual mean percentage change, calculated by dividing the mean change by the mean 'pre' figure, was a 47.8% reduction, and the mean difference from the predicted percentage change a reduction of 45.8%.  At the time of finalising this article, specific questions were being raised about a scheme report from one borough -Enfielddue to a contractor error in collecting follow-up data.While we have left this data in the paper, we note that removing Enfield from the dataset makes little difference to results.

Boundary roads
By contrast, the average boundary road saw very little change in motor traffic.Whether this has been maintained following 'post' monitoring will depend on whether the background small reduction in traffic during those periods represents a longer-term trend.This in turn will partly depend on the introduction and maintenance of measures to discourage driving and to support walking, cycling, and public transport.
Specifically, a median baseline of 11,034 vehicles per day grew very slightly to 11,074.Had the expected trend been followed, the median 'after' count would have been 10,526.The actual median change was a rise of 106 motor vehicles per day, and the median difference from the predicted change was an increase of 242 vehicles per day.The actual median percentage change was a 1.2% increase, and the median difference from the predicted percentage change was an increase of 4.2%.
A mean baseline of 11,706 vehicles per day reduced to 11,505.Had the expected trend been followed, the mean 'after' count would have been 11,429.The actual mean change was a reduction of 201 motor vehicles per day, and the mean difference from the predicted change was an increase of 77 vehicles per day.The actual mean percentage change, calculated by dividing the mean change by the mean 'pre' figure, was a 1.7% reduction, and the mean difference from the predicted percentage change was a rise of 0.7%.This is the same as in Nello-Deakin's study, which had a median adjusted change similar to this study at +3.7%.
As in Nello-Deakin's research, average differences are typically modest as means or medians.Fig. 5 highlights the lack of a clear systematic pattern, by contrast to impacts within LTNs.The actual changes on boundary roads are in different directions for medians and means, although both are small (+1.2% vs. − 1.7%).Taking perhaps the most 'negative' result, roads with around 11,000 motor vehicles per day typically saw a rise of 242 motor vehicles daily during the monitoring period in question, compared to background trends in the relevant functional areas.In practice, due to those background trends being towards slight decline, the typical actual rise was half this (140; plausibly within a margin of error).

A. Thomas and R. Aldred
declines.There is a long tail of outliers seeing very substantial declines, and only two outliers showing substantial growth.Overall, there is a clear tendency for traffic to fall on internal roads, which we consider highly likely to be caused by the schemes.
By contrast, boundary roads show a much more 'normal' distribution; with variation either side of the mean and median (+77, +242), not just identified outliers but also within the curve itself.Combined with the relatively small 'typical' or average changes at the centre of the distribution, 7 this may largely highlight the contribution of exogenous factors to variation at an individual count site, compared to the wider functional area trends from month to month.

Further analysis
As noted above, in the UK a level of 1,000 motor vehicles daily, or 100 at peak hour, is often taken as a tipping point for pedestrians being able to use the entire carriageway (DfT et al., 2007).Thus, we explored how this changed for internal roads (no boundary road count site recorded less than 1,000 motor vehicles daily at baseline).At baseline, 241 of 412 internal roads (58.5%) carried over 1,000 motor vehicles per day.After the interventions, only 139 (33.7%) did.Of the 273/412 internal roads that had under 1,000 motor vehicles per day postintervention, 123 of these were newly quietened.A further 21 roads moved from having under 1,000 to over 1,000 motor vehicles per day.
Table 4 compares changes on boundary roads depending on their classification.It shows that changes on A roads are relatively small both in numerical and in percentage terms, and sometimes in different directions.The largest impact metric is for median difference from predicted volumes, at 292 motor vehicles per day.By contrast, the means show 69 fewer motor vehicles per day than predicted.Local minor roads have a higher median percentage difference from predicted, at 5.0% (+204 motor vehicles daily), although the mean increase is negligible in absolute or percentage terms (+35 motor vehicles per day, or +0.5%).It is B roads that show a larger percentage increase, 6.4% greater than predicted for the median and 4.7% for the mean.Note though that there are only 33 data points and that schemes in Central London (in some cases, near major ongoing roadworks) appear to be driving some of this relatively high mean increase.
Next, we present results split by functional zone (Table 5), and briefly report on sensitivity analysis (see also Appendix C).In aggregate, Inner London schemes perform somewhat better than Outer London schemes: in terms of actual counts, 78.3% (220) of 281 Inner London internal sites saw a reduction in motor traffic, compared to 63.1% (70) of 111 Outer London internal sites.For boundary roads, the 124 Inner London count sites were almost equally split with 63 seeing traffic reduction and 62 seeing traffic growth, whereas 62.2% (23 of 37) Outer London boundary road count sites recorded traffic growth.However, note that due to data availability, many Outer London boroughs and schemes are not in the analysis, with relatively few Outer London boundary road count sites in particular.

Internal roads
For medians, Table 5 shows that the typical count point in Inner London achieves higher relative and absolute reductions than those in Outer London, on roads that are typically initially busier (baseline median 1447 motor vehicles per day compared to 843 in Outer London).When considering means, similar patterns appear.Across all London

Boundary roads
Table 5 shows boundary road changes broken down between the Central, Inner, and Outer zones.There are even fewer Central boundary road count sites than for internal roads − 13 -so we caution against taking too much from these results.For Inner London, the actual figures show no change or a small decrease depending on the averaging methods.Taking into account expected changes, the Inner London median boundary road count is 333 vehicles daily or 4.1% higher than the background predictions, while the mean boundary road count is 97 vehicles daily or 0.8% higher than might have been expected.
Outer London figures show that actual counts on boundary road sites rose by a median average of 355 and fell by a mean average of 131.However, because the background trend at these sites across the relevant monitoring periods was slightly more towards decline at Outer than Inner London sites, these become a slightly higher rise than expected compared to Inner London (6.5% median rise and 1.7% mean increase).These figures are only based on 37 sites, so caution should be taken in interpreting them.

Sensitivity analysis
Most differences found during the sensitivity analysis were very small.For boundary roads the three most prominent deviations from the headline findings occurred when (i) when non-standard counts were removed (5-day counts, counts estimated from telemetry data, baseline counts normalised to a specific month), (ii) when the schemes from the borough of Brent were removed, and (iii) roads were classified only using the borough definitions.This represented (respectively) +2.45, +1.55, and +0.92 changes from the overall figures for median % difference from predicted.
For internal roads, the three most prominent deviations from the headline findings occurred when (i) roads were classified only using the stricter ATA definitions, (ii) the points from the borough of Waltham Forest were removed, and (iii) again when the points from the borough of Brent were removed.This represented − 6.37%, − 4.46%, and − 4.24% respectively in terms of changes from the overall median % difference from predicted.

Discussion
This is the first systematic review of motor traffic volume changes associated with Low Traffic Neighbourhoods in London.We have summarised changes on internal and boundary roads across 46 schemes in 11 boroughs, with nearly 600 data points.This allows us to generalise across the schemes studied, providing median and mean averages and distributional data.Data on seasonal and longer-term changes in three functional zones allows us to separate estimated likely background changes from changes due to LTNs.While as with any study there are limitations related to data sources, methods, and analysis (see above), we believe that these results represent the best evidence so far on the impacts of LTNs in London on motor traffic both within schemes and on their boundary roads.
The results suggest that LTNs in London introduced between May 2020 and May 2021 have typically resulted in a substantial relative reduction in motor traffic inside the scheme area, with particularly strong reductions in Inner London.The typical monitored internal road is now well under 1,000 motor vehicles per day, rather than above this threshold, which may generate a qualitative change in walking and cycling experiences as also suggested by growth in active travel following LTN introduction (Aldred and Goodman, 2020).
On boundary roads, by contrast, we found little change.In terms of raw numbers, 'pre' and 'post' monitoring sites on boundary roads were similar, around 11,000 motor vehicles per day (with a very small rise if comparing medians, or a very small fall if comparing means).This became a small increase when background trends were accounted for.The picture of little change on average in motor vehicle numbers on boundary roads is good news for the potential for exemptions to limit potential journey length and hence time increases for key users, if this is judged necessary.There was some evidence that as LTN boundary roads, B roads may be more likely to see motor traffic increases, compared to A roads or local/minor streets.While B roads are often excluded from LTN schemes, authorities might consider whether local B roads could be suitable for inclusion given this has happened in some cases (e.g., in Hackney, Lambeth, and Waltham Forest schemes) and that the 8 internal count points on B roads included in this study all saw decreases in motor traffic.
Substantial variation in change on boundary roads as well as some extreme outliers highlight the danger in drawing conclusions about causation (in either direction) from a small or selected group of schemes.Even excluding outliers, a quarter of boundary road count sites saw decreases of 1,000-5,000 motor vehicles per day, while a quarter saw growth of 1,750-4,500 motor vehicles per day (adjusted for expected changes).These and even larger variations in either direction may in part or entirely result from exogenous variation or atypicality that cannot easily be controlled for.For instance, Newham's monitoring and evaluation reports highlight the likely large impact of Olympic development on boundary roads; and this will be inadequately controlled for by our functional zone normalisation.However, further research could usefully seek to identify those 'real' scheme factors that may promote motor traffic reductions rather than growth on boundary roads.
The review highlighted the importance of monitoring and evaluation, including the need in the UK for more accessible, transparent, and standardised data.This is a potential role for organisations like Transport for London, Combined Authorities, the Department for Transport or Active Travel England, and devolved national governments.Nello-Deakin's (2022) analysis of traffic reduction measures in Barcelona was facilitated by the municipal authority publishing open access monthly average traffic count figures from locations across the city.At the time of writing, London has no such publicly available data.Its provision across the UK could make it easier for academics and others to explore the impacts of a range of interventions and would permit the use of (for instance) more sophisticated normalisation approaches than we were able to use here.

Conclusions
Mean falls in motor traffic on internal roads are around ten times greater than mean rises in motor traffic on boundary roads, adjusting for background trends.We believe that this result suggests that these LTN schemes may be contributing to 'traffic evaporation' or 'disappearing traffic' (Cairns et al., 2002).Traffic evaporation refers to the many ways in which people adjust habitual behaviour in response to restrictions, which may lead to a reduction in motor traffic.This may simply mean a car trip to a specific destination is directly replaced by a walking or cycling trip.Probably more common, however, are more complex changes: for instance, a person makes fewer trips to the supermarket (by car) and more shopping trips on foot to the local shops, or combining car trip destinations to increase efficiency. 8We cannot say which of these types of behavioural response is dominant in the schemes studied here.
Specifically, this study found substantial reductions in motor traffic within scheme areas, while across boundary roads there was very little aggregate change (+0.7% mean average compared to background trends).We have not attempted to calculate overall traffic reduction due to these schemes, because aggregation is affected by the number of 8 Other research has found evidence of a relative decrease in driving (Goodman et al., 2023) or car ownership (Goodman et al., 2020a,b) associated with London LTN schemes.
A. Thomas and R. Aldred count points, and in most cases, more counters could have hypothetically been placed (particularly on internal roads, more numerous than boundary roads).However, the results indicate that motor traffic has been reduced, and only a small proportion re-routed to boundary roads.This is suggested by the mean increase of 82 vehicles per day on each boundary road being much lower than the mean reduction of 815 vehicles on each internal road.This indicates that LTNs can contribute to wider traffic reduction objectives.It is encouraging to see on average so little change in boundary road traffic volumes.Perhaps the widely fluctuating Covid-19 trends (as well as the tendency of some authorities to introduce schemes in September following the annual August fall in congestion) have contributed to perceptions that such schemes make large contributions to boundary road traffic.And it may still be the case that despite relatively little changes in boundary road traffic, those car journeys which continue to be made as before take longer simply because cut-throughs are no longer available.This study only looked at motor traffic volumes, not journey times or trip distances (e.g. through a travel survey); different methods are needed to test and quantify changes in journey time or trip distances.For instance, Goodman et al. (2023) used Driver and Vehicle Licencing Agency data, which found a 6.4% decrease in distance driven per vehicle among those living in areas in Lambeth, South London, where LTNs were introduced.
Finally, it is important that boundary roads are not forgotten.They do experience often substantial traffic burdens, and just over half the boundary roads in this study saw increases over the monitoring periods (with just under half seeing a reduction).Tools are needed to reduce burdens on boundary roads, whether by traffic reduction and/or mitigation of its effects.A paper by Hajmohammadi and Heydecker (2002) found that the introduction of London's Ultra Low Emission Zone in April 2019 led to reductions in NO, NO 2 , and NO x concentrations both within the implementation zone and the wider low emission zone (LEZ) and Greater London area.LTNs represent one tool alongside many others (such as bus priority, pollution taxes, road charging, and/or car parking levies, which address different aspects of the problem and/or address different behavioural patterns) that may be needed to achieve the ambitious reductions in car use sought by many cities including London.Further research could explore and compare the impacts and different contributions of these tools alone or in combination.

Fig. 2 .
Fig. 2. Map of LTN areas in part of Hackney (yellow), with boundary roads in purple.

Fig. 4 .
Fig. 4. Chart indicating the background traffic trend in each London functional zone from TfL data.

Fig. 5
highlights the different patterns on internal and boundary roads, shown as the difference from the expected change based on functional areas.Three-quarters of monitored internal roads saw

Table 1
Road classification by whether internal or boundary road.
functional zone compared to March 2019.Thus, for instance, in March 2021 the figure for Inner London was − 19%, meaning that average daily motor traffic was 19% less in March 2021 than in March 2019.We initially turned these figures into percentages (and associated ratios) of March 2019.Where average daily motor traffic is 19% less than in March 2019, this is 81% of the March 2019 figure, or a ratio of 0.81 to 1.These percentages were used as 'adjustment factors' as explained below.

Table 2
Count sites.

Table 3
Mean and median internal and boundary road traffic changes.

Table 4
Comparison of changes on major and minor boundary roads.

Table 5
Changes on internal and boundary roads by functional zone of London.The mean change of 0.7% compared to expected background change is comparable in magnitude to similarly calculated changes in Inner London's non-holiday seasonal fluctuation: for instance, such fluctuation meant average daily motor traffic in March 2019 was 1.4% higher than in May 2019.schemes,motor traffic fell on average 45.8% more than predicted; this is − 48.3% in Inner London, but somewhat less in Central and Outer London (− 39.2% and − 36.5%). 7

Table A6 Table of Adjustment
Factors.