This is why building new roads can make the network less efficient – and make traffic congestion worse

Congestion in action. Image: Flickr/Wendell, CC BY-ND.

It’s not been a good year so far for major transport projects in Australia’s capital cities. 

The scale of disagreements over the merits of Melbourne’s proposed East West Link was such that the Victorian Government recently paid out $339m simply for the project not to be built. In Queensland, Annastacia Palaszczuk became the fourth successive Premier to completely throw out her predecessor’s signature infrastructure project – in this case the Bus and Train (BaT) Tunnel.

And while Sydney’s WestConnex project is still going ahead, recent reports concluded that it will either substantially increase traffic on the much-maligned Parramatta Road, or maybe decrease it. It all depends on which report you believe.

I am not going to debate the relative merits of these schemes: my background is in applied mathematics and not in transport planning. As such, my interest is less in the conclusions of some of these predicted usage studies and more in the consequences of the assumptions made in the modelling.

How can one report filed with the NSW Major Project register predict 20,000 fewer cars per day on a section of Parramatta Road, while other reports within the Roads and Maritime Services state that no significant reductions are likely to be seen?

It is very easy to chalk up some of these differences either to wildly overoptimistic developers potentially misleading themselves and others to get a project approved. Similarly, it is sometimes alleged that feasibility studies might be influenced by political biases or pre-established views on the merits of roads or public transport schemes.

While these factors may well influence some decision making, one thing that is often missed in the reporting of such studies is the sheer complexity associated with analysing such networks. Assuming all roads are connected, the behaviour of a whole network can be hyper-sensitive to how individual parts, even seemingly minor ones, function.

A poor estimate of traffic flow in one section of a network can lead to hugely different behaviour across the whole system. Furthermore, even the simplest networks can have the potential to function in some extremely surprising and often counter-intuitive ways.

It is very easy to believe that, if a traveller is offered the choice of two routes for a journey, the addition of a third choice should not worsen his/her travel time. If the new route is slower, the traveller could simply ignore the new route and make the same choice as before.

But as the German mathematician Dietrich Braess pointed out, this is not always the case. Increasing the capacity of a network can, perhaps surprisingly, decrease the efficiency of journeys around it even without increasing the number of trips made, as was pointed out in a recent article.

The Braess’s Paradox

To take a closer look at the reasoning behind this paradox, consider the case illustrated below. There are two major cities, labelled as the Start and End locations for a journey.

Travellers between the two cities have two choices of route, either via Town A or via Town B. The roads from Start to Town A and from Town B to End are both highways, which can handle any number of cars and allow them to make each leg of the journey in 105 minutes.

The roads from Start to Town B and from Town A to End are smaller roads, which are slower when busy. When there are N cars on the road, each leg of the journey takes N minutes. There is an old road linking Town A to Town B with a journey time of 100 minutes. This road is sufficiently slow that no traveller from Start to Finish would choose a route that involves it.

Network illustrating Braess’s Paradox. Travel times along routes are listed in minutes.

If we assume that 100 cars are travelling at the same time from Start to End, then there is no advantage to going via Town A vs going via Town B. The traffic will split approximately 50/50 between the two routes and each car will do the journey in 155 minutes. This is the fastest route. In reality, the split of cars might not be exactly 50/50, but unless the ratio is heavily imbalanced, the average travel time across the network will be 155 minutes.

A driver cannot help the overall network without suffering for it in the form of a slower journey

Now suppose now that the network is “improved” by upgrading the road between Town A and Town B. Rather than taking 100 minutes to travel between the towns, it now takes only 2 minutes.

The fastest route now is for all drivers to go from Start to Town B in 100 minutes, take the 2 minute trip to Town A, then travel from Town A to End in another 100 minutes. This journey now takes 202 minutes – but that’s 47 minutes longer than on the old road layout.

There is no incentive for any driver to choose an alternative route. Opting for either of the 105 minute roads will only lengthen their trip. A driver can improve travel times for all others by selflessly choosing the slowest roads, but cannot help the overall network without suffering for it in the form of a slower journey. This, of course, is not an option which many will choose.

While the old road between Town A and Town B was hugely inefficient, this inefficiency actually ensured that the network as a whole remained reasonably efficient. It served to distribute traffic evenly between the two routes from Start to End. But by improving this relatively unimportant road, it simply redistributes traffic more unevenly and worsens the overall system.

Even more counter-intuitively, Braess’s Paradox is observed in simple physical systems as well as transport networks.

The video (above) illustrates a system whereby a weight is suspended on two springs connected both in series (by an initially tense string) and in parallel (by initially slack strings). Removing the string which is in tension actually leads to the weight being lifted upwards.


This is actually a reverse of our traffic example. The distance the weight hangs represents the longer journey time of the traffic – remove the central string (the new road) and the hanging distance is reduced as is the journey time for the traffic.

Paradox in action

This paradox is not simply a mathematical quirk, or one which can be neglected by network analysts. There are a number of examples where removing roads – rather than building news ones – has improved transport networks.

Probably the most famous example of this comes from South Korea. When the motorway network around Seoul was reworked to remove some of the 1960s-built roadways, the the result was significantly reduced transit times throughout the city. This was not because of fewer journeys through the city, rather a more efficient distribution of cars across the remaining network.

Similar phenomena have been observed during road closures in New York City in the United States and in Stuttgart in Germany.

As Braess’s Paradox points out, even a slight change to a relatively unimportant part of the whole network can lead to massive changes in travel times. While planning reports might focus on headline stories – new road X will cut travel times by Y minutes – the underlying modelling must be more robust and look at the uncertainty around such estimates.

As painstaking as this modelling may be, it is unquestionably something that needs to be answered as fully and as correctly as possible, admitting its own limitations.

A multi-billion dollar infrastructure project cannot afford to fail simply because someone didn’t do their sums correctly. The financial consequences of incorrect projections can be financially catastrophic.

Both Sydney’s Cross City Tunnel and Lane Cove Tunnel drove their initial operators into receivership. The developers of Brisbane’s Clem Jones Tunnel fared no better.

The issue is not just limited to Australia, of course. Of the 15.9m journeys expected to be taken between London and Paris during the Channel Tunnel’s first year of operation, a mere 18 per cent of those actually occurred.The Conversation

Stephen Woodcock is lecturer in mathematics at University of Technology, Sydney.

This article was originally published on The Conversation. Read the original article.

 
 
 
 

Urgently needed: Timely, more detailed standardized data on US evictions

Graffiti asking for rent forgiveness is seen on a wall on La Brea Ave amid the Covid-19 pandemic in Los Angeles, California. (Valerie Macon/AFP via Getty Images)

Last week the Eviction Lab, a team of eviction and housing policy researchers at Princeton University, released a new dashboard that provides timely, city-level US eviction data for use in monitoring eviction spikes and other trends as Covid restrictions ease. 

In 2018, Eviction Lab released the first national database of evictions in the US. The nationwide data are granular, going down to the level of a few city blocks in some places, but lagged by several years, so their use is more geared toward understanding the scope of the problem across the US, rather than making timely decisions to help city residents now. 

Eviction Lab’s new Eviction Tracking System, however, provides weekly updates on evictions by city and compares them to baseline data from past years. The researchers hope that the timeliness of this new data will allow for quicker action in the event that the US begins to see a wave of evictions once Covid eviction moratoriums are phased out.

But, due to a lack of standardization in eviction filings across the US, the Eviction Tracking System is currently available for only 11 cities, leaving many more places facing a high risk of eviction spikes out of the loop.

Each city included in the Eviction Tracking System shows rolling weekly and monthly eviction filing counts. A percent change is calculated by comparing current eviction filings to baseline eviction filings for a quick look at whether a city might be experiencing an uptick.

Timely US eviction data for a handful of cities is now available from the Eviction Lab. (Courtesy Eviction Lab)

The tracking system also provides a more detailed report on each city’s Covid eviction moratorium efforts and more granular geographic and demographic information on the city’s evictions.

Click to the above image to see a city-level eviction map, in this case for Pittsburgh. (Courtesy Eviction Lab)

As part of their Covid Resource, the Eviction Lab together with Columbia Law School professor Emily Benfer also compiled a scorecard for each US state that ranks Covid-related tenant protection measures. A total of 15 of the 50 US states plus Washington DC received a score of zero because those states provided little if any protections.

CityMetric talked with Peter Hepburn, an assistant professor at Rutgers who just finished a two-year postdoc at the Eviction Lab, and Jeff Reichman, principal at the data science research firm January Advisors, about the struggles involved in collecting and analysing eviction data across the US.

Perhaps the most notable hurdle both researchers addressed is that there’s no standardized reporting of evictions across jurisdictions. Most evictions are reported to county-level governments, however what “reporting” means differs among and even within each county. 

In Texas, evictions go through the Justice of the Peace Courts. In Virginia they’re processed by General District Courts. Judges in Milwaukee are sealing more eviction case documents that come through their courtroom. In Austin, Pittsburgh and Richmond, eviction addresses aren’t available online but ZIP codes are. In Denver you have to pay about $7 to access a single eviction filing. In Alabama*, it’s $10 per eviction filing. 

Once the filings are acquired, the next barrier is normalizing them. While some jurisdictions share reporting systems, many have different fields and formats. Some are digital, but many are images of text or handwritten documents that require optical character recognition programs and natural language processors in order to translate them into data. That, or the filings would have to be processed by hand. 

“There's not enough interns in the world to do that work,” says Hepburn.


Aggregating data from all of these sources and normalizing them requires knowledge of the nuances in each jurisdiction. “It would be nice if, for every region, we were looking for the exact same things,” says Reichman. “Instead, depending on the vendor that they use, and depending on how the data is made available, it's a puzzle for each one.”

In December of 2019, US Senators Michael Bennet of Colorado and Rob Portman of Ohio introduced a bill that would set up state and local grants aimed at reducing low-income evictions. Included in the bill is a measure to enhance data collection. Hepburn is hopeful that the bill could one day mean an easier job for those trying to analyse eviction data.

That said, Hepburn and Reichman caution against the public release of granular eviction data. 

“In a lot of cases, what this gets used for is for tenant screening services,” says Hepburn. “There are companies that go and collect these data and make them available to landlords to try to check and see if their potential tenants have been previously evicted, or even just filed against for eviction, without any sort of judgement.”

According to research by Eviction Lab principal Matthew Desmond and Tracey Shollenberger, who is now vice president of science at Harvard’s Center for Policing Equity, residents who have been evicted or even just filed against for eviction often have a much harder time finding equal-quality housing in the future. That coupled with evidence that evictions affect minority populations at disproportionate rates can lead to widening racial and economic gaps in neighborhoods.

While opening up raw data on evictions to the public would not be the best option, making timely, granular data available to researchers and government officials can improve the system’s ability to respond to potential eviction crises.

Data on current and historical evictions can help city officials spot trends in who is getting evicted and who is doing the evicting. It can help inform new housing policy and reform old housing policies that may put more vulnerable citizens at undue risk.

Hepburn says that the Eviction Lab is currently working, in part with the ACLU, on research that shows the extent to which Black renters are disproportionately affected by the eviction crisis.

More broadly, says Hepburn, better data can help provide some oversight for a system which is largely unregulated.

“It's the Wild West, right? There's no right to representation. Defendants have no right to counsel. They're on their own here,” says Hepburn. “I mean, this is people losing their homes, and they're being processed in bulk very quickly by the system that has very little oversight, and that we know very little about.”

A 2018 report by the Philadelphia Mayor’s Taskforce on Eviction Prevention and Response found that of Philadelphia’s 22,500 eviction cases in 2016, tenants had legal representation in only 9% of them.

Included in Hepburn’s eviction data wishlist is an additional ask, something that is rarely included in any of the filings that the Eviction Lab and January Advisors have been poring over for years. He wants to know the relationship between money owed and monthly rent.

“At the individual level, if you were found to owe $1,500, was that on an apartment that's $1,500 a month? Or was it an apartment that's $500 a month? Because that makes a big difference in the story you're telling about the nature of the crisis, right? If you're letting somebody get three months behind that's different than evicting them immediately once they fall behind,” Hepburn says.

Now that the Eviction Tracking System has been out for a week, Hepburn says one of the next steps is to start reaching out to state and local governments to see if they can garner interest in the project. While he’s not ready to name any names just yet, he says that they’re already involved in talks with some interested parties.

*Correction: This story initially misidentified a jurisdiction that charges $10 to access an eviction filing. It is the state of Alabama, not the city of Atlanta. Also, at the time of publication, Peter Hepburn was an assistant professor at Rutgers, not an associate professor.

Alexandra Kanik is a data reporter at CityMetric.