Smart and Safe Kenya Transport (smarTTrans) Using technology, analytics and policy experiments to save lives and promote inclusive growth Context Road traffic crashes (RTCs) devastate the lives of victims and their families, while causing economic destruction in terms of property and lost earnings, especially for the poor. With 1.35 million deaths per year, RTCs are the primary cause of death among young people and the eighth leading cause of death globally (WHO 2018). They cause 50  million nonfatal injuries, imposing heavy burdens on health systems. The overall annual costs are estimated between 1% and 5% of GDP in developing countries (World Bank 2014). Without urgent attention to this issue, the Sustainable Development Goal (SDG) target 3.6 to halve RTC fatalities by 2020 will not be met (WHO 2018). Policy action must focus on Africa, where road traffic deaths are the highest (26.6 deaths per 100,000 people), and in contexts like Kenya, where these rates rank among the highest in the world. This is also where pedestrians make up 37% of RTC fatalities, compared to the 22% incidence worldwide (WHO 2018). In Nairobi, where the Smart and Safe Kenya Transport (smarTTrans) pilot takes place, walking is the main mode of commuting (Pendakur 2005), especially among the poor. Here, pedestrian deaths represent 71% of all RTC deaths reported in police crash records (2012–18). Poor data, no analytics, and the lack of policy experimentation constrain the government’s ability to develop policies and interventions to effectively regulate, monitor, and enforce road safety. In Kenya, the WHO estimates that the number of RTC fatalities is 4.5  times the figure in the official registry. The difference is due to lack of information on the location of crashes and their characteristics, on the dispatch of ambulances and the destination of the victims, on the results of emergency response at trauma centers, and on vital statistics about people who might die at home from complications. As a result, data analytics— Image by Nina Stock from Pixabay which would provide policy makers with a better understanding Image by Nina Stock from Pixabay of the problem—are lacking. Solutions are imported from more infrastructure and institutional approaches to improve road safety. developed contexts without adaptation to local conditions and To this end, smarTTrans aims to (a) build a pilot interactive data factors, and thus have limited impact. Policy experiments that platform for road safety and urban mobility, with particular attention could arise from better data and analytics are limited. Developing to advancing a measurement framework that brings together analytical tools is of paramount importance if Kenya is to create data on crashes and victims; (b) identify blackspots (delimited context-appropriate solutions to road safety. locations with the highest number of crashes or crash outcomes); (c) identify risk factors and characteristics of blackspots; and (d) evaluate interventions to improve road safety and urban mobility. smarTTrans and Its Expected An important innovation of this project stems from its systematic Policy Influence approach to building the data systems that can support the design The Kenyan government and the World Bank started the and implementation of road safety policies in data-resource- smarTTrans collaboration to develop data, analytics, and evidence constrained contexts and develop the tools that can be used to improve the country’s road safety and transport policy. This for both a national scale-up and replications in other contexts. joint effort is supported by the Kenyan ministries, departments, The project aims to directly inform government agencies such as and agencies in the transport sector, as well as by development the National Transport and Safety Authority and the National partners, and aims to test innovations using data, analytics, and Police Service on interventions that can have high returns and experiments to support more effective transport policies. The first how to target them in time and space to maximize effectiveness. focus of this collaboration, chosen in 2018 by these actors, is to The example of Kenya will be carefully documented to provide significantly improve road safety. a roadmap on building innovative data systems in data-scarce settings to monitor and manage road safety policy, and how to The strategy we adopted is to: (i) invest in innovative ways of use data to extract relevant information for more effective policy collecting high-frequency, real-time data, (ii) understand the local making. The tools and instruments developed in this pilot will be constraints to road safety, and (iii) experiment with behavioral, made available for deployment across the African continent by 2 SMART AND SAFE KENYA TRANSPORT (SMARTTRANS) leveraging the portfolio of transport projects of the World Bank Time and other development agencies. Addressing the urgent need Location Injury severity for data systems to facilitate appropriate planning and resource Date Direction of travel allocation is a necessary step to improve road safety. REF. TC/1/14/INDO/42/2017 27/6/2017 PD SUBJECT SELF INVOLVED SERIOUS INJURY ROAD TRAFFIC ACCIDENT PD THIS OCCURED TODAY 27/6/2017 ABOUT 01.00 AM AT DELTA ROUNDABOUT INVOLVING M/V REG. NO. KBW 838G V/W PASSAT DRIVEN BY JAMES WATIITI WAFULA AGED 44 YRS C/O TEL. 0790-810612 PD IT HAPPENED THAT smarTTrans Outputs: Road Safety THE SAID M/V WAS BEING DRIVEN FROM CHARY ROUNDABOUT ALONG HAILE SELASSIE AND ON REACHING AT THE LOCATION OF THE ACCIDENT, THE DRIVER LOST CONTROL OF THE M/V, AND HIT THE PAVEMENT AND LANDED INTO ROUNDABOUT PD DUE TO THE IMPACT THE M/V GOT FIRE PD AS A Nairobi Crash Map (Ongoing) RESULT THE DRIVER SUSTAINED FRACTURES ON BOTH LEGS AND WAS TAKEN TO K.N.H. WHERE HE WAS LEFT UNDERGOING TREATMENT PD SCENE VISITED AND M/V SHELL TAKEN TO THE STATION PD CAUSE CODE – 26 USED PD CASE P.U.I., D.T.O. INDO/AREA DEALING PD Using multiple data sources and machine learning algorithms, the BT smarTTrans team built the first georeferenced multiyear crash Hospital taken to DRAFTER – PC KANDIE Number of vehicles dataset and crash visualizations for the city of Nairobi. This con- and Injuries FOR: D.T.E.O. INDO/AREA Vehicle type stitutes the first step in building data and data systems leading to better road safety. The data allow for the identification of crash Cause code of crash location and time, helping identify the riskiest locations based on used by police actual crashes and crash outcomes. Having five or more years of Figure 1. Police crash report illustration data is critical to support any good analysis of road safety (US FHA Note: This example police report demonstrates the relevant data that was extracted 2017). The data include two types of sources: and coded as variables to produce a single dataset of all reports that can be analyzed to inform policy. All identifiable information has been removed and location information 1. Administrative: National Police Service paper-based crash has been changed to provide an illustration while maintaining privacy. reports between 2012 and 2018 for all 14 police stations in Nairobi (figure 1) 2. Crowdsourcing (contributions from a large online community): Ma3Route @Ma3Route · 15 Dec 2017 (i) Twitter handle Ma3Route (2012–19) (figure 2); (ii) Waze, a GPS 18:19 Major accident on Thika Road, just past Roysambu on your way to Githurai. Traffic jam from Garden City. Police on site navigation software app that provides navigation information Location but not of but also collects crowdsourced data on crashes (2018–19); and crash location (iii) Sendy, a motorcycle delivery phone platform (2018–19) Ma3Route @Ma3Route · 7 Nov 2014 About 10,000 police and 30,000 crowdsourcing reports have Accident at Uhuru Gardens man knocked down traffic snarl up been accessed. Paper reports from police, such as the example @CapitalFM_Kenya provided in figure 1, have been digitized, all relevant data coded Figure 2. Tweet crash report illustration into variables as demonstrated in the figure and the location Note: These tweets demonstrate the type of crowdsourcing reports accessed. The of the crashes geolocated to enable maximum use of the data machine learning algorithms developed by the team extract relevant information for informing policy. The smarTTrans team developed machine on location, as well as discard location variables that are irrelevant to enable unique geolocation of each report. learning algorithms using natural language processing to extract relevant information from crowdsourcing reports to enable their geolocation (figure 2). occur. The goal is a system that integrates multiple sources and Some reports correspond to the same crash but many are produces analytics in real time to facilitate early identification unique, allowing for a better picture of all crashes occurring of adverse events and risk factors, and to reduce reaction time, across the city and for learning about the types of biases that improving planning and response. the different sources may have. For instance, 98% of the police reports contain crashes with a death or injury, indicating that Using crowdsourcing is an inexpensive way to collect data at scale focus is on attending and recording the most severe crashes. on all crashes across a city to complement the administrative data. This also represents the best data to analyze crash outcomes, Administrative and additional data collection have also allowed but it is currently collected using paper forms, limiting its use for us to ground truth the crowdsourcing in order to develop better timely monitoring and action. Crowdsourcing allow for real-time algorithms. Through this process, we have worked to improve the reporting, which facilitates timely monitoring and action. However, administrative data, exploring best-practices for obtaining digital they also contain less standardized and verifiable details on crash information. These methods can be adapted for other contexts for outcomes, and decline late at night when many deadly crashes effective policy action. SMART AND SAFE KENYA TRANSPORT (SMARTTRANS) 3 Nairobi Crash Map National Police Records (2012–2018) C: 75 F: 30 I: 70 71% of road traffic deaths are pedestrians C: 52 Central Business District F: 27 I: 41 C: 69 F: 32 I: 67 C: 56 F: 21 I: 79 C: 63 F: 24 C: 69 I: 53 F: 32 I: 67 C: 26 F: 7 I: 61 C: 78 F: 25 I: 69 C: 47 F: 11 C: 28 I: 67 F: 3 I: 75 C: 59 C: 62 F: 8 F: 7 C: 63 I: 92 C: 63 I: 73 F: 10 F: 14 I: 54 I: 86 C: 80 F: 31 I: 69 Number of Fatalities 200 and Injuries locations (blackspots) 0 25 50 75 >100 represent 52% of crashes Number of Crashes and 55% of deaths 20 40 60 80 Blackspots are locations where a high number of crashes, C = crashes, F = fatalities, I = injuries deaths or injuries have occurred in a delimited space and time. 4 SMART AND SAFE KENYA TRANSPORT (SMARTTRANS) Blackspot Identification (Ongoing) these, 89% of crashes were geolocated using double data entry by local coders (the remaining reports lacked enough information Blackspots are crash locations or hazardous road locations that to locate the crash). The smarTTrans team developed a clustering usually indicate that there are common causes for the crashes. algorithm to identify blackspots. The preliminary analysis defines In developed countries such as Australia and New Zealand, safety a blackspot in two steps. First, we identified those locations with strategies aimed at risky locations have led to reductions in the more than 15 crashes within 300 meters during the analysis concentration of crashes; most fatal and serious crashes now period, identifying 150 blackspots. A second step considered occur at locations on roads with no other injury crashes reported additional crash outcomes. Locations that contained less than 15 in the previous five years (Austroads 2015). In countries where crashes but have either at least 21 injuries or at least 6 deaths were crashes are concentrated, which is likely the case for the deadliest also included as blackspots, since these crash outcomes may be countries in terms of RTCs, the focus of road-safety analysis and related to risk factors at those locations. This added 50 blackspots policy should therefore be on blackspots first. The first step is the for a total of 200. identification of such locations and classification of risk, which will depend on the context. Next, a diagnosis of the main risk factors and typology of blackspots, prioritization, and monitoring Risk Factors Associated with Blackspots and evaluation of the effectiveness of different interventions is (Ongoing) needed to maximize the benefits in scarce-resource contexts. A system of indicators was designed as part of this first For Nairobi, six complete years of police crash reports were used smarTTRans pilot based on internationally recognized standards. from 2013 to 2018 to help identify these blackspots. The data They were discussed with the Kenyan government agencies to includes 9,315 reports from the 14 National Police Stations across analyze risk factors at the 200 blackspots described above. The the city. These paper reports provide a description of where the indicators aim to measure multiple dimensions around the road crash occurred but no GPS to enable visualization on a map. Of segments that compose the blackspots, and use multiple data sources, including onsite data surveys and several months of real-time crowdsourcing through Uber and Waze. Examples of indicators included and their source are as follows: 77 Road physical attributes such as location of sidewalk (if any) relative to the road, quality of delineation, whether there are safety barriers along the road, or whether there are pedestrian crossings and whether they have clear markings on the road and signalization (survey following the International road Assessment Programme (iRAP) road coding indicators) 77 Road user flows, including cars, pedestrians, motorbikes, and trucks (surveys and observation) 77 Economic activity and land usage around the blackspots, such as schools, public transportation stops, and markets (web- scraping of landmarks from Google Maps) 77 Compliance with speed limits based on big data built with average speed in time intervals over each day for several months (Uber and Waze real-time web-scraping around the blackspots) 77 Compliance with select road safety practices, such as motorbike drivers’ use of helmets and the use of pedestrian crossings (five-minute videos recorded at the blackspots) This exercise will contribute to a structured set of indicators that can be used for crash blackspot assessments and road safety planning. They also aim to advance a framework to monitor and improve road safety in Kenya and similar contexts. 6 SMART AND SAFE KENYA TRANSPORT (SMARTTRANS) Pilots of Interventions to Improve 8% 8% Road Safety (Expected) 7% 7% The results from the outputs just noted are the basis for discussions 7% 6% with Kenyan government agencies and all stakeholders regarding 6% 6% 6% the types of interventions that should be tested in order to 5% 5% 5% 5% 5% 5% 5% address some of the high-risk problems identified in Nairobi. 5% 5% 5% 4% 4% Taking into consideration the constrained resources and in line 4% 4% 4% 4% with the Kenyan Road Safety Strategic Plan, this phase will 4% 4% 4% 4% 4% 3% 3% continue improving the data platform, developing a low-cost data 3% 3% 3% 3% 3% 3% 3% 3% platform (systems to assess deaths and injuries outside of the 2% 2% 2% crash site, which will improve the ability to assess the extent 2% 2% 2% 2% 1% and costs of road crashes), and incorporating interventions that address the risk factors identified. This work will be defined with the Kenyan government to identify feasible interventions such as 0:00 1:00 2:00 3:00 4:00 5:00 6:00 7:00 8:00 9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00 19:00 20:00 21:00 22:00 23:00 changes in road attributes (engineering interventions), road safety management, enforcement and behavioral interventions. Total deaths registered Total injuries registered Figure 3. Deaths and Injuries Reported in Police Crash Reports by Time smarTTrans Early Findings of Day, 2013–18 The first result from merging police and crowdsourced data is that there are at least 50% more crashes reported by bystanders than are reported in the police crash reports. Crowdsourcing crash data, 53% (deaths) and 50% (injuries) of the total. The deadliest times in other words, can help us identify more of the universe of crashes, for pedestrians are at night between 7 p.m. and midnight when gain a better understanding of their characteristics, develop policies 41% of pedestrian deaths occur. Implicating factors that need to be to address them as well as estimate the economic costs of RTCs. investigated further could include poor visibility, drunk driving and speed. Cross examining location and time of crashes could help Second, a statistical analysis of the 2013–2018 police reports develop a strategy for remedying street lighting and enforcement reveals the following: of speed and alcohol regulation. Finding 1: 71% of RTC deaths in the city of Nairobi covered by the main 14 Police stations are pedestrians. Pedestrian deaths Moving Forward reached 79% in 2014 and adjusted downward to 66% by 2017. The early results from the analysis of the data we collected and Finding 2: While crashes occur in 1400 different locations digitized show how critical it is to clearly identify the problems across the city, 200 locations represent 55% of the deaths and and prioritize policy interventions around them. This would 53% of the injuries across the city. These locations represent avoid dissipating scarce fiscal resources and instead use them 14% of all locations where severe crashes have occurred in the six strategically to narrow down where policy action can be most years under review. This high concentration of crashes and negative effective in reducing mortality. Targeting 14% of locations where crash outcomes is an opportunity to prioritize these locations for severe crashes have occurred and focusing enforcement efforts in policy action and substantially remedy the loss of life in the city. the most dangerous hours of the day could potentially halve the number of deaths in the city. This will not only allow Kenya to meet Finding 3: 35% of deaths occur within twenty meters of matatu its SDG targets but avoid enormous suffering and economic costs. stages. This staggering number is a call for action. Developing a better understanding of how to regulate and enforce matatu Moving forward, the team will conduct a more comprehensive flows, and driver and pedestrian behavior, while at the same time analysis of both police and crowdsourced data as well as a improving the infrastructure at these sites will be required to detailed analysis of the survey conducted on the priority guide policy action. blackspots. The latter will be used to understand what makes these locations so dangerous and what might be the menu Finding 4: Deaths and injuries are concentrated between of interventions and investments we might want to test and 5 a.m. and 8 a.m. and between 5 p.m. and 11 p.m., representing implement to curb crashes and deaths. SMART AND SAFE KENYA TRANSPORT (SMARTTRANS) 7 smarTTrans Team and Funding ” Berkeley, CA: UC Recommendations with an Evidence-Based Model. Berkeley. https://escholarship.org/uc/item/6dv195p7. The research team comprises of Guadalupe Bedoya, Arianna Pendakur, V. Setty. 2005. “Non-Motorized Transport in African Cities. Legovini, Sveta Milusheva, Sarah Williams, and Robert Marty. ” SSATP Working Paper 80. Lessons from Experience in Kenya and Tanzania. This work is in collaboration with the World Bank Transport Global Africa Transport Policy Program, World Bank, Washington, DC. https://www. Practice, the National Transport and Safety Authority, the National ssatp.org/en/publication/non-motorized-transport-african-cities-lessons- Police Service, and other Kenyan government agencies. Amy experience-kenya-and-tanzania. Dolinger, Meyhar Mohammed, and Robert Tenorio provided research US FHA (US Federal Highway Administration). 2017. Road Safety assistance throughout the project. Elizabeth Resor developed the Fundamentals: Concepts, Strategies, and Practices that Reduce Fatalities first geocoded crash map and provided field support. Funding was and Injuries on the Road. Washington, DC: US FHA. https://rspcb.safety. provided by the DIME Impact Evaluation to Development Impact fhwa.dot.gov/rsf/. ieConnect program, which has been funded with UK aid from the WHO (World Health Organization). 2010. “Data Systems: A Road Safety UK government, and the Transport Global Practice at the World Bank. ” Geneva: WHO. https://www. Manual for Decision-Makers and Practitioners. who.int/roadsafety/projects/manuals/data/en/. WHO. 2018. Global Status Report on Road Safety 2018. Geneva: WHO. References https://www.who.int/violence_injury_prevention/road_safety_status/2018/ Austroads. 2015. Guide to Road Safety, part 8, “Treatment of Crash en/. ” Sydney: Austroads. https://austroads.com.au/safety-and-design/ Locations. World Bank. 2014. Transport for Health: The Global Burden of Disease from road-safety/guide-to-road-safety. Motorized Road Transport. Washington, DC: World Bank. http://documents. Gonzales, Eric J., Celeste Chavis, Yuwei Li, and Carlos F. Daganzo. worldbank.org/curated/en/984261468327002120/Transport-for-health-the- 2009. “Multimodal Transport Modeling for Nairobi, Kenya: Insights and global-burden-of-disease-from-motorized-road-transport. ieConnect has over 30 ongoing impact evaluations across 19 different countries. The IEs focus on urban mobility, transport corridors, road safety, and rural roads sectors with thematic emphasis on gender, female economic empowerment, and fragile situations. From the ieConnect program we will learn how to improve the availability and quality of data that can be used for measuring the impact of transport projects and generate evidence that can be used to improve decision making for transport investments in the long-term. The ieConnect for Impact program is a collaboration between the World Bank’s DIME group and the Transport Global Practice. This program has been funded with UK aid from the UK government.