39 results found
Singh R, Graham D, Horcher D, et al., 2021, The boundary between random and non-random passenger arrivals: robust empirical evidence and economic implications, Transportation Research Part C: Emerging Technologies, Vol: 130, ISSN: 0968-090X
In this paper, we investigate the influence of train headways on passenger platform wait times using automated data from the London Underground metro system. For high frequency services, the literature suggests that passenger arrivals are random and that under perfectly random conditions with all other factors held constant, wait times are equivalent to half of the headway between trains. We test this hypothesis using large-scale smart card and vehicle location data, which enables the extraction of access times from total passenger journey times as well as the precise measurement of train headways. Using a semi para-metric regression modelling framework, we generate non-linear estimates of the relationship between access times and headway while conditioning for other service supply and demand factors. Marginal platform wait times are then derived numerically via an exposure-response model framework which accounts for potential confounding between the walking and waiting components of access times, thus enabling quantification of the unbiased impact of headways on wait times. For three lines in central London, we observe that marginal wait times transition from greater than half of the headway to approximately one third of the headway astrain frequencies decrease. The transition occurs in the range between 2-3 minute headways, lower than earlier estimates in the literature. A series of numerical simulations illustrate the importance of waiting time sensitivity in the optimisation of public transport services. In comparison with the standard wait time assumption, our exercise reveals that the degree of density economies is milder than what the literature suggests, and this may neutralise some of the economic justifications of high public transport subsidies
Xuto P, Anderson R, Graham D, et al., 2021, Optimal infrastructure reinvestment in urban rail systems: A dynamic supply optimisation approach, Transportation Research Part A: Policy and Practice, Vol: 147, Pages: 251-268, ISSN: 0191-2607
The state of infrastructure in many developed countries around the world is an increasingly pressing issue, with mounting costs the longer repairs are deferred. In today’s rapidly urbanising world, the urban rail network is particularly critical, since infrastructure failures can have severe economic consequences for both the operator’s finances and user time costs. This paper thus provides a system-level model of welfare-oriented supply optimisation that integrates asset management with the literature on optimal pricing and capacity provision. Using a simulation approach and calibrating with London Underground data, this paper delivers three key contributions. First, the economic efficiency of long-term capital planning is highlighted, with up to an 87% welfare gain when comparing a 40- versus 5-year planning horizon. Second, in general, the longer the planning horizon, the higher the annual welfare, demand, asset condition, fare and supply, in the steady-state. Third, the analysis explores why policies in reality diverge from the welfare optimum: we show that election cycles can have a detrimental effect, with increased asset neglect and volatility in spending. Economic efficiency improves in the short-term at the expense of the long-term; significant intervention is needed to break this downwards trend, as also reflected in various rail systems’ histories.
Singh R, Graham DJ, Anderson RJ, 2020, Quantifying the effects of passenger-level heterogeneity on transit journey times, Data-Centric Engineering, Vol: 1, Pages: e15-1-e15-28, ISSN: 2632-6736
In this paper we apply flexible data-driven analysis methods on large scale mass transit data to identify areas for improvement in the engineering and operation of urban rail systems. Specifically, we use data from automated fare collection (AFC) and automated vehicle location (AVL) systems to obtain a more precise characterisation of the drivers of journey time variance on the London Underground, and thus an improved understanding of delay. Total journey times are decomposed via a probabilistic assignment algorithm and semiparametric regression is undertaken to disentangle the effects of passenger-specific travel characteristics from network related factors. For total journey times, we find that network characteristics, primarily train speeds and headways, represent the majority of journey time variance. However, within the typically twice as onerous access and egress time components, passenger-level heterogeneity is more influential. On average, we find that intra-passenger heterogeneity represents 6% and 19% of variance in access and egress times, respectively, and that inter-passenger effects have a similar or greater degree of influence than static network characteristics. The analysis shows that while network-specific characteristics are the primary drivers journey time variance in absolute terms, a non-trivial proportion of passenger-perceived variance would be influenced by passenger-specific characteristics. The findings have potential applications related to improving the understanding of passenger movements within stations, for example, the analysis can be used to assess the relative way-finding complexity of stations, which can in turn guide transit operators in the targeting of potential interventions.
Anupriya A, Graham DJ, Horcher D, et al., 2020, Quantifying the ex-post causal impact of differential pricing on commuter trip scheduling in Hong Kong, Transportation Research Part A: Policy and Practice, Vol: 141, Pages: 16-34, ISSN: 0191-2607
This paper quantifies the causal impact of differential pricing on the trip-scheduling of regular commuters using the Mass Transit Railway (MTR) in Hong Kong. It does so by applying a difference-in-difference (DID) method to large scale smart card data before and after the introduction of the Early Bird Discount (EBD) pricing intervention. We find statistically significant but small effects of the EBD in the form of earlier departure times. Leveraging the granularity of the data, we also allow for the treatment effect to vary over observed travel characteristics. Our empirical results suggest that fares and crowding are the key determinants of commuter responsiveness to the EBD policy.
Anupriya, Graham DJ, Carbo JM, et al., 2020, Understanding the costs of urban rail transport operations, Transportation Research Part B: Methodological, Vol: 138, Pages: 292-316, ISSN: 0191-2615
There is considerable variation in the average cost of operations across urban rail transport (or metro) systems. Since metros are typically owned and operated by public authorities, there is a public interest case in understanding the key drivers of their operational costs. This paper estimates short-run cost functions for metro operations using a unique panel dataset from twenty-four metro systems around the world. We use a flexible translog specification and apply dynamic panel generalised method of moments (DPGMM) estimation to control for confounding from observed and unobserved characteristics of metro operations. Our empirical results show that metro systems with a high density of usage are the most cost-efficient. We also find that operational costs fall as metro size increases. These results have important implications for the economic appraisal of metro systems.
Singh R, Graham DJ, Horcher D, et al., 2020, Decomposing journey time variance on urban metro systems via semiparametric mixed methods, Transportation Research Part C: Emerging Technologies, Vol: 114, Pages: 140-163, ISSN: 0968-090X
The availability of automated data for urban metro systems allows operators to accurately measure journey time reliability. However, there remains limited understanding of the causes of journey time variance and how journey time performance can be improved. In this paper, we present a semiparametric regression modelling framework to determine the underlying drivers of journey time variance in urban metro systems, using the London Underground as a case study. We merge train location and passenger trip data to decompose total journey times into three constituent parts: access times as passengers enter the system, on-train times, and egress times as passengers exit at their destinations. For each journey time component, we estimate non-linear functional relationships which we then use to derive elasticity estimates of journey times with respect to service supply and demand factors, including operational and physical characteristics of metros as well as passenger demand and passenger-specific travel characteristics. We find that the static fixed physical characteristics of stations and routes have the greatest influence on journey time, followed by train speeds, and headways, for which the average elasticities of total journey time are −0.54 and 0.05, respectively. The results of our analysis could inform operators about where potential interventions should be targeted in order to improve journey time performance.
Morse L, Trompet M, Barron A, et al., 2020, A benchmarking framework for understanding bus performance in the U.S., Benchmarking: an international journal, Vol: 27, Pages: 1533-1550, ISSN: 1463-5771
Purpose This paper describes a benchmarking framework applied to medium-sized urban public bus agencies in the United States which has overcome the challenges of data quality, comparability and understanding.Design/methodology/approach The benchmarking methodology described in this paper is based on lessons learned through seven years of development of a fixed route key performance indicator (KPI) system for the American Bus Benchmarking Group (ABBG). Founded in 2011, the ABBG is a group of public medium-sized urban bus agencies that compare performance and share best practices with peers throughout the United States. The methodology is adapted from the process used within international benchmarking groups facilitated by Imperial College and consists of four main elements: peer selection, KPI system development, processes to achieve high-quality data, and processes to understand relative performance and change.Findings The four main elements of the ABBG benchmarking methodology consist of eighteen sub-elements, which when applied overcome three main benchmarking challenges; comparability, data quality, and understanding. While serving as examples for the methodology elements, the paper provides specific insights into service characteristics and performance among ABBG agencies.Research limitations/implications The benchmarking approach described in this paper requires time and commitment and thus is most suitably applied to a concise group of agencies. Practical implications This methodology provides transit agencies, authorities and benchmarking practitioners a framework for effective benchmarking. It will lead to high-quality comparable data and a strong understanding of the performance context to serve as a basis for organizational changes, whether for policy, planning, operations, stakeholder communication, or program development. Originality/value The methodology, while consistent with recommendations from literature, is unique in its scale, in-depth validation
Singh R, Graham DJ, Anderson RJ, 2019, Characterizing journey time performance on urban metro systems under varying operating conditions, Transportation Research Record, Vol: 2673, Pages: 516-528, ISSN: 0361-1981
Automated fare collection (AFC) data provide opportunities for improved measurement of public transport service quality from the passenger perspective. In this paper, AFC data from the London Underground are used to measure service quality through an analysis of journey time performance under regular and incident-affected operating conditions. The analysis involves two parts: (i) parametrically defining the shape of journey time distributions, and (ii) defining three performance metrics based on the moments of the distributions to measure the mean and variance of journey times. The metrics show that mean journey times are longest during the afternoon peak across all lines analyzed, and are more variable during the afternoon and off-peak periods depending on the line. Under incident conditions, mean journey times range from 8% to 39% longer compared with regular conditions, depending on the line. Overall, the main application of this work is that the metrics presented here can be directly applied by operators to quantify customer journey time performance, and can be further extended for industry-wide application to compare performance across metro networks.There has been increasing recognition in the transport industry of the need for performance metrics that capture journey time reliability from a passenger perspective as opposed to the traditional operator-oriented indicators. In a report for the Organisation for Economic Co-operation and Development (OECD) on service quality metrics used by metro operators, it is noted that the three most commonly reported metrics relating to journey time are train delay, wait times, and passenger journeys on-time (1). The first two metrics capture train performance from a schedule and headway adherence point of view. The third attempts to capture the experience of the user; however, it is recognized that operator-oriented indicators are rarely able to measure the true impact of passenger delay (2).The journey time distribution on
Canavan S, Barron A, Cohen J, et al., 2019, Best Practices in Operating High Frequency Metro Services, Transportation Research Record, ISSN: 0361-1981
© National Academy of Sciences: Transportation Research Board 2019. Most metro rail systems worldwide are facing increasing demand and the need to deliver additional capacity in key corridors. Although total capacity reflects the combination of train capacity and frequency, increasing frequency is the primary strategy to increase capacity on existing lines where infrastructure is fixed. Higher frequencies also increase efficiency, by attracting more passengers and making existing journeys faster, thereby making better use of expensive rail infrastructure and increasing both metro revenues and wider economics benefits to the cities they serve. This paper is based on a study conducted for the Community of Metros, a worldwide group of metro systems, which surveyed 17 high frequency lines. The paper first documents the characteristics of high frequency lines [with 25 trains per hour (tph) or more defined as “high frequency” and 30 tph or more as “very high frequency”] and presents the various constraints to higher frequency operations, including how they interact and the various possible solutions. Five main categories of constraints were identified, relating to signaling and train control, station and train crowding, fleet, terminal turnarounds, and service complexity. To achieve the highest frequencies, it is essential for metro systems to take a holistic approach and identify not only the immediate constraints but also secondary and tertiary constraints that may prevent the full benefits of improvements from being realized. This paper provides guidance to those operating, funding, planning, and designing metro systems in how to maximize frequency and thereby deliver greater benefits to riders, transit agencies, and stakeholders.
Anupriya A, Graham D, Anderson R, et al., 2018, Cost Function for Urban Rail Transport Systems, Transportation Research Board 98th Annual Meeting
Platform doors are increasingly installed by metros, primarily to improve safety. However, they have the potential for both positive and negative operational impacts, mostly by affecting dwell times at stations. Using data from the CoMET and Nova international metro benchmarking consortia of 33 metro systems, this paper seeks to understand and quantify these operational impacts. Overall, platform doors have a net negative impact on dwell times, leading to between 4 and 15 seconds of extra time per station stop. This is due to additional time for the larger doors to open and close slower passenger movements due to the additional distance between platforms and trains and, most importantly, extended departure delays after both sets of doors are closed caused by the need to ensure safety (that no one is trapped in the gap between the two sets of doors). This is a particular problem in mainland China, where metros conduct manual safety checks that require drivers to step out of trains onto platforms. However, despite longer dwell times, platform doors have a net positive impact on metro operations, largely due to the many safety benefits that also reduce delays and thereby improve service performance. There are also potential benefits regarding energy and ventilation. To mitigate the negative impacts, metros should seek to refine procedures and improve technology to reduce dwell time delays caused by platform doors. Reducing or eliminating these extra delays are essential to delivering efficient service and maximum capacity, provided that safety can be assured.
Canavan S, Graham D, Anderson R, et al., 2018, Urban metro rail demand: evidence from dynamic generalised method of moments (GMM) estimates using panel data, Transportation Research Record, Vol: 2672, Pages: 288-296, ISSN: 0361-1981
This paper estimates elasticities of demand for metro service with respect to fares, income, quality of service, population and network length. Data for 32 world metro systems covering the period from 1996 to 2015 are analysed within a dynamic panel data specification. Three key contributions are made. First, we collate a database for estimation that is more extensive than that used in previous studies. Second, the quality of the data we have available allows us to more accurately represent quality of service than has been possible previously. And lastly, we estimate and compare two different measures of demand. Our analysis finds a statistically significant negative fare elasticity of -0.25 in the long run for a passenger km specified model and -0.4 in the long run for a passenger journeys specified model, and a positive long run income elasticity of 0.17 and 0.18 for the passenger km and passenger journey models respectively. Regarding quality of service we find positive long run elasticities of 0.56 and 0.47 for the passenger km and passenger journey models respectively. Income levels, population and the size of the network are also found to be statistically significant and positive in nature. The results suggest passenger km and passenger journeys will increase more in response to changes in service (here represented by increased capacity) than to changes in fares, with the difference in elasticities of service and fares being more pronounced for passenger km.
Trompet M, Anderson RJ, Graham DJ, 2018, Improved understanding of the relative quality of bus public transit using a balanced approach to performance data normalization, Transportation Research Part A: Policy and Practice, Vol: 114, Pages: 13-23, ISSN: 0965-8564
In order for bus operators and/or their respective authorities to understand where service quality can improve, it is useful to systematically compare performance with organizations displaying similarities in types of services offered, operational characteristics and density of the service area. These similar characteristics enable peer organizations to benchmark performance once their operational data are normalized for differences in scale of operations. The most commonly used normalization factors for the demand side output are passenger boardings and passenger kilometres. For the supply side output these are vehicle kilometres and vehicle hours. Through twelve years of experience in the International Bus Benchmarking Group (IBBG) a better understanding of differences in service characteristics between ‘similar’ peers has been achieved, which highlight a challenge for the interpretation of normalized performance. It became clear that relative performance should often not be concluded from performance indicators normalized in a single dimension. Variety between peers in commercial speed, trip length, vehicle planning capacity, vehicle weight and network efficiency result in the need for a bi-dimensional or balanced approach to data normalization. This paper quantifies the variety within these operational characteristics and provides examples of the interpretation bias this may lead to. A framework is provided for use by bus organization management, policymakers and benchmarking practitioners that suggests applicable combinations of denominators for a balanced normalization process, leading to improved understanding of relative performance.
Anupriya A, Graham D, Horcher D, et al., 2018, The impact of early bird scheme on commuter trip scheduling in Hong Kong: a causal analysis using travel card data, Transportation Research Board 97th Annual MeetingTransportation Research Board
Horcher D, Graham DJ, Anderson RJ, 2017, The economics of seat provision in public transport, Transportation Research Part E: Logistics and Transportation Review, Vol: 109, Pages: 277-292, ISSN: 1366-5545
Seated and standing travelling imply significantly different experience for public transport users. This paper investigates with analytical modelling and numerical simulations how the optimal seat supply depends on demand and supply characteristics. The paper explores the implications of seat provision on the marginal cost of travelling as well. In crowded conditions, we distinguish two types of external costs: crowding density and seat occupancy externalities. We derive, using a realistic smart card dataset, the externality pattern of a metro line, and identify the distorting role of the occupancy externality that makes the welfare maximising fare disproportionate to the density of crowding.
Horcher D, Graham DJ, Anderson RJ, 2017, The economic inefficiency of travel passes under crowding externalities and endogenous capacity, Journal of Transport Economics and Policy, ISSN: 0022-5258
Horcher D, Graham DJ, Anderson RJ, 2017, Crowding cost estimation with large scale smart card and vehicle location data, Transportation Research Part B: Methodological: an international journal, Vol: 95, Pages: 105-125, ISSN: 0191-2615
Crowding discomfort is an external cost of public transport trips imposed on fellow passengers that has to be measured in order to derive optimal supply-side decisions. This paper presents a comprehensive method to estimate the user cost of crowding in terms of the equivalent travel time loss, in a revealed preference route choice framework. Using automated demand and train location data we control for fluctuations in crowding conditions on the entire length of a metro journey, including variations in the density of standing passengers and the probability of finding a seat. The estimated standing penalty is 26.5% of the uncrowded value of in-vehicle travel time. An additional passenger per square metre on average adds 11.9% to the travel time multiplier. These results are in line with earlier revealed preference values, and suggest that stated choice methods may overestimate the user cost of crowding. As a side-product, and an important input of the route choice analysis, we derive a novel passenger-to-train assignment method to recover the daily crowding and standing probability pattern in the metro network.
Canavan S, Graham D, Melo P, et al., 2016, The Impacts of Moving Block Signalling on Technical Efficiency: An Application of Propensity Score Matching on Urban Metro Rail Systems, Transportation Research Record, Vol: 2534, Pages: 68-74, ISSN: 0361-1981
This study tested the effect of introducing moving-block signaling on the technical efficiency of urban metro rail systems. The study used a panel data set of 27 urban metro systems across 20 countries for 2004 to 2012. When moving-block signaling was considered as a treatment, the effect of the associated benefits on output efficiency levels was able to be measured. Stochastic frontier analysis was employed to estimate technical efficiencies for each metro, and then propensity score matching was applied to evaluate the effect of the type of signaling on technical efficiency. The study allowed the selection of appropriate reference groups and accounted for confounding factors. The study is novel in its provision of empirical evidence of this nature. The results indicate that the technical efficiency of a metro can be improved by 11.5%.
Cohen JM, Barron AS, Anderson RJ, et al., 2016, Impacts of Unattended Train Operations (UTO) on Productivity and Efficiency in Metropolitan Railways, Transportation Research Record-Series, Vol: 2534, Pages: 75-83, ISSN: 0361-1981
Urban metro subway systems (metros) around the world are choosing increasing levels of automation for new and existing lines: the global length of metro lines capable of unattended train operation (UTO) is predicted to triple in the next 10 years. Despite significant investment in this technology, empirical evidence for the financial and service quality impacts of UTO in metros remains scarce. This study used questionnaires and semistructured interviews with the Community of Metros and Nova Group benchmarking groups to assemble emerging evidence of how automation affected costs, staffing, service capacity, and reliability. The results from an analysis of data from 23 lines suggested that UTO could reduce staff numbers by 30% to 70%, with the amount of wage cost reduction depending on whether staff on UTO lines were paid more. On the basis of the experience of seven metros, the capital costs of lines capable of UTO were higher, but the internal rate of return had been estimated by two metros at 10% to 15%. Automated lines were capable of operating at the highest service frequencies of up to 42 trains per hour, and the limited available data suggested that automated lines were more reliable. The findings indicated that UTO was a means to a more flexible and reliable operating model that could increase metro productivity and efficiency. The study identified important work needed to understand the impacts of UTO and identify where statistical analyses would add value once sufficiently large data sets became available.
Horcher D, Graham DJ, Anderson RJ, 2016, Merging smart card data and train movement data: How to assign trips to trains?, Merging smart card data and train movement data: How to assign trips to trains?
This report explains the assignment method applied to link trips compiled in smart card data to train movements recorded in the signalling system. Particular attention has been paid to (1) origin-destination pairs with multiple potential route options, (2) peak-hour trips delayed by di culties in boarding crowded trains at the origin station, and (3) trips originating or ending on rail lines not included in the train movement dataset.In the current version of this paper the metro network on which the method has been applied is anonymised.
Horcher D, Graham DJ, Anderson R, 2015, The link between crowding pricing and seat supply in public transport, Transportation Research Board 95th Annual Meeting, Washington D.C.
Brage-Ardao R, graham DJ, anderson RJ, 2015, Determinants of Rolling Stock Maintenance Cost in Metros, Proceedings of the Institution of Mechanical Engineers Part F -Journal of Rail and Rapid Transit, Vol: 230, Pages: 1487-1495, ISSN: 0954-4097
This study examines the economies of scale and the determinants of rolling stock maintenance costs for 24 urban rail transit operators. The estimates reveal significant returns to scale in maintenance for both per car and per car kilometre. The econometric analysis also provides statistically significant cost elasticities for wages and staff hours, suggesting substitution effects between factors. Staff outsourcing is found to significantly decrease costs, whereas higher levels of fleet availability at the peak and rolling stock failures increase it. The effect of the age of rolling stock and the network is negligible on rolling stock maintenance costs; however, the analysis reveals a downward trend in rolling stock costs among the metros in the CoMET and Nova consortia.
Cohen JM, Parasram R, Anderson R, et al., 2015, Global trends in metro station organisation and management, 43rd European Transport Conference
Increased uptake of smart ticketing, mass availability of personal information technology,and roll-out of 4G and WiFi coverage within metropolitan railway systems, are leadingmetros to change the way they manage stations.
Cohen JM, Barron A, Anderson R, et al., 2015, Increased likelihood of injury as a form of transport disadvantage for differently abled and elderly travellers: Evidence from urban metro subway systems, 14th International Conference on Mobility and Transport for Elderly and Disabled Persons
Anderson RJ, Brage-Ardao R, Graham DJ, et al., 2015, Econometric Benchmarking of Metro Operating Costs. Methods and Applications, European Transport Conference 2015
Brage-Ardao R, graham DJ, Anderson RJ, 2015, Determinants of Train Service Costs in Metro Operations, Transportation Research Board 94th Annual Meeting
Cohen JM, Barron AS, Anderson RJ, 2014, Human Operational Support on UTO Lines, Publisher: Imperial College London
Metro automation is an increasing trend worldwide. This study investigated the realities of operating automated lines, focused on the following key questions: What staffing levels are used by metros, and what are the pros and consof each approach? Under what circumstances do metros choose to staff all trains on linesthat are capable of unattended operations? What technology is required to enable automated operations? Do the benefits of automation outweigh the additional investment?
Anderson R, Condry B, Findlay N, et al., 2013, Measuring and Valuing Convenience and Service Quality: A Review of Global Practices and Challenges from Mass Transit Operators and Railway Industries, International Transport Forum Discussion Papers, No. 2013/16, PCED Publishing, Paris, ISSN: 2223-439X
Barron A, Melo PC, Cohen JM, et al., 2013, Passenger-Focused Management Approach to Measurement of Train Delay Impacts, Transportation Research Record, Vol: 2351, Pages: 46-53, ISSN: 0361-1981
Train delay incidents have major effects on transit service reliability and on customer satisfaction. Operators have long focused efforts on preventing such incidents. While this action is important, the fact that all transit operations inevitably face some degree of delay and disruption from incidents means that operators must also dedicate attention to reducing the duration of incidents and the time to restore normal operations after incidents occur. To be able to do this, it is necessary to measure the total impact of incidents on train service and customers. This research uses data from the CoMET and Nova metro benchmarking groups to investigate the ways in which transit operators can better measure the full effects of incidents on train service and customers. The key benefit of such a passenger-focused approach is that it enables transit managers to direct resources for incident response and recovery better, as well as support the case for strategic investments. This research has shown that most operators measure and report only the frequency of incidents. Of the 22 metros interviewed, only two were able to provide detailed data to estimate the number of passengers affected by incidents. It is no coincidence that the only two metros able to provide detailed data are in fact two of the most reliable in the group.
Trompet M, Parasram R, Anderson RJ, 2013, Benchmarking Disaggregate Customer Satisfaction Scores Between Bus Operators In Different Cities and Countries
Directly comparing the satisfaction of customers of urban bus operators in different cities and countries is methodologically challenging due to the different surveys used, different sample frames, different response collection methods and the possibility of cultural bias. Nonetheless, due to the importance of customer satisfaction, the members of the International Bus Benchmarking Group started a research project in 2009 to overcome these challenges. The objective was for bus operators to understand the relative performance in meeting their customer’s expectations and to be able to target those areas in which they relatively underperform. Between 2009-2012, eight to ten participating organizations annually posted identical surveys on their website homepages in the same period. This paper describes the survey and data normalization methodology developed within the International Bus Benchmarking Group that provides managers of these organizations with a comparable view of their customer satisfaction. The described methodology has been successfully tested in the bus industry but can also be applied to other industries where there is a wish to benchmark customer satisfaction amongst other national and international peers.
This data is extracted from the Web of Science and reproduced under a licence from Thomson Reuters. You may not copy or re-distribute this data in whole or in part without the written consent of the Science business of Thomson Reuters.