Public Transport Congestion Relief Measurement A New Framework and Its Impacts (8)

Transcription

1 Australasian Transport Research Forum 2015 Proceedings 30 September - 2 October 2015, Sydney, Australia Publication website: Public Transport Congestion Relief Measurement A New Framework and Its Impacts (8) Nguyen Phuoc Quy Duy 1, Graham Currie 1, Bill Young 1 1 Institute of Transport Studies, Monash University, Clayton, VIC, for correspondence: graham.currie@monash.edu Abstract This study presents a new method that can be used to estimate the traffic congestion relief associated with urban public transport (PT). In order to calculate the impact of PT, it is assumed that a proportion of PT riders shift to car if PT service were to cease. As a result, the level of congestion on the highway network will increase because of the increase in the number of car trips. In this research, variation in the share of PT users switching to car is based on the traffic characteristics in each of Melbourne s Local Government Areas (LGAs) explored. Four predictor factors affecting this parameter are presented in this paper. These are the share PT users who: (1) Park and Ride, (2) make long-distance PT trips, (3) have car available and (4) have a driving license. The major finding is that the share of mode shift to car when PT is removed is lower in inner areas and higher for regions further from the CBD. Further, by using the Victoria Integrated Transport Model (VITM), the level of congestion relief in Melbourne is compared for the base and without PT scenarios. The results show that when all modes of PT are removed, the diversion to private cars generates an additional 1,500 congested road links, an increase of over 40%. The results of the new methodology are compared to previous research which used a fixed share of mode shift (32.4%) for all trip ends. Congestion relief patterns differ significantly using the new method in inner and outer areas. Mapping using ArcGIS illustrates these findings. The new approach offers a more precise model for assessing the impact of PT on traffic congestion. The paper closes with suggestions for further methodology development. 1. Introduction Congestion is a major issue in the daily lives of commuters, especially those living in big cities. As vehicles on the network in metropolitan areas grow, congestion has an increasing direct effect on commuters. Congestion cost increase through higher fuel consumption, delay, accidents, or air pollution. It is expected that congestion cost in Australia s cities could be over $20 billion AUD in 2020 (Garnaut 2012). In order to reduce the effect of traffic congestion, public transport offers a method of increasing person throughput. It has been encouraged and applied in many cities around the world. Transit service provides an alternative mode of travel, resulting in changes of trip making by automobile to transit, affects land-use activity and leads to direct and indirect employment. In addition, congestion relief impacts are also considered as one of the main rationales for providing transit service in cities (Gray 1992, Larwin 1999) There has been limited research on the congestion relief resultant from PT (Aftabuzzaman et al. 2010). Thus, transit management authorities have a difficult task in assessing the effectiveness of the existing PT system, particularly in terms of congestion relief. This paper outlines a methodology developed in Melbourne, Australia to estimate the traffic congestion relief impact of PT. A conventional four step transport model (The Victorian Integrated Transport Model VITM) is used to forecast congestion relief consequent on the new method. In addition, multiple performance measures that reflect many aspects of traffic 1

2 congestion are selected from existing methods to present the impact of PT on reducing traffic congestion. GIS is used to illustrate the research results. This paper is structured as follows: the next section outlines previous approaches to measure congestion relief. This is followed by a description of the study methodology. The results are then presented. The paper concludes with a summary and an outline of areas for further research. 2. Research Context Traffic congestion relief of transit was explored by (Lo and Hall 2006). In order to investigate the benefit of PT systems, they explored the impact of transit strikes that took place in Los Angeles over a 35-day period in October and November, Traffic conditions during the strike were measured to understand how transit actually affects congestion experienced by drivers. They measured the traffic speed on freeways before and after the strike by using various sensors. They found that there was a traffic speed decrease of 20% during the strike. The implication is that removing PT may act to reduce traffic speed by 20%. Parry and Small (2009) estimated the optimal transit operating subsidy by developing an analytical model of a transportation system. One input to the model is the impact of PT on reducing traffic congestion. In order to determine this effect they assumed that each passenger mile travelled on PT diverts nearly 0.9 passenger miles from roadways. The outcome from their model suggests that the PT system reduces the travel delay by 5%. In 2012, the annual urban mobility report from the Texas Transportation Institute explored the effect of PT on saving road travel time for 498 urban areas in America (Schrank et al. 2012). In this report, all commuter rail travellers are assumed to shift to private cars travelling on freeways if a PT service shutdown occurs. Thus, all the ridership from this mode is assigned to freeways. All travel from light rail, heavy rail and bus is assigned to both freeways and arterial streets in ratios that equal the existing travel portions on these types of roadways. With the additional transit vehicle-miles of travel, new operating speeds and volumes are computed and the delay is also estimated from these values. The report shows that if PT service is eliminated, the riders would contribute an additional 865 million hours of delay or approximately a 15% increase in the total delay in the 498 urban areas. Another study that measured the congestion relief benefit of public transport at a corridor level came from Washington, D.C (Federal Transit Adminstration 2000). Both of these studies are aggregate in nature (citywide and/or corridor level), and their fundamental assumption for measuring congestion relief benefit is that all public transport users switch to private vehicles when the PT service does not exist. These methodologies might be considered limited or simplistic because there are still many alternative travel modes that PT riders can be choose other than cars. In addition, there are a large percentage of PT users who do not have cars or driving licenses so only a percentage of PT users shifting to cars in the absence of PT service is a more likely outcome. Anderson (2013) explored whether transit generates a much larger congestion relief impact than earlier estimates. Using a choice model and data from Los Angeles, he predicts that PT travellers are likely to drive on routes with the most travel delay. A regression discontinuity design is then used to calculate the travel delay if PT is not available. He found that the average highway delay would increase 47% when PT ceases. Aftabuzzaman et al. (2010) demonstrated that in practice not all PT users would shift to using a car if PT is removed. Indeed, they assembled real world evidence that only a percentage of PT riders could switch to car. From secondary research, they suggested that on average 32% of PT users would shift to car use. They adopted this as a fixed value for all trips and applied it to a transport network model in Melbourne and estimated the congestion relief impact associated with PT. Some 32 percent of PT travel was added to the existing car trip matrix and this was then is assigned to the road network. They applied this on a fixed 2

3 basis for all trip ends in the model. Results found that removing PT is estimated to increase the number of congested links by about 1,400 or 30%. This research paper adds to the overall analytical process utilized by Aftabuzzaman et al. (2010) but adopts a new method for varying the share of PT users who shift to car. This method is based on the location and distribution travel behaviours related to car use amongst public transport users. 3. Analysis Objective In this paper, Aftabuzzaman s approach is adopted to estimate congestion relief however a variable rather than a fixed ratio of PT users who would shift to car when PT is removed is adopted. An approach is developed to estimate a variable rate of PT users who might shift to car driving. This is developed using an analysis of mode shift behaviour and travel characteristics of PT users in parts of Melbourne. The objective is to provide a more accurate assessment of congestion relief based on travel behaviour research. Figure 1: Mode choice options for PT riders when PT is removed Total of Public Transport Travellers Mode shift Trip cancelled Trip retiming Transportation characteristics for each location Trips switch to car Walking to destination Cycling to destination Car as drivers Car as passengers Figure 1 illustrates some of the processes of travel mode choice for PT users if PT is not available and factors which may impact on this mode shift. These are explored further below. The rest of the paper is structured as follows. In section 4, explanatory factors that might influence the share of mode shift to car if PT is removed are introduced. Section 5 presents new method to estimate variation in the share of PT travellers who would shift to cars. Section 6 then presents the results indicating the impact of new methodology on congestion relief modelling including comparison with older methods. A Discussion and Conclusions then close the paper. 4. Identifying Explanatory Factors 4.1 Aim The main aim of this section is to explore factors that might influence the mode shift from transit to private car if PT is not available. These factors regarding to transportation characteristics can be used to vary the share of mode shift to car for each area. 4.2 Spatial focus and data resources Spatial unit of analysis Local Government Areas are the base unit of analysis used this study. There are 31 LGA s in Melbourne VicRoads (2005) which are grouped into three categories (inner, middle 3

4 and outer) based on the classification of the local road authority for this study (see Figure 2). The local government areas in each category are: Inner: Melbourne, Port Phillip, Stonnington, Yarra Middle: Banyule, Bayside, Boroondara, Brimbank, Darebin, Glen Eira, Hobsons Bay, Kingston, Manningham, Maribyrnong, Monash, Moonee Valley, Moreland, Whitehorse. Outer: Cardinia, Casey, Frankston, Greater Dandenong, Hume, Knox, Maroondah, Melton, Mornington Peninsula, Nillumbik, Whittlesea, Wyndham, Yarra Ranges. Figure 2: Local Government Area in Melbourne Data sources In this research, both the 2011 Census (sourced from the Australian Bureau of Statistics) and the Victorian Integrated Survey of Travel and Activity 2007 (VISTA) (from the Department of Transport) are used. The census obtains data from the entire population (3,100,000 people) of Melbourne while VISTA explores a more detailed sample of over 17,000 households including travel information on every member of the household. Census data provides information on Park and Ride in each LGA. This source is utilized in this research because the sample size is much higher than that in VISTA. Indeed, in some outer LGAs the number of PT users included in VISTA in quite small. Research data is also obtained from VISTA that provides information regarding PT trips for all trip purpose while the census only provides work trip information. VISTA is a very comprehensive household travel survey that was conducted by Department of Transport, and it includes over 17,000 households, almost 44,000 people who made over 145,000 trips. So in most cases the sample sizes are very large, giving a small margin of error. The data has been weighted to match the demographics of Melbourne in the 2006 census. This provides a control to test under or over sampling of particular demographic groups. 4.3 Predictor factors affecting the ratio of PT user mode shift to private car There are several behavioural factors that can affect the proportion of PT travellers switching to car. In this research, four transportation characteristics are considered (1) Park and Ride use (PNR), (2) longer-distance PT trips (>3km), (3) PT riders with car available in the household and (4) PT riders with driving licenses. - The first factor that might influence on the percentage of mode shift is the share of PNR (V 1) access. PNR schemes often aim to reducing auto use to CBDs so PNR services are often subsidized to attract car users to use PT (Meek et al. 2008). Indeed, travellers who 4

5 use PNR services for access to PT might be considered an obvious group who might shift to car use if PT were removed since they have already used a car for this trip. - The share of long-distance PT riders (V 2) might be a critical factor affecting the share of mode shift to car if PT is removed. Trip length is considered as an important feature for choosing travel mode by road users (Bergström and Magnusson 2003, McConville et al. 2011). For short trips, auto trip can be replaced by three alternatives including public transport, walking and cycling (Carse et al. 2013). Panter et al. (2011) suggests, three kilometres is a maximum acceptable walking distance so travellers who have the trip length over than that value would tend to use cars for travelling. In this paper, the percentage of long distance PT trips (>3km) is calculated for each LGA. Hence, regions with the high values might have a high share of PT users switching to car if PT service is not available. In VITM, Melbourne is divided into 2,959 zones for analysis. A GIS approach is using to measure the distance between zones. It is assumed that PT users start and finish their trips at the centroid node of each zone (a point inside the zone). The number of long distance trips in each LGA is then calculated from the PT trip matrix obtained from VITM. - PT riders who have driving licenses and cars available may use them as alternative mode if PT is removed. In fact, road users tend to use private cars rather than walk and bike if they are licensed drivers and cars are available in their household (Ewing et al. 2004). Thus, these characteristics are likely to be related to the share of mode shift to car in the absence of PT service. Figure 3 shows the process for estimating the portion of PT riders having licenses and available cars (V 3). The number of adults (>18 year olds), vehicles in a PT traveller is provided in VISTA so it is possible to calculate V 3 value from that resource. Figure 3: Calculation of PT users with available car and license process PT user in a household Ye Have license N Number of vehicles - (number of adults - 1) I = 0 >0 N Ye I = 1 - The share of PT users owning driving licenses (V 4) is another factor that may impact on the share of mode shift. From a survey, Delbosc and Currie (2013) found that people getting driving licenses intend to have private car whilst those without a license do not plan to get one in the near future. Hence, if PT is removed, PT users owning a driving license have more chances to use a car. If a car is not available in their households, they could borrow or even rent. The VISTA 07 database is used to compute the percentage of PT users having a driver s license for each LGA. 5

6 4.4 Results Table 1 and Figure 4 show the values of four predictor factors. It can be seen that the share of PNR and long-distance PT trips is lower in inner areas such as the City of Melbourne, Port Phillip or Stonnington and higher in outer regions. By contrast, inner areas show higher shares of PT users owning driving licenses than outer and middle areas. Figure 4: Distribution of traffic characteristics for each LGA in Melbourne: (a) Share of PNR (%) (b) Share long-distance PT trip (%) (c) PT Users with available HH car (%) (d) PT Users with Driving License (%) Table 1: LGA share of: PNR, long-distance PT trip, available car and driving license. LGA_Name PNR Long-distance PT trip Available car Driving license V 1 (%) V 2 (%) V 3 (%) V 4 (%) Banyule (C) Bayside (C) Boroondara (C) Brimbank (C) Cardinia (S) Casey (C) Darebin (C) Frankston (C) Glen Eira (C) Greater Dandenong (C) Hobsons Bay (C) Hume (C) Kingston (C) Knox (C)

7 Manningham (C) Maribyrnong (C) Maroondah (C) Melbourne (C) Melton (S) Monash (C) Moonee Valley (C) Moreland (C) Mornington Peninsula (S) Nillumbik (S) Port Phillip (C) Stonnington (C) Whitehorse (C) Whittlesea (C) Wyndham (C) Yarra (C) Yarra Ranges (S) New Method To Predict The Share Of Mode Shift From Transit To Car 5.1 Aim The new method aims to estimate variation in the share of PT users who would shift to car driving if PT were removed. 5.2 Model structure A model is used to estimate mode shift using the four explanatory factors. A key input requirement is the relative strength of each of the four factors. In the absence of any primary research indicating relative strength, the following assumptions are made using simple deductions: The average mode shift share of LGAs is assumed to equal the fixed value (32.4%) suggested by Aftabuzzaman et al. (2010); this value was based on a range of secondary evidence and is considered a reasonable starting point for estimate mode shift. The share of mode shift (MS) for each LGA is the weighted mean of four mode shift shares (MS 1, MS 2, MS 3, MS 4) which are computed by splitting the effect of each factor using the following assumptions. The share of PNR (MS 1) might be assumed to have the highest contribution to mode shift share because PT users are already using a car for that trip. Only two modes (car and PT) are considered while they travel. Thus, if PT is removed, most of them might be expected to shift to car. The share of long-distance PT trips (>3km, MS 2) seems to be the second strongest contributing factor. In fact, private car is likely to be the only alternative mode for longdistance trips if PT is not available. Car available (MS 3) and driving licence (MS 4) might be surmised to have less contribution to mode shift since access to a car does not mean a PT user can drive and it is not clear a PT user with a license has a car available. Clearly the above arguments might be improved with primary research of users. But in the absence of this the following weights of four factors are adopted using the arguments above: β 1 : β 2 : β 3 : β 4 = 4: 3: 2: 1 The share of mode shift to car in LGA j by the effect of factor i is calculated by the equation: 7

8 MS i j = α i V i j (1) with α i = n 32.4% 31 j j=1 V i (2) MS j i : Share of mode shift to car for LGA j when splitting only one factor i impacts on the ratio of mode shift to car, α i : the weight of factor i, j: LGA that is considering, n: number of LGA (there are 31 LGAs in Melbourne) The value of α i is assumed as a constant for all LGAs so the mode shift share when splitting one factor is high if the share of that factor is high. The weighted mean method is then applied to compute the share of PT users switching to automobiles for each LGA. Thus the share of mode shift to private car for each LGA based on four impacting factors is calculated by following formula: MS j = β j j j j 1MS 1+β2 MS 2+β3 MS 3+β4 MS 4 (3) β 1 +β 2 +β 3 +β 4 MS j : the weighted mean of share of mode shift in LGA j β 1, β 2, β 3, β 4 : the weights of four factors. 5.3 Results Table 2 and Fig. 5 present the percentage of PT travellers who shift to individual cars if PT is removed from each of Melbourne LGAs. The Fig. 5 illustrates a considerable contrast in the ratio of mode shift to private vehicles between inner and outer Melbourne. These indicate the following: - The ratio of mode shift to car varies considerably by LGA On average 18.8% of PT users divert to cars in inner Melbourne. This figure is the lowest in the city of Melbourne by only 15%. - By contrast outer Melbourne has an average of approximately 40% of PT riders shifting to cars, about 10% higher than middle Melbourne (about 29.3%) and about 10% higher than inner Melbourne. Region with the highest share of mode shift to private car is Mornington Peninsula (47.6%). Table 2: Share of PT users shifting to car when split each factor and the average LGA_Name MS 1 MS 2 MS 3 MS 4 MS Banyule (C) Bayside (C) Boroondara (C) Brimbank (C) Cardinia (S) Casey (C) Darebin (C) Frankston (C) Glen Eira (C) Greater Dandenong (C) Hobsons Bay (C) Hume (C) Kingston (C) Knox (C) Manningham (C) Maribyrnong (C) Maroondah (C) Melbourne (C) Melton (S) Monash (C) Moonee Valley (C) Moreland (C) Mornington Peninsula (S)

9 Nillumbik (S) Port Phillip (C) Stonnington (C) Whitehorse (C) Whittlesea (C) Wyndham (C) Yarra (C) Yarra Ranges (S) Average Figure 5: Spatial distribution of share of mode shift to car for each LGA (%) 6. Impact Of New Approach On Congestion Relief Modelling 6.1 Aim The purpose of this section is to understand the impact of the new approach on congestion relief modelling in Melbourne. This includes comparison with the original approach of Aftabuzaman s which uses a fixed mode shift share to illustrate the impacts of the different methods. 6.2 Victoria Integrated Transport Model (VITM) VITM is used for the assessment of both current and future travel demand of Victoria. It can estimate the volume and travel time on each links of the network. In VITM, the road network that is considered as the input of this model is presented by a set of links and nodes, and the space is divided into 2959 zones. Each zone is represented by a centroid node that is a point inside the zone. Nodes usually represent an intersection or a change in road characteristics. Links usually represent the segments of actual roads in the network or they represent centroid connector. The links are given associated various road characteristics such as posted speed, capacity or travel time. In fact, VITM contains a number of submodels which work together to create the required output for each link such as actual speed, volume or travel time. 9

10 VITM was recalibrated by a team of transport modellers in The purpose of the recalibration was to update the VITM s key sub-models (trip generation, trip distribution and mode choice calculations) with new travel behaviour data obtained from the Victorian Integrated Survey of Travel and Activity (VISTA). VITM s structure was also redesigned to incorporate four time periods, a car ownership model and other numerous enhancements. VITM is built on a large database that includes information describing regional travel patterns, the regional transportation network, and regional socioeconomic/demographic patterns. 6.3 Modelling application The modelling analysis the congestion conditions is determined by using the trip matrix in the morning peak hour (7-9am). The modelling comprises three major steps: - Estimate the travel demand for each zone based on the demographic database. With the help of VITM, the travel data such as travel time, travel speed or volume of each link in traffic network is computed under the scenario with public transport. Multimodal Performance Measures are utilized to identify and assess the level of congestion of Melbourne roadway network. - The share of mode shift to car from transit for each LGA is adopted from the PT trip matrix. - These trips are added to a modified car trip matrix in the scenario without PT by adding the modified PT trip matrix (mode shift) to the existing car trip matrix. The result of this step is then assignment to the road network to explore congestion conditions after PT is removed. The comparison of the level of congestion between two scenarios with PT and without PT is undertaken to see the effect of the new approach on traffic congestion relief modelling. 6.4 Selecting multiple congestion performance measures The approach to congestion measurement starts by evaluating the level of congestion in the base case. Identifying congestion performance measures and threshold value of congestion is problematic since a wide range of methods and values are used in both the research literature and practice. In this research two approaches are chosen to estimate the congestion relief impacts of PT because of the availability of data: (a) volume-based approach and (b) a delay-based approach. The volume based approach measures congestion as the volume (V) of traffic on a road in relation to its capacity (C). Southeast Michigan Council of Governments introduces two threshold (V/C) ratios: 0.9 for severe congestion and 0.8 for moderate congestion (SEMCOG 2011). They are used in this research to classify a link as congested. The delay-based approach measures the delay of vehicles when the travel speed for a link is lower than the free-flow speed. The speed-flow relationship for this analysis which is embedded in VITM has been determined according to Akçelik (1991). In this research, multiple performance measures are applied at the regional level in order to compare the performance of transportation network in two scenarios: with public transport (base case) and without public transport. Thus, various aspects of congestion can be presented in many dimensions to explore outcomes. 6.5 Results Table 3 shows the congestion impacts for removal of PT on the whole Melbourne road network using both Aftabuzzaman's method and the new method described in this paper. Results suggest that: 10

11 - The number of moderate congested links is not significantly different between methods. However the severely congested links increase to 86.9% in the new method from 84% in the old. - The length of congested links increases in the new method to 50.2% from 47.8%. - Almost all congestion metrics rise slightly with the new method. In addition travel time per km is slightly slower. Overall the new method suggests slightly higher levels of congestion relief are associated with public transport than that suggested by the old method. The spatial contribution of congestion relief is also subtly different. Table 3: Comparing the congestion relief of PT in Melbourne: Aftabuzzaman's method and new method Aftabuzzaman's method (A) Difference A-B (%) New Approach (B) Congestion Measures Base Without Increase Without PT Increase (%) Case PT (%) Number of severe congested links (V/C>0.9) 2,075 3, , Number of moderate congested links (V/C>0.8) 1,999 2, , Length of congested Links (Km) 1, , , Congested link (%) Congested Lane (%) Number of vehicles experiencing congestion (millions) Vehicle distance travelled(million veh-km) Vehicle time travelled (million veh-hr) Total delay on roadway (million veh-hr) Average Travel Time Speed (Km/h) Actual travel time per km (min) Fig. 6a and Fig. 6b illustrate the number and the level of congested links on Melbourne network in the base case and the Without PT case (New Method). It is clear that the number of heavy congested road links increase significantly if PT is removed, particularly in inner Melbourne. Figure 6 Distribution of congested road links in Melbourne Congestion Level of Links No congested V/C<0.8 Moderate 0.8 V/C<0.9 Severe V/C 0.9 (a) Base Case (Including PT) 11

12 Congestion Level of Links No congested V/C<0.8 Moderate 0.8 V/C<0.9 Severe V/C 0.9 (b) No PT (New Method) When impacts are explored at an LGA level, there are considerable differences between the results of two methods. Fig. 7 illustrates road links in which the congestion levels change between the two methods. These figures show that: - There are congested road links in inner city in the Aftabuzzaman s method which are not congested in the new approach. - In contrast, when using the new approach to estimate the benefit of PT, more congested links occur in outer Melbourne if PT is removed. Table 4 compares the congestion relief effects of PT in inner, middle and outer Melbourne between the two approaches. This shows that: - In inner city areas, the new method estimates a lower level of congestion compared to previous method. The number heavy congested links only increases about 155% due to ceasing PT, nearly 9 % less than the result obtained from Aftabuzzaman s approach. Total delay estimates in inner areas were approximately 14% less with the new method. - By contrast, the congestion relief impacts of PT removal in outer areas are higher using the new method. The number of heavily congested links in the new method is higher by over 10% than estimated using Aftabuzzaman s approach. The total delay on roadway if PT ceases obtained by the improved method is about 10.4 million vehicle-hours whilst this figure evaluated by old method is only around 10.1 million vehicle-hours. - In terms of middle areas, there is an increase in the number of heavy congested links in the new method compared to old method (by 2.9%). Other congestion measures also show a slightly increase using the new approach compared to Aftabuzzaman s method. 12

13 Figure 7: Spatial distributions of road links which are only congested in each method in inner, middle and outer Melbourne Congested links Improved Method Abtabuzzaman Method Table 4: Comparing level of congestion in inner, middle and outer Melbourne between Aftabuzzaman's method and new method Aftabuzzaman's New Approach Difference method (A) (B) Region Congestion Measures A-B Base Without Increase Without PT Increase (%) Case PT (%) (%) Inner Middle Outer Number of severe congested links (V/C>0.9) 413 1, , Total delay on roadway (million veh-hr) Actual travel time per km (min) Number of vehicles experiencing congestion (millions) Number of severe congested links (V/C>0.9) 1,087 1, , Total delay on roadway (million veh-hr) Actual travel time per km (min) Number of vehicles experiencing congestion (millions) Number of severe congested links (V/C>0.9) Total delay on roadway (million veh-hr) Actual travel time per km (min) Number of vehicles experiencing congestion (millions) Discussion And Conclusions This paper has presented the result of research aimed at an objective estimation of the congestion relief associated with public transport system in Metropolitan Melbourne. It has also explored the previous studies regarding to the impact of PT on congestion reduction. A review of the literature shows that most recent researches determining the benefit of PT in reducing traffic congestion are based a fixed share of mode shift to car for all regions. These percentages of mode shift however vary among areas which have different traffic 13

14 characteristics. Four predictor factors that could have an effect on the share of PT users switching to car are manifested: (1) Park and Ride (PNR), (2) long-distance PT trips, (3) car available and (4) driving license available. Depending on the assumed weights of these traffic characteristic, the values of mode shift are calculated in each of 31 Melbourne LGAs. On average the outer Melbourne has the highest percentage of mode shift to private car by around 40% compared to middle Melbourne (29.3%) and inner Melbourne (18.8%). When applying the various share of mode shift to car, the analysis of congestion level shows that about 2,000 (increase of over 85%) additional network links become heavily congested because of the expand in the number of car trips when PT is removed. The total delay on the network also goes up by around 44%. Results from the new approach are different to those in Aftabuzzaman s method but not significantly different. However the spatial distribution of congestion relief varies significantly with the new method. In inner Melbourne, the average share of mode share obtained from new method is 18.8%, much less than the fixed value (32.4%) used in Aftabuzzaman s methodology. Hence, less PT riders are shifting to cars in congested inner areas and the congestion relief impact of PT is smaller in new method compared to previous one. By contrast, the share of mode shift in outer Melbourne areas calculated by new methodology is much higher than the fixed value so the PT benefits estimated in new method are higher than in Aftabuzzaman s approach. In middle Melbourne, the congestion relief impact of PT is slightly higher under the new method. Overall the revised approach proposed is likely to be more robust than the Aftabuzzaman approach however this does not mean to suggest that it could not be improved. The new approach adopted the fixed share of mode shift (32.4%) proposed by Aftabuzzaman as a starting point for estimates. This value is based on an average of a wide range of values from previous published research. Clearly a primary survey of local expectations of travel if public transport were removed might improve estimates. In addition the new method makes assumptions about the relative scale of impact of each of the explanatory factors for mode shift. This approach would also be improved by exploring the likely influences of park and ride, travel distance and car/license access on actual driving behaviours if public transport were removed. This research is part of wider project exploring how to evaluate the impacts of PT on traffic congestion. A range of opportunities are being explored to improve methodological approaches. There are many metrics used to identify congested links. This study used a V/C ratio threshold to determine the number of congested links. However, in some cases, this ratio cannot express fully the level of congestion in stop-and-go conditions. In this situation, traffic congestion can be really serious but the V/C ratio remains low. Thus, finding the most suitable congestion measurement approach to estimate PT impacts can be further explored. Another issue worthy of further exploration is the negative impacts which PT itself has on congestion at places like rail at-grade crossings or on roads where frequently stopping buses and trams might slow traffic. Removing PT in these cases should act to IMPROVE congestion. To understand the TOTAL effects of PT on traffic congestion it will be necessary to understand both the negative impacts as well as the positive impacts discussed in this paper. Overall the new methods described in this paper are considered an improvement to methodological approaches to assessing one of public transport most significant impacts on Australian cities; acting to reduce urban traffic congestion. It is clear these impacts are growing as Australian urban population rise. There is much scope to further improve methods in future research. 14

15 References Aftabuzzaman, M., G. Currie and M. Sarvi (2010). "Evaluating the Congestion Relief Impacts of Public Transport in Monetary Terms." Journal of Public Transportation 13(1): Akçelik, R. (1991). "Travel Time Functions for Transport Planning Purposes: Davidson's Function, its Time-Dependent Form and an Alternative Travel Time Function." Australian Road Research 21(3): Anderson, M. L. (2013). "Subways, strikes, and slowdowns: The impacts of public transit on traffic congestion." NBER Working paper series. Bergström, A. and R. Magnusson (2003). "Potential of transferring car trips to bicycle during winter." Transportation Research Part A 37: Carse, A., A. Goodman, R. L. Mackett, J. Panter and D. Ogilvie (2013). "The factors influencing car use in a cycle friendly city: the case of Cambridge." Journal of Transport Geography 28: Delbosc, A. and G. Currie (2013). Exploring Attitudes of Young Adults toward Cars and Driver Licensing. Australasian Transport Research Forum. Brisbane, Australia. Ewing, R., W. Schroeer and W. Greene (2004). "School Location and Student Travel: Analysis of Factors Affecting Mode Choice." Transportation Research Record: Journal of the Transportation Research Board 1895 (Transportation Planning and Analysis 2004): Federal Transit Adminstration (2000). Transit Benefits Working Papers: A Public Choice Policy Analysis. Federal Transit Administration, Washington, D.C. Garnaut, R. (2012). Transforming Transport. The Garnaut climate change review: Gray, G. E. (1992). Perceptions of public transportation. Public transportation. 2nd. Prentice Hall, Englewood Cliffs, New Jersey. Larwin, T. F. (1999). Urban transit. Transportation planning handbook. 2nd. Institute of Transportation Engineers, Washington, D.C. Lo, S.-C. and R. W. Hall (2006). "Effects of the Los Angeles transit strike on highway congestion." Transportation Research Part A 40: McConville, Megan, Rodríguez, D. A., Clifton, K. J, Cho, Gihyoug and F. Sheila (2011). "Disaggregate Land Uses and Walking." American Journal of Preventive Medicine 40(1): Meek, S., S. Ison and M. Enoch (2008). "Role of Bus Based Park and Ride in the UK: A Temporal and Evaluative Review." Transport Reviews: A Transnational Transdisciplinary Journal 28(6): Panter, J., S. Griffin, A. Jones, R. Mackett and D. Ogilvie (2011). "Correlates of time spent walking and cycling to and from work: baseline results from the commuting and health in Cambridge study." International Journal of Behavioral Nutrition and Physical Activity 8: 124. Parry, I. W. H. and K. A. Small (2009). "Should Urban Transit Subsidies Be Reduced?" American Economic Review 99 3: Schrank, D., B. Eisele and T. Lomax (2012). TTI s 2012 Urban mobility report, Texas A&M Transportation Institute. SEMCOG (2011). Congestion Management Process (CMP). Michigan Southeast Michigan Council of Governments. VicRoads (2005). Austroads Metropolitan Network Performance Database. Melbourne, Australia. VISTA-07 (2008). Victorian Integrated Survey of Travel and Activity Victorian Government, Melbourne, Australia. 15