Impact of a congestion pricing exemption on for new energy-efficient vehicles

The impact of demographics, built environment attributes, vehicle characteristics, and gasoline prices on household vehicle holdings
Dr.SagaLindberg Profile Pic
Dr.SagaLindberg,Philippines,Researcher
Published Date:18-12-2017
Your Website URL(Optional)
Comment
The impact of a congestion pricing exemption on the demand for new energy-efficient vehicles in Stockholm 1, 2 3 Jake Whitehead , Joel P. Franklin , Simon Washington 1 Junior Scholar and Double PhD Candidate, Royal Institute of Technology - KTH, Teknikringen 72, Stockholm, Sweden SE- 100 44 and Queensland University of Technology, 1 George Street, Brisbane, Australia 4000; Tel. (Sweden) +46 7 6252 1284; Tel. (Australia) +61 4 3040 4974; Email: jake.whiteheadabe.kth.se 2 Associate Professor, Royal Institute of Technology - KTH, Teknikringen 72, Stockholm, Sweden SE-100 44; Tel. (Sweden) +46 87 908 374; Email: joel.franklinabe.kth.se 3 Professor and TMR Chair, Queensland University of Technology, 1 George Street, Brisbane, Australia 4001; Tel. (Australia) +61 7 3138 9990; Email: simon.washingtonqut.edu.au ARTICLE INFO ABSTRACT Keywords: As governments seek to transition to more efficient vehicle fleets, one strategy has Energy-Efficient been to incentivize ‘green’ vehicle choice by exempting some of these vehicles from Vehicles; road user charges. As an example, to stimulate sales of Energy-Efficient Vehicles Alternatively (EEVs) in Sweden, some of these automobiles were exempted from Stockholm’s Fuelled Vehicles; congestion tax. In this paper the effect this policy had on the demand for new, Congestion Pricing; privately-owned, exempt EEVs is assessed by first estimating a model of vehicle Incentive Policies; choice and then by applying this model to simulate vehicle alternative market Multinomial Logit; shares under different policy scenarios. The database used to calibrate the model Revealed Preferences. includes owner-specific demographics merged with vehicle registry data for all new private vehicles registered in Stockholm County during 2008. Characteristics of individuals with a higher propensity to purchase an exempt EEV were identified. The most significant factors included intra-cordon residency (positive), distance from home to the CBD (negative), and commuting across the cordon (positive). By calculating vehicle shares from the vehicle choice model and then comparing these estimates to a simulated scenario where the congestion tax exemption was inactive, the exemption was estimated to have substantially increased the share of newly purchased, private, exempt EEVs in Stockholm by 1.8% (+/- 0.3%; 95% C.I.) to a total share of 18.8%. This amounts to an estimated 10.7% increase in private, exempt EEV purchases during 2008 i.e. 519 privately owned, exempt EEVs. 1. Introduction Numerous initiatives have been employed around the world in order to address rising greenhouse gas (GHG) emissions originating from the transport sector. These measures have included: travel demand management (congestion pricing), increased fuel taxes, alternative fuel subsidies and energy-efficient vehicle (EEV) rebates. Incentivizing the purchase of EEVs has been one of the more prevalent approaches in attempting to tackle this global issue. EEVs, whilst having the advantage of lower emissions and, in some cases, more efficient fuel consumption, also bring the downsides of increased purchase cost, reduced convenience of vehicle fuelling, and operational uncertainty. To stimulate demand in the face of these challenges, various incentive-based policies, such as toll exemptions, have been used by national and local governments to encourage the purchase of these types of vehicles. In order to address rising GHG emissions in Stockholm, and to achieve the Swedish Government’s ambition to operate a fossil-fuel free fleet by 2030, a number of policies were implemented, targeting the transport sector. Foremost amongst these was the combination of a congestion tax – initiated to discourage peak-hour emissions-intensive travel – and an exemption from this tax for some EEVs, established to encourage a transition towards a ‘green’ vehicle fleet. Although both policies shared the aim of reducing GHG emissions, the exemption for EEVs carried the risk of diminishing the effectiveness of the congestion-pricing scheme. As the number of vehicle owners choosing to transition to an eligible exempt EEV increased, the congestion-reduction effectiveness of the pricing scheme weakened. In fact, policy makers quickly recognized this potential issue and consequently phased out the EEV exemption less than 18 months after its introduction (Hultkrantz and Liu, 2012). Whitehead, J., Franklin, J., & Washington, S. 2   Several studies have investigated the demand for EEVs through stated-preference (SP) surveys across multiple countries, including: Denmark (Mabit and Fosgerau, 2011) Germany (Hackbarth and Madlener, 2013; Ziegler, 2012), Norway (Dagsvik et al., 2002), United Kingdom (Batley et al., 2004), Canada (Ewing and Sarigöllü, 1998), USA (Brownstone et al., 1996; Bunch et al., 1993; Hess et al., 2012; Musti and Kockelman, 2011) and Australia (Beck et al., 2013). Although each of these studies differed in their approach, all involved SP surveys where characteristics were varied among various types of vehicles including EEVs and presented to respondents, who in turn made hypothetical choices about which vehicle they would be most likely to purchase. As described in Section 2, although these studies have revealed a number of interesting findings regarding the potential demand for EEVs, they relied on SP data. In contrast, this paper employs an approach where EEV choice data are obtained retrospectively by collecting and using revealed preference (RP) data based on private vehicle registrations. By examining the revealed preferences of vehicle owners in Stockholm, this study overcomes one of the principal limitations of SP data - that stated preferences may not in fact reflect individuals’ actual choices, such as when cost, time, and inconvenience factors are hypothetical rather than real. While the RP data used in this study are sufficient, a follow up SP survey of vehicle owners in Stockholm could be interesting for comparing RP and SP results across a variety of dimensions. This paper’s RP approach involves modeling the characteristics of private individuals who purchased new EEVs, whilst estimating the effect of the congestion tax exemption on marginal demand. The study specifically builds on work undertaken by Bunch et al. (1993), Musti and Kockelman (2011), Campbell et al. (2012), Graham-Rowe et al. (2012) and Ziegler (2012) in Transportation Research Part A: Policy and Practice, in attempting to identify individuals that are most likely to purchase a energy-efficient vehicle. This paper also contributes to the current literature by examining the effectiveness of a tax exemption under revealed preference conditions, and by assessing the total effect of the policy based on key indicators for policy makers, including: vehicle owner home and work locations, commuting patterns, number of children, number of vehicles, age, gender and income. The two main research questions motivating this study were: • Which private individuals chose to purchase different types of new EEVs in Stockholm in 2008?; and, • How did the congestion tax exemption affect the marginal demand for new EEVs in Stockholm in 2008? In order to answer these research questions the analysis was split into two stages. Firstly, a multinomial logit (MNL) model was used to identify which demographic characteristics were most significantly related to the purchase of an EEV over a conventional vehicle. The three most significant variables were found to be: intra-cordon residency (positive); commuting across the cordon (positive); and distance of residence from the CBD (negative). In order to estimate the effect of the exemption policy on vehicle purchase choice, the model included variables to control for geographic differences in preferences, based on the location of the vehicle owners’ homes and workplaces in relation to the congestion tax cordon boundary. These variables included one indicator representing commutes across the cordon and another indicator representing intra-cordon residency. The effect of the tax exemption policy on the probability of purchasing EEVs was estimated in the second stage of the analysis by focusing on the groups of vehicle owners that were most likely to have been affected by the policy i.e. those commuting across the cordon boundary (in both directions). Given the inclusion of the indicator variable representing commuting across the cordon, it is assumed that the estimated coefficient of this variable captures the effect of the exemption policy on the utility of choosing to purchase an exempt EEV for these two groups of vehicle owners. The intra-cordon residency variable also controls for differences between the two groups, based upon direction of travel across the cordon boundary. A counter-hypothesis to this assumption is that the coefficient of the variable representing commuting across the cordon boundary instead only captures geo-demographic differences that lead to variations in EEV ownership across the different groups of vehicle owners in relation to the cordon boundary. In order to address this counter-hypothesis, an additional analysis was performed on data from a city with a similar geo-demographic pattern to Stockholm, Gothenburg - Sweden’s second largest city.    Whitehead, J., Franklin, J., & Washington, S. 3   Based upon this framework, the vehicle alternative market shares were calculated using the estimated coefficients of the MNL model and compared to predicted vehicle type shares from a simulated scenario where the exemption policy was inactive. This simulated scenario was constructed by setting the coefficient for the variable representing commutes across the cordon boundary to zero for all observations to remove the utility benefit of the exemption policy. Overall, the procedure of this second stage of the analysis led to results showing that the tax exemption had a substantial effect upon the probability of purchasing a new, exempt EEVs in Stockholm during 2008 i.e. the policy lead to an increase in marginal demand. As part of an additional analysis, panel data from both Gothenburg and Stockholm were used to compare the changes in vehicle shares between 2004 (before the tax exemption and congestion pricing) and 2008 in both cities, with Gothenburg acting as the control group (without a congestion pricing scheme and/or tax exemption). This additional analysis provides evidence to support the method used in this paper to estimate the increase in demand for exempt EEVs in Stockholm in 2008 due to the congestion tax exemption. The estimation results of two additional MNL models have also been included in the results section of this paper. The first of these models is a binomial model that has been included to compare with the principle MNL model employed for the policy analysis. The results of the third MNL model have been included in order to further explore what differences arise in regards to the demographic makeup of different vehicle owners, when the vehicle choice set is expanded based upon differences in vehicle purchase price, vehicle weight (as a proxy for size) and congestion tax exemption eligibility. Section 2 details the broader background of EEV choice, along with providing an overview of policies implemented to encourage the purchase of EEVs. In Section 3 a case study from Stockholm is presented, including a short overview of the history of EEV policies and summary of results from other studies that have investigated the effects of EEV incentive policies in Stockholm. Section 4 provides details of the research methodology, while section 5 documents the exploratory analysis of the dataset used in this study, and Section 6 discusses the results of the investigation, including the model estimation results and predicted shares of the vehicle alternatives. Finally, the study implications are discussed in Section 7, where the potential consequences of the findings are examined, particularly in relation to the effects of the congestion tax exemption on marginal demand for EEVs. 2. Background As energy independence and climate change have gained societal importance, many countries have sought to bring about a large-scale transition in the composition of national vehicle fleets. The policies introduced to encourage such transitions vary widely, including such measures as: subsidies for clean vehicle research and development; information campaigns to raise the importance of environmental concerns among households (Siriwardena et al., 2012); and financial incentives to make the choice of a clean vehicle more attractive (de Haan et al., 2009). There are several other incentive-based policies proposed by leading authors in this field, including Beck, Rose & Hensher’s (2013) paper investigating the effect of emissions charging on vehicle choice. The diversity of incentive-based policies makes it difficult to assess the demand for these vehicles, as the definitions of a ‘clean’ / ‘energy-efficient’ / ‘environmentally-friendly’ vehicle vary substantially between these policies. In some cases, different definitions have even been applied within the same country and/or region. Despite the complications of these varying definitions, it is clear from the literature that without incentives a substantial increase in the adoption of energy- efficient vehicles is unlikely. Most successful cities and/or countries have provided incentives that at least partially offset the typical disadvantages of adopting a EEV: lower driving ranges; smaller vehicle size; reduced engine power; limited fuel availability, etc. Other successful case studies have involved the introduction of regulations to counteract these disadvantages, e.g. mandatory supply of alternative fuels, electric charging stations at parking locations, etc. Literature investigating the marginal demand for different types of EEVs in different countries is increasing; however, as mentioned previously, the main approach in most of these studies has been to conduct a SP survey, in which a number of hypothetical scenarios were presented to the respondent. These scenarios involved a number of different vehicles; a number of different policies or incentives; or a combination of both various policies and vehicles. After collecting the data from    Whitehead, J., Franklin, J., & Washington, S. 4   the surveys, the information was then analyzed through the use of discrete choice models in order to identify which variables or indicators were the most significant within the survey sample. Both Ziegler (2012) and Hackbarth and Madlener (2013) found in their SP data that in Germany those individuals who were younger and had higher environmental preferences were the most likely group to purchase EEVs, with hydrogen vehicle owners most likely to be male. Mabit and Fosgerau (2011) in an analysis of the demand for alternatively fuelled vehicles (AFVs) in Denmark and Dagsvik et al. (2002) in a similar analysis for Norway, both found that if the cost and performance between AFVs and conventional vehicles were equal, due to environmental preferences, AFVs would be chosen. Contrary to the German studies, however, both studies found that females were more likely to purchase electric and/or hydrogen vehicles. Some studies have found that car ownership, in particular owning more than one vehicle, was a significant characteristic of EEV adopting individuals (Campbell et al., 2012; Graham-Rowe et al., 2012). Campbell et al. (2012) also found that individuals living further away from the city center were most likely to be early adopters of EEVs, in Birmingham, UK. Choo and Mokhtarian (2004) found in their study of vehicle choice in San Francisco, USA, that inner-city residents may have a tendency to own larger, less fuel-efficient vehicles, suggesting a reduced sensitivity to environmental concerns. These two studies are contrary to other evidence, particularly amongst new urbanist proponents, that inner-city residents tend to have higher environmental preferences and/or support environmental political parties (Bhat et al., 2009; Kahn, 2007), and in turn, individuals with higher environmental preferences tend to live more environmentally-friendly lifestyles, including purchasing smaller, more fuel-efficient vehicles (Kahn, 2007) and fewer vehicles (Flamm, 2009). In terms of incentive policies, Musti and Kockelman (2011) found that under both the hypothetical scenarios of a doubling in fuel prices or a rebate for EEVs, there would be little effect on the share of these vehicles in Texas, USA. Given the scenario of a ‘feebate’, however, where individuals would be compensated or charged in a carrot-and-stick approach depending on the fuel economy of the vehicle used, the share of EEVs would be increased by approximately 10%. Similarly, Gallagher and Muehlegger (2011) found in their analysis of state-level hybrid vehicle sales data across the USA that ‘feebate’ programs may be more effective in increasing the demand for EEVs compared to sales tax waivers and/or emissions testing fees. Another study, partially based on RP data from household surveys, found that monetary incentives had little to no effect on the adoption of EEVs, however, that an incentive such as a High- Occupancy Vehicle (HOV) lane exemption, when placed in congested areas of the USA, did lead to an increase in EEV shares (Riggieri, 2011). One could argue that the congestion pricing scheme in Stockholm acted somewhat similarly to a hybrid combination of the HOV lane exemption and the ‘feebate’ program, at least during 2008, where users had to pay a fee for using high-emission vehicles in the city, but gained an exemption from this fee if they had purchased and used an eligible energy-efficient vehicle. Considering the range of issues addressed by these prior studies, a literature gap exists on whether incentive-based policies, such as a congestion tax exemption, have affected the demand for EEVs. A better understanding of the characteristics of private individuals who have purchased these types of vehicles is needed in order for policy makers to better target such initiatives. This paper attempts to address this knowledge gap using Stockholm as the case study. 3. Case study – Stockholm, Sweden Since 1994, the City of Stockholm has had a EEV project in place, promoting the adoption and usage of these vehicles and their associated fuel types. From 1994 to 2005, two of the main achievements of this project were to replace conventional vehicles in the government fleet with EEVs and to put in place a number of tax incentives in order to increase the attractiveness, and in turn supply of alternative fuels within the Swedish market. From 2005, the demand for alternatively fuelled vehicles started to increase, largely due to a number of financial incentives that were introduced during the same period. In May of 2005, free residential parking was introduced for inner-city residents in Stockholm who owned alternatively fuelled vehicles, a policy that remained in place until the conclusion of 2008. The introduction of this policy was shortly followed by the commencement of a seven-month long congestion tax trial starting in January 2006, parallel to the introduction of a exemption from the congestion tax for all alternatively-fuelled EEVs e.g. vehicles running on ethanol, electricity, biogas, etc. After the trial, there was a 12-month period in which neither policy was active. During this    Whitehead, J., Franklin, J., & Washington, S. 5   period, a public referendum was held in order to gauge support for the policy with 52.5% of the population voting against the scheme (Börjesson et al., 2012). The 2006 general election also lead to a change in government from the center-left party to a center-right coalition. Despite this period of policy instability and the referendum result (which was not legally binding), the new government reintroduced the congestion tax permanently, starting from August, 2007, with the revenue raised to be hypothecated for the construction of new roads, in contrast with the environmentally driven motivations of the previous government. In 2006, during the congestion tax trial, only 2% of cordon boundary crossings were made by alternatively fuelled vehicles. By the end of 2008, this share had increased to 14% (Börjesson et al., 2012). The incentive policy of exempting alternatively fuelled EEVs from the congestion tax, the main subject of this paper, was so successful that policy makers became concerned that the congestion reduction effectiveness of the greater pricing scheme was being weakened. As such, the st tax exemption was phased out for all new EEVs purchased from the 1 of January, 2009, less than 18 months after its introduction. The policy did, however, remain valid for all existing EEVs that were already exempt until the beginning of August, 2012 (Birath and Pädam, 2010). Concurrent with the introduction of congestion tax exemption in Stockholm, in April, 2007, a 10,000 SEK (1,000 EUR) national purchase rebate was also introduced for all newly purchased alternatively fuelled and low CO petrol/diesel EEVs in Sweden. This last policy expanded the 2 definition of EEVs to also include petrol/diesel vehicles that emitted less than 120 grams of CO per 2 km (Börjesson et al., 2012; Pädam et al., 2009). It is this combination of policies that appears to have led to record growth in the sale of EEVs in Stockholm. The effect of the congestion tax exemption policy was the main focus of this study seen as the most significant EEV policy incentive introduced in Stockholm. This assertion has been echoed by several experts in the field (Börjesson et al., 2012; Hugosson and Algers, 2010) as well as established through a number of different studies, including: - Analysis of Swedish market level vehicle sales data combined with vehicle characteristics and fuel data using a Nested Logit model (Lindfors and Roxland, 2009); - Analysis of monthly reported new car registrations in Sweden using times series and cross sectional OLS regression (Pädam et al., 2009); and, - Results of an opinion survey sent to new clean vehicle owners in Stockholm in 2008 that was conducted by ‘Clean Vehicles in Stockholm’ (Birath and Pädam, 2010). Lindfors and Roxland’s (2009) paper primarily focuses on analyzing the effect the national purchase rebate had on the sales of alternatively fuelled vehicles throughout Sweden. They employ a similar method to that described in this paper where a variable representing the purchase rebate is included in a Nested Logit Model. Market shares are predicted based on this estimation and compared to shares predicted from the same model with the rebate coefficient set to zero i.e. removing the effect of the incentive policy. They estimate that the purchase rebase led to a 12% increase in alternatively fuelled vehicle sales throughout Sweden during 2008. Although brief, they also separate out data for Stockholm and include an additional variable representing the congestion tax exemption. Comparing this variable to the purchase rebate variable, they suggest that the exemption effect was at least twice as large as that of the purchase rebate i.e. a 24% increase. However, these figures refer to total sales in Stockholm, including company vehicles, which were subject to a number of other incentive policies active in Sweden during this period. As mentioned previously, only private vehicle owners are considered in this current study. In a separate analysis, Pädam et al. (2009) have analyzed monthly car registrations in Sweden, using both time series and cross-section OLS regressions, and found that the congestion tax exemption appears to have increased the sales of alternatively fuelled EEVs in Stockholm Country in 2008 by 23%. Again, this data included company and leased vehicles, and as such, we can expect the estimates from this paper’s analysis to be less than these figures. Finally, an opinion survey conducted by ‘Clean Vehicles in Stockholm’ during 2008 showed that EEV owners saw the congestion tax exemption and lower fuel costs as the most important incentives to purchasing an EEV in Stockholm (Birath and Pädam, 2010). The results of these studies have shown that the congestion tax exemption appears to have been the most significant incentive policy introduced in Stockholm in terms of increasing the demand for exempt EEVs. For this reason, this paper focuses primarily on annual vehicle registration data from 2008 - since this was the only period in which the congestion tax exemption    Whitehead, J., Franklin, J., & Washington, S. 6   was active for the entire year for all alternatively fuelled (exempt) EEVs. Importantly, the effects of other policies cannot be ignored thus this paper attempts to isolate the effect of the congestion tax. It should be noted that given the exemption was only implemented in mid-2007, there could be an argument that consumer behavior had not fully settled by 2008, however, given the policy was phased out by 2009, this was only full 12-month period in which it could be analyzed. 2008 was also the latest year of data available at the time of analysis. 4. Methodology Vehicle type choice can be conceptualized using econometric models of a discrete choice among mutually exclusive alternatives (Train, 2009; Washington et al., 2011). Here, it is assumed that when individuals choose a vehicle to purchase they maximize an unobserved utility function (unknown to the researcher). This function can be separated into an observable portion and an unobservable portion, written as: U =V +ε , where the unobservable portion (ε ) is i.i.d. from a nj nj ni ni Gumbel Type-2 distribution and captures all the factors that affect utility but that are not captured by observable factors (V ). In logit models, it is also assumed that the unobserved factors are nj uncorrelated over alternatives, which, although restrictive, provides a convenient form for calculating the choice probability (P ) – see Equation 1: ni V ni e P = ni V (1) nj e ∑ j Once the choice probabilities have been calculated, the model must then be estimated by using the maximum-likelihood function shown in Equation 2: N LL(β)= y lnP ∑∑ ni ni (2) n=1 i where y is an indicator for whether the decision maker chooses alternative i and β ni represents the parameter that maximizes this function. As shown by McFadden (1974), LL(β) is globally concave for linear-in-parameters utility. It can therefore be said that the β value that maximises this function is that where the derivative of the function is equal to zero – thus the global maximum. In this paper, this procedure was carried out using both Biogeme (BIerlaire, 2003) and STATA, testing a number of different model specifications and forms in order to find the best fit for new vehicle choice in Stockholm during 2008. In order to estimate the effect of the congestion tax exemption, it was necessary to operationalize which vehicle owners would be considered "treated" versus "untreated". This is not obvious: in some sense, a large portion of the total population might be considered "treated", since all who might sometimes travel by car, even as a passenger, across the cordon during peak periods might be affected by the presence of a congestion toll or a tax exemption. However, in this study the treatment group was designed to focus on private vehicle owners whose homes and workplaces were located on opposite sides of the cordon boundary. Hence, the vehicle owner population was split into four groups based on their home-work locations relative to the cordon, to separate out those owners that were more likely to have been affected by the policy. These four groups were: A. Living and working within the cordon; B. Living within but working outside the cordon (commute across the cordon); C. Living outside but working within the cordon (commute across the cordon); and, D. Living and working outside the cordon. Based on these groups, three variables were defined to control for geographic differences in preferences towards EEVs: • Living within the cordon (or not); • Commuting across the cordon (home-work trips); • Working within the cordon (or not). All three variables were not included within the model due to extreme multi-collinearity. The first two variables were included in the model specifications since the location of the workplace was    Whitehead, J., Franklin, J., & Washington, S. 7   seen to have had the least bearing upon vehicle choice. The second variable, representing commutes across the cordon, was seen as critical to this analysis as it was assumed to have a strong relationship with the effect of the congestion tax exemption policy. It is also possible that the estimated coefficient of this variable would instead capture other effects of geography on vehicle choice, such as vehicle owners who lived in the suburbs and worked in the city center having a set of attitudes and preferences that made them more likely to choose EEVs. To test this counter- hypothesis and provide evidence to support the validity of the method employed in this study, an additional analysis of panel data comparing new, exempt EEV shares in both Stockholm and in Sweden’s second largest city - Gothenburg (acting as a control group i.e. no congestion pricing scheme/tax exemption), has been included in Section 7.1. To assess the total effect of the exemption policy upon the demand for new, exempt EEVs in Stockholm, the predicted vehicle type shares were calculated based upon the estimated coefficients of the best model. By making the assumption that the estimated coefficient of the variable representing commuting across the cordon captured the effect of the exemption policy upon the utility of choosing to purchase an exempt EEV, this coefficient was then set to equal zero for all observations to simulate removing the benefit of the EEV exemption for crossing the cordon boundary. Predicted vehicle shares were then recalculated based upon this new scenario where the exemption was effectively inactive. By comparing the predicted shares from these two scenarios, an estimate of the effect that the congestion tax exemption had upon the demand for exempt EEVs could then be calculated. The estimation procedure was first carried out using Biogeme (BIerlaire, 2003), and then repeated in STATA using bootstrapping (1000 repetitions) in order to provide 95% confidence intervals for the reported results. Although factors other than the congestion tax exemption could have affected the decision of vehicle owners to commute across the cordon, by including a number of other variables to control for many of the demographic and geographic differences between vehicle owners, the calculated difference in vehicle shares could largely be attributed to the effect that the exemption policy had upon the demand for exempt EEVs. The findings of the additional panel data analysis (see Section 7.1) also assist providing evidence to support these conclusions. As stated earlier, the focus in this study is on the treatment effects on commute trips over the cordon. The resulting estimation is regarded as conservative since the demand for exempt EEVs by other vehicle owners, such as those who did not commute across the boundary, could have also been affected by the congestion tax exemption. This is likely to especially be true for those who lived and worked within the toll cordon. The data used in this analysis did not include detailed trip data, precluding estimation of the effect that the exemption policy would have had based on other trips. Regardless, the estimation provides some insight into the extent of the effect of the exemption policy, with the additional panel data analysis yielding estimated effects of the tax exemption on vehicle owners not commuting across the cordon boundary. It should also be noted that the free residential parking policy could have conflated the results obtained for inner-city residents. 5. Data and exploratory analysis Swedish vehicle ownership and distance travelled data, analyzed by Pydokke (2009), revealed that rural vehicle owners’ usage is higher compared to urban vehicle owners, and that car ownership is slow to change throughout Sweden. A subset of the same data analyzed in that paper, obtained from Sweden’s Central Bureau of Statistics (SCB), is used in this study and consists of vehicle registrations for the year 2008 combined with demographic characteristics for private vehicle owners in Sweden. The dataset used in this paper was created by first merging all vehicles with their respective owners and disregarding any entries that either had no vehicle or no owner. At this level the dataset included all owners in Sweden, so the study was further reduced to only those individuals who lived and worked in Stockholm County. Additionally, approximately 50% of the observations related to company-owned or -leased vehicles. Since it was impossible to determine whether the home locations were true to the vehicle owner, these entries were also discarded. Note that here, ‘new vehicles’ are defined as encompassing all vehicles with a manufactured date of 2007, 2008 or 2009, due to some 2007 and 2009 models being sold and registered during 2008. In the analysis it was assumed that the registered owner of the vehicle was also the predominant driver of that vehicle; the vehicle was used for home-work trips; and for the small group    Whitehead, J., Franklin, J., & Washington, S. 8   of owners with multiple vehicles, the most driven vehicle was the predominant vehicle for home- work trips. In calculating the predicted vehicle shares, the refined dataset was subdivided into groups based upon home and work locations. In particular focus were the groups commuting across the cordon in order to assess the impact of the congestion tax exemption upon the demand for new EEVs. A frequency table of the four groups, based on home-work locations, along with the annotation of incentive policies applicable to each of these groups, can be found in Table 1. Home and Work locations were based on postal code groupings, with Stockholm divided into approximately 50 areas. It should be highlighted that although the number of electric vehicles was relatively small, from an analysis of the summary statistics (see Table 2) it was clear that this group of owners was distinctly different from Ethanol EEVs, and thus the two groups were kept separate. It should also be noted that there were also a very small number of other alternatively-fuelled EEVs running on biogas, however, these observations were excluded from the data used in this analysis. TABLE 1 – Number of New Vehicles by Vehicle Alternative and Home-Work Group, including applicable incentives Living inside Cordon Living outside Cordon All Owners Working inside Working outside Working inside Working outside Cordon Cordon Cordon Cordon Conventional 1 144 (64.5%) 700 (49.0%) 4 974 (71.0%) 13 827 (75.6%) 20 645 (72.43%) Low CO 2 101 (5.7%) 99 (6.9%) 343 (4.9%) 985 (5.4%) 1 528 (5.36%) Petrol Low CO 63 (4.4%) 2 67 (3.8%) 206 (2.9%) 638 (3.5%) 974 (3.42%) Diesel Electric 47 (2.7%) 41 (2.9%) 94 (1.3%) 149 (0.8%) 331 (1.16%) Ethanol 415 (23.4%) 526 (36.8%) 1 386 (19.8%) 2 697 (14.7%) 5 024 (17.63%) Total 1 774 1 429 7 003 18 296 28 502 Key: Dotted = National Government Purchase Rebate; Dashed = Inner-City Residential Parking Exemption; Solid = Congestion Tax exemption; Represents those owners crossing the cordon. Through inspection of Table 1, it is apparent that the group with the highest share of exempt EEVs (electric, ethanol) was those owners commuting across the boundary but living inside the cordon. This is expected as these owners benefited from all three policies shown. The share of exempt EEVs was highest amongst those living within the cordon, but was also substantial for those living outside the cordon but still commuting across the boundary. Table 2 includes summary statistics for each vehicle alternative, providing average values for the various socio-demographic characteristics included in this analysis. Mean and Median values for i the Purchase Price and Total Weight (as a proxy for size) of each Vehicle Alternative have also been included in order to provide some insight into the alternative specific differences. It can be seen that on average Conventional, Ethanol and Electric vehicles in this sample were approximately the same size and price, although there was a large range of variation within Electric Vehicle category. Low CO Petrol models were the smallest and the cheapest, followed by Low CO Diesel 2 2 models. The number of alternatives included in the vehicle choice model could have been significantly greater given the range of vehicles in the dataset. The chosen level of aggregation of alternatives is motivated by the main research question of this study to better understand the impact of the tax exemption on the marginal demand for EEVs in Stockholm, and to gain some insight into individual preferences towards different aggregate EEV types, and not to analyze individual preferences towards every new vehicle available on the market. The five alternatives outlined in Table 2 are each distinctly different in regards to either: tax exemption eligibility; the demographic makeup of owners in that vehicle category; and/or the specific characteristics of that alternative. Two additional models have also been included in this paper to examine how the demographics of owners varied depending on exemption eligibility, vehicle purchase price and/or vehicle weight; however, all three models yield the same results in regards to policy implications.    Whitehead, J., Franklin, J., & Washington, S. 9   TABLE 2 – Summary Statistics for each Vehicle Alternative Low Low Electric/ Conventional CO CO Hybrid Ethanol Attribute Averages 2 2 Vehicles Petrol Diesel Electric No. of Observations 20 645 1 528 974 331 5 024 Mean Vehicle Purchase Price (EUR) 22 956 12 255 19 165 19 349 21 669 Median Vehicle Purchase Price (EUR) 19 130 11 120 18 400 23 040 19 290 Mean Vehicle Total Weight (kg) 1 958 1 229 1 728 1 935 1 918 Median Vehicle Total Weight (kg) 1 950 1 190 1 700 1 730 1 900 Owner Age (Years) 47.50 45.56 46.57 49.70 47.05 Owner 30 Years 4.98% 9.23% 5.54% 3.32% 4.32% Females 34.17% 57.72% 37.78% 35.95% 33.88% No. of Children 0.93 0.86 0.89 0.80 0.91 No. of Cars 1.28 1.30 1.30 1.29 1.22 Yearly Income (EUR) 47 504 34 314 42 860 90 628 41 087 Home inside Cordon 8.93% 13.09% 13.35% 26.59% 18.73% Commuting across Cordon Boundary 27.48% 28.93% 27.62% 40.79% 38.06% Home Distance from Cordon Boundary (km) 14.03 12.41 14.90 8.83 11.27 Home-Work Trip (km) 15.25 15.17 17.86 14.15 14.85 Finally, considering literature reviewing findings, the primary research questions, and the summary statistics shown in Tables 1 and 2, three research hypotheses were developed: 1.) Intra-Cordon residency had a significant, positive influence on an individual’s likelihood to purchase a tax exempt EEV i.e. electric or ethanol; 2.) The congestion tax cordon crossing exemption had a significant, positive influence on an individual’s likelihood to purchase an exempt EEV; and, 3.) Residential distance to the CBD had a significant and negative influence on the likelihood of purchasing an ethanol or electric vehicle. 6. Results Several model forms were tested and estimated, with these models varying in how vehicle alternatives were grouped and whether some of the explanatory variables were common across alternatives. The modeling aim was to develop the most plausible and defensible discrete choice model for capturing the relationships between vehicle choice and the demographics of private individuals purchasing new vehicles in Stockholm during 2008. It was also important for the model to provide the greatest insight into the effect of the congestion tax exemption on the demand for EEVs. The three different choice structures tested include: • Model 1: Binomial logit with two alternatives: a) EEVs exempt from congestion tax (electric/ethanol), b) Non-exempt vehicles (conventional, low CO petrol, CO diesel); 2 2 • Model 2: Multinomial logit model with five alternatives: a) conventional vehicles, b) low CO 2 petrol vehicles c) low CO diesel vehicles, d) electric/hybrid vehicles, d) ethanol; and, 2 • Model 3: Multinomial logit model with 8 alternatives based on the tax exemption (eligible or not), vehicle purchase price (cheap or expensive) and vehicle weight (light or heavy). Several alternative model specifications were also tested for each choice set, but ultimately deemed to be redundant when compared to the three models outlined previously. The model specifications tested include: • Binomial logit with 2 alternatives (EEVs vs. non-EEVs); • Multinomial logit with 3 alternatives (conventional vs. tax exempt EEVs vs. non-exempt EEVs);    Whitehead, J., Franklin, J., & Washington, S. 10   • Multinomial logit with 4 alternatives (conventional vs. low CO petrol/diesel vs. ethanol vs. 2 electric); • Nested logit version of Model 2 with 2 nests (tax exempt EEVs nested and non-exempt vehicles nested); • Nested logit version of Model 2 with 3 nests (tax exempt EEVs nested, non-exempt EEVs nested, non-exempt vehicles nested); • Nested logit version of Model 3, with varying nesting structures; and, • Nested logit with 20 alternatives based on fuel type, vehicle purchase price and vehicle weight, with varying nesting structures. The results of the binomial logit model (Model 1) aligned with expectations, however, it was apparent from the analysis of the summary statistics that there were substantial variations in the demographic makeup and alternative specific characteristics of categories within both the exempt EEV and non-exempt alternatives, and that the model could more accurately represent vehicle type choice by dividing the exempt group into two categories: electric versus ethanol; and by also dividing the non-exempt group into three categories: conventional versus low CO petrol and low 2 CO diesel. Aggregating vehicle type choice to five alternatives provided an opportunity to compare 2 between different types of EEVs, both exempt and non-exempt, whilst still principally allowing for the calculation of the effect of the tax exemption upon the demand for exempt EEVs, including how the policy affected different types of exempt EEVs. Among the nested logit structures, all tested specifications were found to have nesting parameters not statistically different from one, thus collapsing back to multinomial logit (MNL). Through iteratively experimenting with the available parameters and verifying the progressive improvement of the model with log-likelihood ratio tests, the final iteration resulted in a five- alternative MNL model with 33 estimated parameters. Correlation amongst coefficients for this model was reviewed, with no significant issues identified. The estimation results of Models 1 and 2 are provided in Table 3. When interpreting the results for Model 2, it is interesting to compare these findings with the estimates for Model 1 to understand the effects of aggregating the number of alternatives, and justifying the use of five alternatives compared to the simpler binomial model. One of the clearest results from Model 2 was that the ‘Living inside Cordon’ coefficient was significant for all five types of EEV, with positive values for all EEV alternatives (Electric = 0.815; Ethanol = 0.527; LowCO Diesel = 0.596; LowCO Petrol = 0.342). Electric vehicles had a 2 2 coefficient approximately two times greater than the coefficient for low CO petrol vehicles, whilst for 2 ethanol it was approximately one and half times larger. This relative difference is due to low CO 2 petrol vehicles that were not exempt from congestion tax. Interestingly, however, low CO diesel 2 vehicles had a coefficient higher than ethanol vehicles. This suggests that those owners that preferred low CO diesel were less sensitive to the congestion tax and also to the incentive of free 2 residential parking for inner-city residents. Overall, this positive coefficient supports hypothesis 1 - that higher preference towards energy-efficient vehicles exists for those residing within the cordon. The coefficients representing crossing the cordon boundary for home-work trips were, as to be expected, positive for exempt EEVs (Electric = 0.365; Ethanol = 0.311). This coefficient for both low CO petrol and low CO diesel vehicles was not statistically significant. This is reasonable given that 2 2 these vehicles were not exempt from the congestion tax. This coefficient, unsurprisingly, was estimated at a similar magnitude in Model 1. These findings support hypothesis 2, that crossing the cordon was a significant factor in determining an individual’s likelihood of purchasing an exempt EEV. An additional interaction variable was included to represent owners living inside the cordon and commuting across the boundary for work. This variable was only statistically significant for ethanol EEVs (Ethanol = 0.303), and corresponds with the findings from the initial analysis shown in Table 1; owners living within the cordon and crossing the boundary for work, being the only group that benefitted from all three major incentive policies (congestion tax exemption, free residential parking for inner-city residents and national purchase rebate), had the highest likelihood of purchasing exempt EEVs. Another noteworthy variable was an owners’ residential distance from the inner-city. This variable's coefficient was statistically significant for electric, ethanol and low CO diesel vehicles, 2 with all having negative values (Electric = -0.353; Ethanol = -0.129; Low CO Petrol = -0.156). This 2    Whitehead, J., Franklin, J., & Washington, S. 11   result supports the finding that those individuals within or close to the cordon had the highest preference towards purchasing an EEV and confirms hypothesis 3. Campbell et al. (2012) found the opposite, that the further an individual lived from the city center in Birmingham, UK, the more likely there were to purchase an EEV; however, their analysis was based upon the assumption that EEV owners were early adopters. Moreover, the effect of income distribution of Birmingham may have confounded the distance effect. TABLE 3 – Estimated Parameters of Multinomial Models 1 and 2 Model 1 Non-Exempt Vehicles = Base Alternative Log-Likelihood = -13 310.66 Exempt EEV Attributes: β S.E. Living inside cordon .510 .064 Commuting across boundary .314 .037 (CAB) Living inside cordon CAB .261 .086 Distance from inner-city -.128 .018 Home-work trip distance .031 .016 Income in 10k SEK -.004 .002 Number of children .020 .015 Number of vehicles -.125 .030 Owner under 30 years old -.174 .059 Female -.101 .033 ASC -1.23 .071 Model 2 Conventional Vehicles = Base Alternative Log-Likelihood = -23 991.78 Low CO Petrol Low CO Diesel Electric Ethanol 2 2 Attributes: β S.E. β S.E. β S.E. β S.E. Living inside cordon .342 .090 .596 .100 .815 .157 .527 .067 Commuting across boundary .016 .061 -.052 -.074 .365 .118 .311 .038 (CAB) Living inside cordon CAB .303 .088 Distance from inner-city -.156 .031 -.353 .082 -.129 .019 Home-work trip distance .084 .026 .166 .022 .117 .062 .043 .017 Income in 10k SEK -.046 .011 .005 .001 -.014 .003 Number of children -.093 .056 -.026 .015 Number of vehicles .189 .033 .082 .042 -.112 .030 Owner under 30 years old .578 .080 .337 .107 -.605 .260 Female .972 .055 .194 .068 ASC -4.11 .124 -3.77 .130 -4.08 .140 -1.33 .052 Key: = significant at 𝒑≤𝟎.𝟎𝟓; = significant at 𝒑≤𝟎.𝟏 Owner income was statistically significant for electric, ethanol and low CO petrol vehicles, 2 with a positive relationship for electric vehicles (Electric = 0.005) as opposed to the negative relationship for both ethanol (Ethanol = -0.014) and low CO petrol vehicles (Low CO Petrol = - 2 2 0.046). This overall trend suggests that wealthier owners were less sensitive to incentive-based policies. The positive coefficient for electric vehicles is presumed to arise because these vehicles were generally more expensive than other types of EEVs, and thus, it was predominantly wealthier individuals who could afford to purchase this vehicle type. This notion will be explored further upon examination of Model 3. This is one variable that highlights the differences between Models 1 and 2. Upon analysis of Model 1, it appears that owner income, although statistically significant and slightly negative, did not have a substantial influence on the utility of purchasing an exempt EEV. As stated above, however, this is not the case; rather that Model 1’s owner income coefficient for exempt EEVs was merely reflecting the opposite signs held by the two different exempt EEV alternatives (electric and ethanol) in Model 2. The variables gender and number of children were statistically significant for some of the alternatives in Model 2. The number of children held a negative coefficient for ethanol (Ethanol = - 0.026) and electric (-0.093). Gender was statistically significant for low CO petrol and diesel 2 vehicles, with females more likely to purchase these alternatives (Low CO Petrol = 0.972; Low CO 2 2    Whitehead, J., Franklin, J., & Washington, S. 12   Diesel = 0.194). This could reflect a tendency for women to be more environmentally-conscious than men, as found by some other studies (Dagsvik et al., 2002; Golob and Hensher, 1998; Mabit and Fosgerau, 2011), however, this does not appear to apply to exempt EEVs. Individuals under the age of 30 had a positive coefficient for low CO petrol and diesel 2 vehicles (Low CO Petrol = 0.578; Low CO Diesel = 0.337); however, this coefficient was negative 2 2 for electric vehicles (Electric = -0.605). That young owners were attracted to some EEVs is at least partially consistent with the findings of the previously discussed SP studies (Hackbarth and Madlener, 2013; Ziegler, 2012). The equivalent coefficient in Model 1 was also negative and statistically significant, but with a lower magnitude (Exempt EEV = -0.174). It is unclear exactly why vehicle owners under 30 years were less likely to purchase exempt EEVs; however, this may be a price and/or income consideration. An interaction variable for income age was tested, but was not found to be significant. Finally, contrary to the findings of both Campbell et al. (2012) and Graham-Rowe et al. (2012), in Stockholm it appears that owners of exempt EEVs tend to have less vehicles, where as owners of non-exempt EEVs (Low CO Petrol/Diesel) have higher number of vehicles. This is likely a size 2 consideration, which again will be explored further in the analysis of the results from Model 3. The estimation results of Model 3 have been included in Table 4. This model is an extension of the binomial model – Model 1 – where non-exempt and exempt alternatives were split into eight alternatives based on both purchase price (cheap or expensive) and vehicle weight as a proxy for vehicle size (light or heavy). The estimated coefficients of Model 3 largely reflect the same trends outlined for Model 2. It is interesting to note that the ‘Commuting across the cordon boundary’ (CAB) variable is statistically significant and positive for all exempt EEV alternatives, and relatively similar in magnitude regardless of purchase price or vehicle weight. The CAB variable is also positive for the non-exempt alternatives, however, as can be seen in Figure 1, the magnitude of these coefficients is, on average, approximately half that of the exempt EEV alternatives. FIGURE 2 – Percentage of New Exempt EEVs in Stockholm and Gothenburg for four Home-Work Groups 0.5 0.45 0.4 0.35 0.3 Non-Exempt 0.25 β Exempt 0.2 0.15 0.1 0.05 0 Light, Cheap Light, Expensive Heavy, Cheap Heavy, Expensive Recall that the alternatives estimated in Model 3 are relative to the base alternative – Light, Cheap, Non-Exempt, therefore it is reasonable to expect that the CAB coefficient would be positive and statistically significant for other non-exempt alternatives. It is likely that the CAB coefficients for non-exempt vehicles in this model are positive since vehicle owners commuting across the cordon would have had higher incomes and preferred larger or more expensive vehicles (whether exempt or not) relative to the Light, Cheap, Non-Exempt vehicles. This assertion is supported by all owner income coefficients being statistically significant and positive for non-exempt vehicle alternatives, except for Heavy, Cheap, Non-exempt, which does have a slightly positive and significant CAB variable, but with a much lower magnitude (0.079). Referring to Figure 1, we can also note that both of the Heavy, Exempt alternatives have the highest CAB estimate values. It is hard to say exactly what this result is reflecting, but it may be that heavy vehicle owners are more sensitive to the congestion tax as they already have higher operating costs, and as such, prefer exempt EEVs.     Whitehead, J., Franklin, J., & Washington, S. 13   TABLE 4 – Estimated Parameters of MNL Model with Eight Alternatives based on Price, Weight and Eligibility for Congestion Tax Exemption Model 3 Light, Cheap, Non-Exempt Vehicles = Base Alternative Light, Light, Light, Log-Likelihood = -43 233.53 Expensive, Expensive, Cheap, Exempt Non-Exempt Exempt Attributes: β S.E. β S.E. β S.E. Living inside cordon .884 .088 .927 .129 .650 .260 Commuting across boundary (CAB) .409 .050 .202 .085 .328 .160 Living inside cordon CAB -.066 .118 -.386 .188 .030 .337 Distance from inner-city -.161 .025 -.129 .040 -.342 .091 Home-work trip distance .028 .022 .048 .035 .155 .069 Income in 10k SEK .016 .009 .097 .006 .089 .010 Number of children .130 .021 -.023 .037 -.043 .071 Number of vehicles -.118 .044 .220 .042 -.110 .140 Owner under 30 years old -.119 .072 .179 .111 -.779 .301 Female -.381 .043 -.730 .075 -.273 .136 ASC -.784 .103 -2.145 .144 -3.577 .310 Heavy, Heavy, Heavy, Heavy, Cheap, Cheap, Expensive, Expensive, Non-Exempt Exempt Non-Exempt Exempt Attributes: β S.E. β S.E. β S.E. β S.E. Living inside cordon .079 .103 .503 .208 .655 .074 .803 .123 Commuting across boundary (CAB) .079 .047 .413 .107 .281 .039 .476 .068 Living inside cordon CAB -.401 .158 .057 .270 -.616 .112 -.051 .163 Distance from inner-city .026 .189 -.076 .051 -.035 .017 -.072 .034 Home-work trip distance .025 .017 -.003 .046 -.025 .016 .021 .030 Income in 10k SEK .008 .009 .009 .020 .096 .006 .095 .006 Number of children .273 .018 .161 .045 .276 .016 .255 .028 Number of vehicles -.027 .035 -.052 .087 -.130 .027 .015 .051 Owner under 30 years old -.178 .070 -.537 .189 -.379 .064 -.608 .132 Female -1.032 .042 -1.144 .106 -1.120 .036 -1.085 .067 ASC .077 .091 -1.621 .218 .255 .075 -1.345 .132 Key: = significant at 𝒑≤𝟎.𝟎𝟓; = significant at 𝒑≤𝟎.𝟏 Other points to highlight from the Model 3 estimation results include: - Individuals under 30 years of age were more likely to purchase non-exempt, lighter vehicles regardless of purchase price; - Females were more likely to purchase non-exempt, light, cheap vehicles, relative to all other alternatives; - Wealthier individuals were more likely purchase more expensive vehicles, regardless of exemption eligibility or vehicle weight/size; - Individuals living closest to the city were more likely to have smaller vehicles and also to have exempt EEVs – possibly reflecting higher environmental preferences amongst these residents (Bhat et al., 2009; Kahn, 2007), increased parking demands, and higher likelihood of crossing the cordon; - Owners of exempt EEVs had fewer vehicles (see Flamm (2009)) and owners of non- exempt vehicles had more vehicles –revealing perhaps that the economic benefits of the tax exemption outweighed the practical limitations of a smaller vehicle; and finally, - Owners with more children tended to have larger vehicles, however, there was little difference in preference among larger cars due to tax exemption or purchase price. 6.1 Policy Simulation    Whitehead, J., Franklin, J., & Washington, S. 14   The following section of this paper details the effect of the congestion tax exemption on the demand for new, exempt EEVs. Prior to detailing the results of the policy simulation outlined in the methodology, an additional analysis of panel data from 2004 to 2008, comparing new, exempt EEV shares in both Gothenburg and Stockholm, is included. Table 5 presents summary statistics of the panel data, comparing averages between the two metropolitan regions. The home-work grouping was constructed to be largely based around a cordon in Gothenburg where tolls have more recently been implemented, although those tolls were not active in that city during 2008, when this data was collected. TABLE 5 – Summary Statistics of dataset used for Stockholm compared to Gothenburg Summary Statistics Stockholm Gothenburg Difference No. of Observations 28 502 15 547 +13 186 Age (Years) 47.41 45.99 +1.334 Females 35.52% 38.65% -3.20%pts No. of Children 0.92 0.86 +0.07 Owner Income (EUR/Year) 46 008 33 099 +13 160 Living inside Cordon/Inner-city 11.24% 12.14% -0.90%pts Commuting across Cordon Boundary 29.58% 30.42% -0.93%pts EEVs 27.52% 38.25% -10.51%pts Congestion Tax Exempt EEVs 18.79% 15.66% +3.17%pts Congestion Tax Exempt EEVs and 7.18% 5.13% +2.07%pts Commuting across Cordon Boundary Home-Work Trip Distance (km) 15.25 12.57 +2.64 Distance of Residence from Cordon (km) 13.42 9.38 +4.01 Analysis Region: Population (persons) 1 925 735 745 317 +1 180 418 Land Area (sq.km) 4 509 1 892 +2 616 Population Density (p/sq.km) 427 394 +33 As shown in Table 5, comparing averages of each city reveals a great deal of similarity, with the most notable difference being income levels. Although the total population of the Stockholm region was much greater in 2008, the two study areas were very similar in terms of population density. Furthermore, both regions had similar shares of vehicle owners commuting across the cordon/inner-city boundary, similar shares of EEVs and similar shares of EEVs eligible for the congestion tax exemption in Stockholm. The vehicle registration data for Gothenburg came from the same source as the data used for Stockholm in the previous analysis. Additional demographic data was also sourced from Sweden’s Central Bureau of Statistics (SCB). Using the panel data described, the percentage of new, exempt EEVs in both Stockholm and Gothenburg have been presented in the four graphs displaying in Figure 2. Each graph refers to one of the four Home-Work Groups; as described previously: A. Living and working within the cordon; B. Living within but working outside the cordon (commute across the cordon); C. Living outside but working within the cordon (commute across the cordon); and, D. Living and working outside the cordon. As shown in Figure 2, particularly for the two groups commuting across the cordon (Groups B and C), the percentage of new, exempt EEVs increased over time in both cities between 2007 and 2008 (when the congestion tax exemption was introduced), while the demand in Stockholm increased at a much greater rate. Comparing these results to Group D – the group of vehicle owners that were least likely to be affected by the congestion tax exemption – there was relatively no difference in the rate of increase in demand for new, exempt EEVs between the two metropolitan areas. Interestingly, in Group A there was a greater increase in demand in Stockholm compared to Gothenburg, although these vehicle owners were not commuting across the cordon boundary and not directly affected by the congestion tax. This could be due to the free residential parking policy that was also active during this period, or perhaps a social marketing effect of increased visibility of    Whitehead, J., Franklin, J., & Washington, S. 15   EEVs in Stockholm. It is expected that this group would also be affected by the exemption tax given the high probability that these vehicle owners would need to drive across the cordon boundary regularly for other, non-commute based trips. FIGURE 2 – Percentage of New Exempt EEVs in Stockholm and Gothenburg for four Home-Work Groups % New Exempt EEVs - Group A % New Exempt EEVs - Group B 45.00% 45.00% Stockholm 40.00% 40.00% 35.00% 35.00% Gothenburg 30.00% 30.00% 25.00% 25.00% 20.00% 20.00% 15.00% 15.00% 10.00% 10.00% 5.00% 5.00% 0.00% 0.00% 2003 2004 2005 2006 2007 2008 2009 2003 2004 2005 2006 2007 2008 2009 Year Year % New Exempt EEVs - Group C % New Exempt EEVs - Group D 45.00% 45.00% 40.00% 40.00% 35.00% 35.00% 30.00% 30.00% 25.00% 25.00% 20.00% 20.00% 15.00% 15.00% 10.00% 10.00% 5.00% 5.00% 0.00% 0.00% 2003 2004 2005 2006 2007 2008 2009 2003 2004 2005 2006 2007 2008 2009 Year Year By taking a difference-in-differences approach, an estimate of the effect of the congestion tax exemption was calculated. Table 6 details the differences in the increases in demand for new, exempt EEVs between Stockholm and Gothenburg, over the time periods: 2007 to 2006, 2008 to 2007 and 2008 to 2006. The difference-in-differences have been calculated as a whole, as well as for each of the four home-work groups outline above (Groups A, B, C, D). These particular time periods were selected for comparison since: the congestion tax trial was only valid during the first 6 months of 2006; the permanent exemption was in place for the last 5 months during 2007; but the exemption was in place for the entire 12 months during 2008. Although it may seem more suitable to compare 2005 with 2008, during 2005 the free residential for inner-city EEV owners was introduced in Stockholm, further conflating the results. It should also be noted that the purchase rebate policy for EEVs was introduced in mid-2007, however, since this was a national policy, it was assumed to affect vehicle owners in both cities equally and, therefore, not affect this analysis. As shown in Table 6, over the course of congestion tax exemption (2006 to 2008), there was a 1.56% greater increase in the market share of new, exempt EEV registrations in Stockholm compared to Gothenburg. Focusing specifically on 2008 compared to 2007, the increase in market share of exempt EEV registrations in Stockholm was 1.76% greater than in Gothenburg. The differences between the increases in two metropolitan areas have also been included in Table 6 by the four home-work groups. The difference in market share increases for Group D was negligible, with the largest difference occurring amongst Group B, followed by Groups A and C. Interestingly, the differences between Stockholm and Gothenburg from 2006 to 2007 appear to have been negligible. There was a definitive increase in the market shares of exempt EEVs for the two groups most likely to be affected by the congestion tax exemption (Groups B and C), but these were offset by the reductions for Groups A and D. The policy instability during these two years could have affected demand during this period, with the effect not settling down until 2008.    Whitehead, J., Franklin, J., & Washington, S. 16   In this analysis we can try to separate the effect of the free residential parking policy and the tax exemption by comparing Groups A and B. Assuming the upper bound of the free residential parking effect on CBD residents was 3.46% between 2007 and 2008 (assuming the congestion tax did not affect this group), means that the congestion tax likely resulted in a minimum of a 5.53% exempt EEV increase amongst CBD residents. It is more difficult to separate out the general CBD environmental preferences, which could have also influenced purchasing decisions for this group of vehicle owners. Such information could be obtained through follow-up SP surveys. TABLE 6 – Increase in market share of new, exempt EEV registrations: Stockholm compared to Gothenburg Stockholm vs. Gothenburg: % increase in market share of new exempt EEV registrations Time Period Group A: Group B: Group C: Group D: All Comparison Live in/Work in Live in/Work out Live out/Work in Live out/Work out 2007 vs. 2006 -0.87% 4.55% 1.27% -1.06% -0.20% 2008 vs. 2007 3.46% 8.99% 2.00% 0.95% 1.76% 2008 vs. 2006 2.59% 13.55% 3.27% -0.11% 1.56% Using the estimated coefficients of Model 2, a policy simulation was carried out in order to assess the effect that the congestion tax exemption had upon the demand for EEVs in Stockholm during 2008 using an alternative method. This simulation applied the methodology outlined in Section 4. By assuming the variable representing commuting across the cordon largely captured the effect of the exemption policy, this variable was set to zero for all observations, removing the utility benefit of this variable for exempt EEVs and simulating a scenario where the exemption was not active. The predicted shares were then recalculated and compared to the predicted shares of the original model. Referring to the predicted shares in Table 7, by first focusing on ‘All Owners’ it can be seen that, overall, the congestion tax exemption increased the share of exempt EEVs by 1.82% (+/- 0.32%; 95% C.I.). For owners living inside but working outside the cordon (Group B) the effect was substantially higher at 13.08% (+/- 3.18%; 95% C.I.), whilst for owners living outside but working inside cordon (Group C) the effect was a 4.76% increase (+/- 1.13%; 95% C.I.). Interestingly, these figures closely mirror the results of the panel data analysis comparing Stockholm and Gothenburg between 2008 and 2007. The overall estimate of the effect of the policy is almost identical at a 1.78% market share increase. The magnitude of the increase for each of the four groups does vary between the two methods, largely as a result of the policy simulation approach, which focused on the two groups that were affected by the policy vehicle owners that commuted across the boundary (Groups B and C). Moreover, the trend was similar for vehicle owners living inside the cordon and commuting out of it for work (Group B). The differences in shares of each vehicle type calculated through the policy simulation analysis reflected an increase in the total number of exempt EEVs in Stockholm by 10.7%, corresponding to an increase of 49.2% for those living inside and working outside the cordon, and an increase of 29.0% for those living outside and working inside the cordon. In other words, the congestion tax exemption appears to have had a substantial effect, leading to an increase of 519 (+/- 91; 95% C.I.) exempt EEVs in Stockholm during 2008, out of the 5 355 purchased that year (10.7% increase). Recall that this estimate is based upon the assumption that only those vehicle owners that commuted across the cordon were affected by the congestion tax exemption, when in fact non- commuting across the cordon vehicle owners would have also been influenced by this policy. This may be one reason as to why there are some differences in the estimates between the two methods outlined in this section. Most likely, the effect for Group B has been overestimated, capturing the effect of the policy upon those vehicle owners also living within the cordon but not commuting across the boundary (Group A). Since it is not possible to differentiate the effect of the inner-city free residential parking incentive, this policy partially conflates the estimates for Groups A and B. Previous studies, however, combined with the results of the panel data analysis, suggest that the congestion tax exemption policy had a relatively greater effect. Overall, both methods yield very similar estimates for the total effect of the congestion tax exemption policy on the demand for new, exempt EEVs in Stockholm during 2008, providing additional evidence to support the accuracy of these results. How these results compare with the results obtained in previous analyses of this policy will be explored further in the discussion.    Whitehead, J., Franklin, J., & Washington, S. 17   TABLE 7 – Predicted Vehicle Alternative Market Shares from MNL Model 2   Estimated Vehicle Alternative Market Shares in % (with 95% C.I.) Low CO Low CO Electric/ Exempt EEV 2 2 Conventional Ethanol Petrol Diesel Hybrid Total All Owners With 72.38 5.39 3.44 1.16 17.64 18.80 Exemption (+/- 0.51) (+/- 0.26) (+/- 0.21) (+/- 0.13) (+/- 0.46) Without 73.95 5.50 3.58 1.05 15.92 16.97 Exemption (+/- 0.61) (+/- 0.31) (+/- 0.25) (+/- 0.16) (+/- 0.52) Exemption Effect -1.57 -0.11 -0.14 0.11 1.72 1.82 Market Share (%) (+/- 0.36) (+/- 0.17) (+/- 0.13) (+/- 0.10) (+/- 0.31) (+/- 0.32) Exemption Effect -2.12 -2.01 -3.94 10.46 10.08 10.73 Annual Sales (%) (+/- 0.49) (+/- 3.08) (+/- 3.65) (+/- 9.68) (+/- 1.81) (+/- 1.88) Exemption Effect -447 -31 -40 31 490 519 Annual Sales (+/- 103) (+/- 48) (+/- 37) (+/- 29) (+/- 88) (+/- 91) (Vehicles) Owners Living inside + Working outside Cordon With 49.91 6.20 3.85 3.07 36.98 40.05 Exemption (%) (+/- 2.37) (+/- 0.97) (+/- 0.74) (+/- 0.70) (+/- 2.46) Without 60.71 7.40 4.92 2.59 24.37 26.96 Exemption (%) (+/- 2.16) (+/- 1.00) (+/- 0.89) (+/- 0.65) (+/- 2.15) Exemption Effect -10.81 -1.20 -1.08 0.48 12.60 13.08 Market Share (%) (+/- 2.80) (+/- 0.79) (+/- 0.62) (+/- 0.65) (+/- 3.26) (+/- 3.18) Exemption Effect -18.08 -14.76 -19.68 20.09 52.05 49.18 Annual Sales (%) (+/- 4.68) (+/- 9.47) (+/- 11.47) (+/- 26.36) (+/- 13.59) (+/- 11.84) Exemption Effect -154 -17 -15 7 180 187 Annual Sales (+/- 40) (+/- 11) (+/- 9) (+/- 9) (+/- 47) (+/- 45) (Vehicles) Owners Living outside + Working inside Cordon With 70.77 5.10 3.08 1.29 19.76 21.05 Exemption (%) (+/- 1.04) (+/- 0.49) (+/- 0.38) (+/- 0.24) (+/- 0.93) Without 74.96 5.32 3.43 0.95 15.34 16.29 Exemption (%) (+/- 0.70) (+/- 0.36) (+/- 0.26) (+/- 0.15) (+/- 0.60) Exemption Effect -4.19 -0.22 -0.35 0.34 4.42 4.76 Market Share (%) (+/- 1.25) (+/- 0.56) (+/- 0.43) (+/- 0.27) (+/- 1.11) (+/- 1.13) Exemption Effect -5.57 -4.30 -10.63 33.92 28.75 29.07 Annual Sales (%) (+/-1.67) (+/-10.88) (+/-13.01) (+/-27.07) (+/-7.25) (+/-6.89) Exemption Effect -293 -15 -25 24 310 333 Annual Sales (+/- 88) (+/- 39) (+/- 30) (+/- 19) (+/- 78) (+/- 79) (Vehicles) Key: = significant at 𝒑≤𝟎.𝟎𝟓 7. Discussion This study provides an overview of not only the differences between individual preferences towards exempt EEVs (ethanol and electric) compared to other vehicles, but also estimates the differences between the various categories of EEVs. This study presents a number of important findings in addition to assessing the effect that the congestion tax exemption had upon the demand for new EEVs in Stockholm in 2008. Focusing on the estimation results, a few key variables differentiate among individuals’ preferences for EEVs. One of the most significant variables is the distance of residency from the CBD, with a statistically significant, negative relationship for both exempt EEV alternatives and low CO petrol vehicles. This finding suggests that the further an individual lived from the CBD, the less 2 likely they were to purchase an EEV. Complementing this finding is the positive relationship of intra- cordon residency, which is statistically significant for all four categories of EEVs. Individuals living closer to the inner-city may be motivated by financial incentives, may have higher levels of environmental awareness, may be more motivated to adopt cutting edge technologies, and as a result exhibit a preference towards ‘green’ alternatives (Hackbarth and Madlener, 2013; Jones and Dunlap, 1992; Mabit and Fosgerau, 2011; Ziegler, 2012).    Whitehead, J., Franklin, J., & Washington, S. 18   Focusing on cordon boundary crossings, the coefficient representing this commuting pattern in MNL model 2 was not statistically significant for low CO EEVs, suggesting that low CO owners 2 2 were not detectibly sensitive to congestion pricing. The coefficient for intra-cordon residency for low CO vehicle owners was, however, significant and positive. This could mean that the benefits of 2 improved fuel economy and, potentially, more widely distributed fuel types, outweighed the toll deterrence, and that the congestion charge was not set high enough. Intuitively, owners living within the cordon should have incurred higher costs due to the congestion pricing scheme, and therefore should have been more sensitive to the potential exemption, decreasing their likelihood of purchasing a non-exempt EEV i.e. low CO vehicles. An interaction variable representing owners 2 living inside the cordon and commuting across the boundary was also tested for low CO vehicles, 2 but was not found to be statistically significant. A potentially important and omitted factor that could influence vehicle demand is the price response to the exemption from local dealers across all vehicle types, but unfortunately these data were unavailable. Intra-cordon residents had the highest preferences towards electric vehicles, demonstrated by a coefficient of distance from the cordon for electric vehicles being greater than the magnitude of this coefficient for other EEVs. This finding could again point towards higher environmental attitudes of inner-city residents, with electric vehicles being seen as the most ‘green’ EEV alternative, higher disposable incomes of these residents, and other constraints such as parking. An interaction variable between income and residency within the cordon was not statistically significant. Driver age was significant for electric and low CO petrol/diesel vehicles, and shows a trend of 2 younger individuals (under the age of 30) preferring low CO petrol/diesel vehicles compared to 2 individuals over the age of 30 preferring electric vehicles, somewhat contrary to other studies suggesting that younger people prefer more ‘environmentally-friendly’ alternatives (Hackbarth and Madlener, 2013; Ziegler, 2012) Contrary to the findings of Campbell et al. (2012), no relationship was found between large car owners and number of vehicles owned, however, for smaller, exempt EEVs, vehicle owners tended to own fewer vehicles, supporting Flamm’s (2009) finding that EEV owners have fewer vehicles. Conversely, smaller, non-exempt vehicle owners in Stockholm own more vehicles. 7.1 Policy Effect The congestion tax exemption policy increased the demand for exempt EEVs in Stockholm during 2008. The variable representing vehicle owners commuting across the cordon boundary was, as expected, most significant for exempt EEVs. An indicator variable representing individuals who crossed the boundary for work and lived within the cordon was also found to be significant for ethanol EEVs, providing further evidence that the exemption was significant in inducing demand for exempt EEVs and that the policy had the strongest effect on owners living within the city and commuting across the boundary for work (Group B). The operationalization of the "treatment" was based on those working and living on opposite sides of the cordon. As stated previously, this effect could capture other socio-demographic effects. To control for possible socio-demographic effects, an additional analysis based upon panel data from 2004-2008 for both Stockholm and Sweden’s second largest city, Gothenburg, was conducted. Gothenburg has a similar geographic distribution of socio-economic and demographic groups to Stockholm, although at a reduced scale. Contrary to Stockholm, however, in 2008 there was no congestion pricing scheme in Gothenburg, nor were there exemption policies for EEVs. Using EEV demand in Gothenburg as a case control, a parallel estimate of the effect of the exemption policy on the demand for new, exempt EEVs in Stockholm was estimated and shown in Section 6.1. Overall, this case control methodology revealed that the tax exemption increased demand for new, exempt EEVs in Stockholm from 2007 to 2008 by an estimated 1.78%. The policy simulation suggests that the exemption policy increased the share of EEVs in Stockholm by 1.82% to a total share of 18.8%, corresponding to a 10.7% increase in the number of exempt EEVs sold during 2008 (519 exempt EEVs). These similar results suggest that the case control methodology is appropriate and strengthens belief in the estimates obtained using Stockholm data. Although the policy simulation did not capture the effect that the exemption policy had on owners not commuting across the cordon, the findings of the case control analysis using Gothenburg data suggests that the exemption increased the share of new, exempt EEVs for these vehicle owners (Group A) by up to 3.5%. Again, the effect of the inner-city free residential parking incentive on inner-city residents could not be separated from either these estimates, no doubt    Whitehead, J., Franklin, J., & Washington, S. 19   conflating the estimates for these two groups of vehicle owners. Regardless, previous studies and a comparison of the differences in market share increases between Groups A and B suggest that the congestion tax had a much larger effect on the demand for new, exempt EEVs, compared to the free residential parking policy for inner-city residents. Previous analyses of the tax exemption in Stockholm found that the increase in market share of exempt EEVs during 2008 due to the policy was 23% (Pädam et al., 2009) and 24% (Lindfors and Roxland, 2009). These estimates are significantly higher than the 10.7% sales increase estimated in this study. A primary reason for this discrepancy is that these two prior studies were based on aggregate level data that included both private and company-owned vehicles. Since aggregate level data were used, socio-demographic factors could not be taken into account, and admittedly, the two analyses were relatively less rigorous, with one focused mainly on the effect of the national purchase rebate, and the other based on a simpler OLS regression. Regardless of these discrepancies, given that approximately 50% of new vehicles in Stockholm during 2008 were company-owned, this paper’s estimate of the exemption policy effect is reasonable and within expectations. The compared results suggest that the congestion tax had a much greater impact upon company vehicle purchases compared to private vehicle purchases. 8. Conclusions By making use of unique evidence from revealed preferences of EEV owners in Stockholm, this study has identified the common characteristics of new EEV owners and estimated the effect of Stockholm's congestion tax exemption upon the demand for new, exempt EEVs during 2008. Individual’s with the greatest propensity towards purchasing an exempt EEV included: intra-cordon owners; owners living closest to the CBD, and owners commuting across the cordon boundary. It was also determined that owners under the age of 30 years and females preferred non-exempt EEVs (low CO petrol/diesel), whilst those over the age of 30 years preferred electric vehicles. The 2 results of this study also tend to suggest that EEV owners in fact own fewer vehicles. By calculating the predicted shares from the estimated MNL model for two different scenarios, the effect of the congestion tax exemption upon the demand for new EEVs in Stockholm during 2008 was estimated. Overall, the congestion tax exemption was found to have increased the share of exempt EEVs in Stockholm by 1.82%, with, as expected, a much stronger effect on those commuting across the boundary, with those living inside the cordon having a 13.08% increase, and those owners living outside the cordon having a 4.76% increase. This increase in demand corresponded to an additional 519 (+/- 91; 95% C.I.) new exempt EEVs purchased in Stockholm during 2008 or a 10.7% increase in private sales. This estimate is consistent with the existing literature. One limitation of these estimates was that the effect of the CBD free residential parking incentive, particularly in regards to inner-city residents, could not be separated from the effect of the congestion tax exemption. Despite this shortcoming, other studies have asserted that the free parking policy effect was minimal. In conclusion, policy makers can take note that an incentive-based policy can increase the demand for EEVs and it appears to be an appropriate approach to adopt when attempting to reduce transport emissions through encouraging a transition towards a ‘green’ vehicle fleet. In future studies it would be interesting to examine the potential rebound effects of the congestion tax exemption in regards to EEV usage. There is also a need to better understand vehicle-pricing responses by vehicle manufacturers in response to incentive policies that could in turn influence vehicle purchase decisions. A follow-up state-preference survey of Stockholm vehicle owners could also be useful for comparing with the revealed-preference based results and conclusions of this study. 9. Acknowledgements The authors wish to thank Anders Karlström, Carl Hamilton, Gunnar Isacsson, Yusak Susilo and Jonas Eliasson for their detailed and invaluable comments. Also thanks to Emma Frejinger and Roger Pydokke for their assistance in accessing key data sources and to the two reviewers for their constructive comments. This project was financed by a grant from the Centre for Transport Studies, Stockholm; by an Australian Postgraduate Award (APA) scholarship from Queensland University of    Whitehead, J., Franklin, J., & Washington, S. 20   Technology, Brisbane; and by a scholarship from the Automotive Australia 2020 Co-operative Research Centre (AutoCRC). 10. Reference List 1. Batley, R.P., Toner, J.P., Knight, M.J. (2004) A mixed logit model of uk household demand for alternative-fuel vehicles. International Journal of Transport Economics = Rivista Internazionale de Economia dei Trasporti 31, 55-77. 2. Beck, M.J., Rose, J.M., Hensher, D.A. (2013) Environmental attitudes and emissions charging: An example of policy implications for vehicle choice. Transportation Research Part A: Policy and Practice 50, 171-182. 3. Bhat, C.R., Sen, S., Eluru, N. (2009) The impact of demographics, built environment attributes, vehicle characteristics, and gasoline prices on household vehicle holdings and use. Transportation Research Part B: Methodological 43, 1-18. 4. BIerlaire, M. (2003) BIOGEME: A free package for the estimation of discrete choice models. Ascona, Switzerland. 5. Birath, K., Pädam, S. (2010) Clean vehicles in Stockholm: historical retrospect 1994-2010. Stockholm, Sweden. 6. Börjesson, M., Eliasson, J., Hugosson, M.B., Brundell-Freij, K. (2012) The Stockholm congestion charges—5 years on. Effects, acceptability and lessons learnt. Transport Policy 20, 1-12. 7. Brownstone, D., Bunch, D.S., Golob, T.F., Ren, W. (1996) A transactions choice model for forecasting demand for alternative-fuel vehicles. Research in Transportation Economics 4, 87-129. 8. Bunch, D.S., Bradley, M., Golob, T.F., Kitamura, R., Occhiuzzo, G.P. (1993) Demand for clean-fuel vehicles in California: A discrete-choice stated preference pilot project. Transportation Research Part A: Policy and Practice 27, 237-253. 9. Campbell, A.R., Ryley, T., Thring, R. (2012) Identifying the early adopters of alternative fuel vehicles: A case study of Birmingham, United Kingdom. Transportation Research Part A: Policy and Practice 46, 1318-1327. 10. Choo, S., Mokhtarian, P.L. (2004) What type of vehicle do people drive? The role of attitude and lifestyle in influencing vehicle type choice. Transportation Research Part A: Policy and Practice 38, 201-222. 11. Dagsvik, J.K., Wennemo, T., Wetterwald, D.G., Aaberge, R. (2002) Potential demand for alternative fuel vehicles. Transportation Research Part B: Methodological 36, 361-384. 12. de Haan, P., Mueller, M.G., Scholz, R.W. (2009) How much do incentives affect car purchase? Agent-based microsimulation of consumer choice of new cars—Part II: Forecasting effects of feebates based on energy-efficiency. Energy Policy 37, 1083-1094. 13. Ewing, G.O., Sarigöllü, E. (1998) Car fuel-type choice under travel demand management and economic incentives. Transportation Research Part D: Transport and Environment 3, 429-444. 14. Flamm, B. (2009) The impacts of environmental knowledge and attitudes on vehicle ownership and use. Transportation Research Part D: Transport and Environment 14, 272- 279.